【转】Convolutional Neural Network

卷积神经网络

Posted by Brian on June 7, 2017

TensorFlow实现卷积神经网络

学习个人笔记,共大家学习交流,欢迎拍砖 作者: 谢中朝

1. TensorFlow 实现简单的卷积神经网络

'''
载入MNIST数据集,并创建默认的Interactive Session
'''
from tensorflow.examples.tutorials.mnist import input_data
import tensorflow as tf
mnist = input_data.read_data_sets("MNIST_data/", one_hot = True)
sess = tf.InteractiveSession()
Extracting MNIST_data/train-images-idx3-ubyte.gz
Extracting MNIST_data/train-labels-idx1-ubyte.gz
Extracting MNIST_data/t10k-images-idx3-ubyte.gz
Extracting MNIST_data/t10k-labels-idx1-ubyte.gz
'''
创建卷及神经网络的权重和偏置
(1)我们需要给权重制造随机噪声来打破完全对称,比如截断的正态分布噪声,标准差设为0.1
(2)使用ReLU给偏置增加一些小的正值(0.1)用来避免死亡节点
'''
def weight_variable(shape):
        initial = tf.truncated_normal(shape, stddev=0.1)
        return tf.Variable(initial)

def bias_variable(shape):
        initial = tf.constant(0.1, shape = shape)
        return tf.Variable(initial)
'''
创建卷积层和池化层
(1)参数x为输入,W为卷积参数,[5,5,1,32]:前面两个数字代表卷积尺寸,第三个数字代表有多少个channel,灰度单色为1,彩色RGB图片为3,
最后一个数字代表卷积数量.
(2)Strides代表模板卷积模板移动的步长,都是1代表会不遗漏地划过图片的每一个点.
(3)Padding代表边界的处理方式,这里的SAME代表给边界加上Padding让卷积的输出和输入保持同样的(SAME)的尺寸.
'''
def conv2d(x, W):
        return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')

def max_pool_2x2(x):
        return tf.nn.max_pool(x, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')
'''
定义placeholder,x是特征,y_是真实的label
(1)卷积神经网络会利用空间信息,因此需要将1D的输入向量转化为2D图片结构,1*784转化原始的28*28,只有一个颜色通道,
最终尺寸[-1,28,28,1],前面的-1代表样本数量不固定,最后1代表颜色通道数量.
'''
x = tf.placeholder(tf.float32, [None, 784])
y_ = tf.placeholder(tf.float32, [None, 10])
x_image = tf.reshape(x, [-1,28,28,1])
'''
定义第一个卷积层:
(1)参数初始化,包括weights和bias,[5,5,1,32]代表卷积核尺寸为5*5,1个颜色通道,32个不同卷积核.
(2)使用conv2d进行卷积,再加上偏置.
(3)接着使用ReLU激活函数进行非线性处理.
(4)最后用最大池化函数max_pool_2x2对卷积的结果进行池化.
'''
W_conv1 = weight_variable([5,5,1,32])
b_conv1 = bias_variable([32])
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
h_pool1 = max_pool_2x2(h_conv1)
'''
定义第二个卷积层:
(1)参数初始化,包括weights和bias,[5,5,1,64]代表卷积核尺寸为5*5,1个颜色通道,64个不同卷积核.
(2)使用conv2d进行卷积,再加上偏置.
(3)接着使用ReLU激活函数进行非线性处理.
(4)最后用最大池化函数max_pool_2x2对卷积的结果进行池化.
'''
'''
经历两次步长为2*2的最大池化,图片尺寸28*28变成7*7,由于第二层卷积核数量为64,输出尺寸即为7*7*64
'''
W_conv2 = weight_variable([5,5,32,64])
b_conv2 = bias_variable([64])
h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
h_pool2 = max_pool_2x2(h_conv2)
print (h_pool2.shape)
(?, 7, 7, 64)
'''
定义一个全连接层:
(1)将第二层卷积的输出tensor进行变形,将其转化为1D向量,然后连接一个全连接层,隐含节点为1024
(2)并使用ReLU激活函数
'''
W_fc1 = weight_variable([7 * 7 * 64, 1024])
b_fc1 = bias_variable([1024])
h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64])
print (h_pool2_flat.shape)
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)
print (h_fc1.shape)
(?, 3136)
(?, 1024)
'''
定义一个Dropout层:
(1)为了减轻过拟合,使用Dropout,通过一个placeholder传入keep_prob比率进行控制.
(2)在训练过程中,我们随机丢弃一部分节点的数据来减轻过拟合.
(3)预测时则保留全部的数据来追求最好的预测性能.
'''
keep_prob = tf.placeholder(tf.float32)
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)
print h_fc1_drop.shape
(?, 1024)
'''
最后将Dropout层的输出连接一个Softmax层,得到最后的概率输出
'''
W_fc2 = weight_variable([1024, 10])
b_fc2 = bias_variable([10])
y_conv = tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)
print y_conv.shape
(?, 10)
'''
定义损失函数cross entropy, 优化器使用Adam, 并给予一个比较小的学习速率1e-4
'''
cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y_conv), reduction_indices=[1]))
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
'''
定义评定准确率的操作:
'''
correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
'''
开始训练过程:
(1)首先初始化所有参数
(2)设置训练时的Dropout的keep_prob比率为0.5
(3)然后使用大小为50的mini-batch,进行20000次的训练迭代,参与训练样本的数量总共为100万
(4)其中每训练100次,对准确率进行一次评测(评测时keep_prob设为1),用于实时监控模型性能
'''
tf.global_variables_initializer().run()
for i in range(20000):
        batch = mnist.train.next_batch(50)#label:batch[1]代表50行*10列; input:batch[0]代表1024行*50列
        if i%100 == 0:
                train_accuracy = accuracy.eval(feed_dict={x:batch[0], y_:batch[1], keep_prob: 1.0})
                print ("step %d     , training accuracy %g"%(i, train_accuracy))
        train_step.run(feed_dict={x: batch[0], y_:batch[1], keep_prob: 0.5})
step 0     , training accuracy 0.06
step 100     , training accuracy 0.8
step 200     , training accuracy 0.86
step 300     , training accuracy 0.92
step 400     , training accuracy 0.94
step 500     , training accuracy 0.9
step 600     , training accuracy 0.98
step 700     , training accuracy 0.92
step 800     , training accuracy 0.96
step 900     , training accuracy 0.98
step 1000     , training accuracy 0.96
step 1100     , training accuracy 0.94
step 1200     , training accuracy 0.98
step 1300     , training accuracy 1
step 1400     , training accuracy 0.96
step 1500     , training accuracy 0.98
step 1600     , training accuracy 0.98
step 1700     , training accuracy 1
step 1800     , training accuracy 0.94
step 1900     , training accuracy 0.98
step 2000     , training accuracy 1
step 2100     , training accuracy 1
step 2200     , training accuracy 1
step 2300     , training accuracy 0.98
step 2400     , training accuracy 0.96
step 2500     , training accuracy 1
step 2600     , training accuracy 0.94
step 2700     , training accuracy 0.98
step 2800     , training accuracy 1
step 2900     , training accuracy 0.94
step 3000     , training accuracy 0.96
step 3100     , training accuracy 0.96
step 3200     , training accuracy 1
step 3300     , training accuracy 0.98
step 3400     , training accuracy 1
step 3500     , training accuracy 0.96
step 3600     , training accuracy 1
step 3700     , training accuracy 1
step 3800     , training accuracy 1
step 3900     , training accuracy 0.98
step 4000     , training accuracy 1
step 4100     , training accuracy 1
step 4200     , training accuracy 0.96
step 4300     , training accuracy 0.98
step 4400     , training accuracy 0.98
step 4500     , training accuracy 0.96
step 4600     , training accuracy 1
step 4700     , training accuracy 1
step 4800     , training accuracy 1
step 4900     , training accuracy 1
step 5000     , training accuracy 0.98
step 5100     , training accuracy 1
step 5200     , training accuracy 1
step 5300     , training accuracy 1
step 5400     , training accuracy 1
step 5500     , training accuracy 0.98
step 5600     , training accuracy 1
step 5700     , training accuracy 1
step 5800     , training accuracy 0.98
step 5900     , training accuracy 1
step 6000     , training accuracy 1
step 6100     , training accuracy 1
step 6200     , training accuracy 1
step 6300     , training accuracy 1
step 6400     , training accuracy 1
step 6500     , training accuracy 0.98
step 6600     , training accuracy 1
step 6700     , training accuracy 0.98
step 6800     , training accuracy 1
step 6900     , training accuracy 1
step 7000     , training accuracy 1
step 7100     , training accuracy 0.98
step 7200     , training accuracy 0.98
step 7300     , training accuracy 1
step 7400     , training accuracy 1
step 7500     , training accuracy 1
step 7600     , training accuracy 1
step 7700     , training accuracy 1
step 7800     , training accuracy 0.98
step 7900     , training accuracy 0.98
step 8000     , training accuracy 1
step 8100     , training accuracy 0.96
step 8200     , training accuracy 1
step 8300     , training accuracy 0.98
step 8400     , training accuracy 1
step 8500     , training accuracy 1
step 8600     , training accuracy 1
step 8700     , training accuracy 1
step 8800     , training accuracy 1
step 8900     , training accuracy 1
step 9000     , training accuracy 1
step 9100     , training accuracy 1
step 9200     , training accuracy 0.98
step 9300     , training accuracy 0.98
step 9400     , training accuracy 1
step 9500     , training accuracy 0.98
step 9600     , training accuracy 0.98
step 9700     , training accuracy 0.98
step 9800     , training accuracy 1
step 9900     , training accuracy 1
step 10000     , training accuracy 1
step 10100     , training accuracy 1
step 10200     , training accuracy 1
step 10300     , training accuracy 0.98
step 10400     , training accuracy 1
step 10500     , training accuracy 1
step 10600     , training accuracy 1
step 10700     , training accuracy 1
step 10800     , training accuracy 1
step 10900     , training accuracy 0.98
step 11000     , training accuracy 1
step 11100     , training accuracy 1
step 11200     , training accuracy 1
step 11300     , training accuracy 1
step 11400     , training accuracy 1
step 11500     , training accuracy 1
step 11600     , training accuracy 1
step 11700     , training accuracy 1
step 11800     , training accuracy 1
step 11900     , training accuracy 1
step 12000     , training accuracy 1
step 12100     , training accuracy 1
step 12200     , training accuracy 1
step 12300     , training accuracy 1
step 12400     , training accuracy 1
step 12500     , training accuracy 1
step 12600     , training accuracy 1
step 12700     , training accuracy 1
step 12800     , training accuracy 1
step 12900     , training accuracy 1
step 13000     , training accuracy 1
step 13100     , training accuracy 1
step 13200     , training accuracy 1
step 13300     , training accuracy 1
step 13400     , training accuracy 1
step 13500     , training accuracy 1
step 13600     , training accuracy 1
step 13700     , training accuracy 1
step 13800     , training accuracy 1
step 13900     , training accuracy 0.96
step 14000     , training accuracy 1
step 14100     , training accuracy 1
step 14200     , training accuracy 1
step 14300     , training accuracy 1
step 14400     , training accuracy 1
step 14500     , training accuracy 1
step 14600     , training accuracy 1
step 14700     , training accuracy 1
step 14800     , training accuracy 1
step 14900     , training accuracy 1
step 15000     , training accuracy 1
step 15100     , training accuracy 1
step 15200     , training accuracy 1
step 15300     , training accuracy 1
step 15400     , training accuracy 1
step 15500     , training accuracy 1
step 15600     , training accuracy 0.98
step 15700     , training accuracy 1
step 15800     , training accuracy 1
step 15900     , training accuracy 1
step 16000     , training accuracy 1
step 16100     , training accuracy 1
step 16200     , training accuracy 1
step 16300     , training accuracy 1
step 16400     , training accuracy 0.98
step 16500     , training accuracy 1
step 16600     , training accuracy 1
step 16700     , training accuracy 1
step 16800     , training accuracy 1
step 16900     , training accuracy 1
step 17000     , training accuracy 1
step 17100     , training accuracy 1
step 17200     , training accuracy 1
step 17300     , training accuracy 1
step 17400     , training accuracy 1
step 17500     , training accuracy 1
step 17600     , training accuracy 1
step 17700     , training accuracy 1
step 17800     , training accuracy 1
step 17900     , training accuracy 1
step 18000     , training accuracy 1
step 18100     , training accuracy 1
step 18200     , training accuracy 1
step 18300     , training accuracy 1
step 18400     , training accuracy 1
step 18500     , training accuracy 1
step 18600     , training accuracy 1
step 18700     , training accuracy 1
step 18800     , training accuracy 1
step 18900     , training accuracy 1
step 19000     , training accuracy 1
step 19100     , training accuracy 1
step 19200     , training accuracy 1
step 19300     , training accuracy 1
step 19400     , training accuracy 1
step 19500     , training accuracy 1
step 19600     , training accuracy 1
step 19700     , training accuracy 1
step 19800     , training accuracy 1
step 19900     , training accuracy 1
'''
在最终的数据集上进行全面测试,得到整体的分类的准确率
'''
print("test_accuracy %g"%accuracy.eval(feed_dict={x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0}))
test_accuracy 0.9923

2. TensorFlow 实现进阶的卷积神经网络

cd /home/xzz/code/python_tensorflow/models-master/tutorials/image/cifar10/
/home/xzz/code/python_tensorflow/models-master/tutorials/image/cifar10
#载入常用的库
import cifar10
import cifar10_input
import tensorflow as tf
import numpy as np
import time
'''
定义batch_size, 训练轮数max_steps,以及下载CIFAR-10数据的默认路径
'''
max_step = 100000
batch_size = 128
data_dir = '/home/xzz/download/cifar-10-batches-bin/'
'''
(1)定义初始化weight的函数,使用tf.truncated_normal截断的正态分布来初始化权重,给weight加一个L2的loss,进行L2正则化处理.
(2)在机器学习中,不管是分类任务还是回归任务,都可能因特征过多而导致过拟合,一般可以通过减少特征或者惩罚不重要特征的权重来缓解这一问题.
(3)正则化就是帮组惩罚特征权重的,特征权重也会成为模型的损失函数的一部分.一般来说,L1正则会制造稀疏的特征,大部分无用的特征权重会被置为0,
而L2正则会让特征的权重不过大,使得特征的权重比较平均.
(4)我们使用w1控制L2 loss的大小, 使用tf.nn.l2_loss函数计算weight的L2 loss,再使用tf.multipy让L2 loss乘以w1得到weight loss.接着使用
tf.add_to_collection把weight loss统一存在一个collection,这个colloction称为'losses',在计算神经网络的总体loss时被用上.
'''
def variable_with_weight_loss(shape, stddev, wl):
    var = tf.Variable(tf.truncated_normal(shape, stddev = stddev))
    if wl is not None:
        weight_loss = tf.multiply(tf.nn.l2_loss(var), wl, name = 'weight_loss')
        tf.add_to_collection('losses', weight_loss)
    return var 
#使用cifar10类下载数据集,并解压,展开到默认位置
cifar10.maybe_download_and_extract()
'''
(1)使用cifar10_input类中的distorted_inputs函数产生训练需要使用的数据,包括特征及其对应的label.
(2)对图像的进行数据增强的操作耗费大量的CPU时间,因此disorted_inputs使用了16个独立的线程来加速任务,函数内部会产生线程池,
在需要使用的时候会通过TensorFlow queue进行调度
'''
images_train, labels_train = cifar10_input.distorted_inputs(data_dir = data_dir, batch_size = batch_size)
Filling queue with 20000 CIFAR images before starting to train. This will take a few minutes.
'''
cifar10_input.inputs函数生成测试数据,不需要太多的处理,将图片裁剪为24*24大小的区块,进行数据的标准化处理
'''
images_test, labels_test = cifar10_input.inputs(eval_data=True, data_dir=data_dir, batch_size=batch_size)
'''
(1)创建输入数据placeholder,包括特征和label
(2)数据尺寸中的图片尺寸为24*24,即是裁剪后的大小,颜色通道数则设为3,代表图片是彩色有RGB三条通道
'''
image_holder = tf.placeholder(tf.float32, [batch_size, 24, 24, 3])
label_holder =  tf.placeholder(tf.int32, [batch_size])
'''
(1)开始创建第一个卷积层
(2)先对variable_with_weight_loss函数创建卷积核参数初始化,第一个卷积层使用5*5卷积核大小,3个颜色通道,64个卷积核
(3)不对第一个卷积层的weight进行L2正则,因此w1(weight loss)这项设为0
(4)使用tf.nn.conv2d函数对输入数据image_holder进行卷积操作,步长为1,padding模式为SAME,将这层bias全部初始化为0,
再将卷积结果加上bias,最后使用ReLU激活函数进行非线性化
(5)在激活函数后,使用3*3且步长为2*2最大池化层处理数据,这里最大池化尺寸和步长不一致可以增加数据的丰富性
(6)最后进行LRN,可以使其中响应比较大的值变得更大,并抑制其他反馈较小的神经元,增强模型的泛化能力,适合没有上界的激活函数
'''
weight1 = variable_with_weight_loss(shape=[5,5,3,64], stddev=5e-2, wl=0.0)
kernel1 = tf.nn.conv2d(image_holder, weight1, [1,1,1,1], padding='SAME')
print 'kernel', kernel1
bias1 = tf.Variable(tf.constant(0.0, shape=[64]))
print 'bias1', bias1
conv1 = tf.nn.relu(tf.nn.bias_add(kernel1, bias1))
print 'conv', conv1
pool1 = tf.nn.max_pool(conv1, ksize=[1,3,3,1], strides=[1,2,2,1],padding='SAME')
print 'pool', pool1
norm1 = tf.nn.lrn(pool1, 4, bias=1.0, alpha=0.001/9.0, beta=0.75)
print 'norm', norm1
kernel Tensor("Conv2D:0", shape=(128, 24, 24, 64), dtype=float32)
bias1 Tensor("Variable_1/read:0", shape=(64,), dtype=float32)
conv Tensor("Relu_2:0", shape=(128, 24, 24, 64), dtype=float32)
pool Tensor("MaxPool:0", shape=(128, 12, 12, 64), dtype=float32)
norm Tensor("LRN:0", shape=(128, 12, 12, 64), dtype=float32)
'''
(1)开始创建第二个卷积层
(2)64个通道,64个卷积核,注意bias全部初始化为0.1
(3)注意该层先进行LRN层处理,再使用最大池化层
'''
weight2 = variable_with_weight_loss(shape=[5,5,64,64], stddev=5e-2, wl=0.0)
kernel2 = tf.nn.conv2d(norm1, weight2, [1,1,1,1], padding='SAME')
bias2 = tf.Variable(tf.constant(0.0, shape=[64]))
conv2 = tf.nn.relu(tf.nn.bias_add(kernel2, bias2))
norm2 = tf.nn.lrn(conv2, 4, bias=1.0, alpha=0.001/9.0, beta=0.75)
pool2 = tf.nn.max_pool(norm2, ksize=[1,3,3,1], strides=[1,2,2,1],padding='SAME')
print 'pool', pool2
print batch_size
pool Tensor("MaxPool_1:0", shape=(128, 6, 6, 64), dtype=float32)
128
'''
(1)使用全连接层,将两个卷积层的输入结果全部flatten,使用tf.shape函数把每一个样本都变成一维向量,get_shape可以获得长度
(2)variable_with_weight_loss函数对全连接层的weight进行初始化,隐含节点数384,正态分布标准差设为0.04,bias设置为0.1
(3)为了这个全连接层不要过拟合,设置了非零weight loss即w1 = 0.004,让这一层所有参数都被L2正则所约束,最后ReLU激活函数进行非线性化
'''
reshape = tf.reshape(pool2, [batch_size, -1])
print 'reshape', reshape.shape
dim = reshape.get_shape()[1].value
print reshape.get_shape()[0]
print reshape.get_shape()[1]
print 'dim', dim
weight3 = variable_with_weight_loss(shape=[dim, 384], stddev = 0.04, wl = 0.004)
print 'weight3', weight3
bias3 = tf.Variable(tf.constant(0.1, shape=[384]))
print 'bias3', bias3
local3 = tf.nn.relu(tf.matmul(reshape, weight3) + bias3)
print 'local3', local3
reshape (128, 2304)
128
2304
dim 2304
weight3 Tensor("Variable_4/read:0", shape=(2304, 384), dtype=float32)
bias3 Tensor("Variable_5/read:0", shape=(384,), dtype=float32)
local3 Tensor("Relu_4:0", shape=(128, 384), dtype=float32)
'''
与前一个全连接层一样,就是隐藏节点下降一半,只有192个,其他的超参数保持不变
'''
weight4 = variable_with_weight_loss(shape=[384, 192], stddev = 0.04, wl = 0.004)
bias4 = tf.Variable(tf.constant(0.1, shape=[192]))
local4 = tf.nn.relu(tf.matmul(local3, weight4) + bias4)
print 'local4', local4
local4 Tensor("Relu_5:0", shape=(128, 192), dtype=float32)
'''
最后一层,先创建这一层的weight,其正态分布的标准差设为上一个隐藏层的节点数的倒数,并且不计入L2的正则
'''
weight5 = variable_with_weight_loss(shape=[192, 10], stddev = 1/192.0, wl = 0.0)
print 'weight5',weight5
bias5 = tf.Variable(tf.constant(0.0, shape=[10]))
logits = tf.add(tf.matmul(local4, weight5), bias5)
print 'logits', logits.shape
weight5 Tensor("Variable_8/read:0", shape=(192, 10), dtype=float32)
logits (128, 10)
'''
为了完成模型inference部分构建,接下来计算CNN的loss,将全部的loss整合到一起,
其中包括cross entropy loss,还有后两个全连接层中的weight的L2 loss
'''
def loss(logits, labels):
    labels = tf.cast(labels, tf.int64)
    cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
        logits = logits, labels = labels, name = 'cross_entropy_per_example')
    cross_entropy_mean = tf.reduce_mean(cross_entropy, name='cross_entropy')
    tf.add_to_collection('losses', cross_entropy_mean)
    return tf.add_n(tf.get_collection('losses'), name = 'total_loss' )
#将logits节点和label_placeholder传入loss函数获得最终的loss
loss = loss(logits, label_holder)
#优化器Adam Optimizer, 学习速率设置为1e-3
train_op = tf.train.AdamOptimizer(1e-3).minimize(loss)
#使用tf.nn.in_top_k函数求出输出结果中top k的准确率,默认使用top 1,输出得分最高的那一类准确率
top_k_op = tf.nn.in_top_k(logits, label_holder, 1)
#tf.InteractiveSession创建默认的session,接着初始化全部模型参数
sess = tf.InteractiveSession()
tf.global_variables_initializer().run()
'''
启动图片数据增强线程队列,一共有16个线程来进行加速
'''
tf.train.start_queue_runners()
[<Thread(Thread-4, started daemon 139821564057344)>,
 <Thread(Thread-5, started daemon 139821555664640)>,
 <Thread(Thread-6, started daemon 139821547271936)>,
 <Thread(Thread-7, started daemon 139821538879232)>,
 <Thread(Thread-8, started daemon 139821530486528)>,
 <Thread(Thread-9, started daemon 139820781270784)>,
 <Thread(Thread-10, started daemon 139820772878080)>,
 <Thread(Thread-11, started daemon 139820764485376)>,
 <Thread(Thread-12, started daemon 139820756092672)>,
 <Thread(Thread-13, started daemon 139820747699968)>,
 <Thread(Thread-14, started daemon 139820739307264)>,
 <Thread(Thread-15, started daemon 139820730914560)>,
 <Thread(Thread-16, started daemon 139820244399872)>,
 <Thread(Thread-17, started daemon 139820236007168)>,
 <Thread(Thread-18, started daemon 139820227614464)>,
 <Thread(Thread-19, started daemon 139820219221760)>,
 <Thread(Thread-20, started daemon 139820210829056)>,
 <Thread(Thread-21, started daemon 139820202436352)>,
 <Thread(Thread-22, started daemon 139820194043648)>,
 <Thread(Thread-23, started daemon 139820185650944)>,
 <Thread(Thread-24, started daemon 139820177258240)>,
 <Thread(Thread-25, started daemon 139820168865536)>,
 <Thread(Thread-26, started daemon 139820160472832)>,
 <Thread(Thread-27, started daemon 139820152080128)>,
 <Thread(Thread-28, started daemon 139820143687424)>,
 <Thread(Thread-29, started daemon 139820135294720)>,
 <Thread(Thread-30, started daemon 139820126902016)>,
 <Thread(Thread-31, started daemon 139820118509312)>,
 <Thread(Thread-32, started daemon 139820110116608)>,
 <Thread(Thread-33, started daemon 139820101723904)>,
 <Thread(Thread-34, started daemon 139820093331200)>,
 <Thread(Thread-35, started daemon 139820084938496)>,
 <Thread(Thread-36, started daemon 139820076545792)>,
 <Thread(Thread-37, started daemon 139820068153088)>]
'''
现在正式开始训练.
(1)在每一个step训练过程中,我们需要使用session的run方法执行images_train, labels_train
(2)获得一个batch的训练数据,在将这个batch的数据传入train_op和loss的计算
(3)我们记录,每一个step花费的时间,每隔10个step会计算和展示当前的loss,每秒训练的样本数量,
以及训练一个batch数据所花费的时间
'''
for step in range(max_step):
    start_time = time.time()
    image_batch, label_batch = sess.run([images_train, labels_train])
    _, loss_value = sess.run([train_op, loss],
                            feed_dict={image_holder: image_batch, label_holder: label_batch})
    duration = time.time() - start_time
    if step%10 == 0 :
        examples_per_sec = batch_size / duration
        sec_per_batch = float(duration)
        
        format_str=('step %d, loss=%.2f  (%.1f  examples/sec;  %.3f  sec/batch)' )
        print(format_str % (step, loss_value, examples_per_sec, sec_per_batch))
step 0, loss=4.68  (170.5  examples/sec;  0.751  sec/batch)
step 10, loss=3.66  (1896.5  examples/sec;  0.067  sec/batch)
step 20, loss=3.11  (1757.2  examples/sec;  0.073  sec/batch)
step 30, loss=2.75  (1859.3  examples/sec;  0.069  sec/batch)
step 40, loss=2.42  (1860.1  examples/sec;  0.069  sec/batch)
step 50, loss=2.25  (1932.3  examples/sec;  0.066  sec/batch)
step 60, loss=2.21  (1881.6  examples/sec;  0.068  sec/batch)
step 70, loss=2.02  (1929.8  examples/sec;  0.066  sec/batch)
step 80, loss=1.96  (1898.5  examples/sec;  0.067  sec/batch)
step 90, loss=2.01  (1902.7  examples/sec;  0.067  sec/batch)
step 100, loss=1.95  (1187.7  examples/sec;  0.108  sec/batch)
step 110, loss=2.03  (1928.6  examples/sec;  0.066  sec/batch)
step 120, loss=1.87  (1935.3  examples/sec;  0.066  sec/batch)
step 130, loss=1.88  (1802.0  examples/sec;  0.071  sec/batch)
step 140, loss=1.81  (1964.1  examples/sec;  0.065  sec/batch)
step 150, loss=1.86  (1899.4  examples/sec;  0.067  sec/batch)
step 160, loss=1.83  (1890.8  examples/sec;  0.068  sec/batch)
step 170, loss=1.99  (1129.3  examples/sec;  0.113  sec/batch)
step 180, loss=1.82  (1953.1  examples/sec;  0.066  sec/batch)
step 190, loss=1.75  (1931.8  examples/sec;  0.066  sec/batch)
step 200, loss=1.79  (1813.3  examples/sec;  0.071  sec/batch)
step 210, loss=1.55  (1550.7  examples/sec;  0.083  sec/batch)
step 220, loss=1.64  (1886.9  examples/sec;  0.068  sec/batch)
step 230, loss=1.71  (1889.5  examples/sec;  0.068  sec/batch)
step 240, loss=1.57  (1797.4  examples/sec;  0.071  sec/batch)
step 250, loss=1.78  (1909.6  examples/sec;  0.067  sec/batch)
step 260, loss=1.76  (1901.4  examples/sec;  0.067  sec/batch)
step 270, loss=1.73  (1283.8  examples/sec;  0.100  sec/batch)
step 280, loss=1.57  (1898.1  examples/sec;  0.067  sec/batch)
step 290, loss=1.53  (1974.6  examples/sec;  0.065  sec/batch)
step 300, loss=1.67  (2031.1  examples/sec;  0.063  sec/batch)
step 310, loss=1.69  (1917.7  examples/sec;  0.067  sec/batch)
step 320, loss=1.81  (1857.9  examples/sec;  0.069  sec/batch)
step 330, loss=1.52  (1961.6  examples/sec;  0.065  sec/batch)
step 340, loss=1.55  (1857.5  examples/sec;  0.069  sec/batch)
step 350, loss=1.63  (1730.8  examples/sec;  0.074  sec/batch)
step 360, loss=1.75  (1848.8  examples/sec;  0.069  sec/batch)
step 370, loss=1.59  (1582.6  examples/sec;  0.081  sec/batch)
step 380, loss=1.67  (1652.3  examples/sec;  0.077  sec/batch)
step 390, loss=1.64  (926.2  examples/sec;  0.138  sec/batch)
step 400, loss=1.62  (1666.3  examples/sec;  0.077  sec/batch)
step 410, loss=1.34  (1907.2  examples/sec;  0.067  sec/batch)
step 420, loss=1.40  (1929.3  examples/sec;  0.066  sec/batch)
step 430, loss=1.38  (1904.4  examples/sec;  0.067  sec/batch)
step 440, loss=1.57  (1899.8  examples/sec;  0.067  sec/batch)
step 450, loss=1.61  (1939.0  examples/sec;  0.066  sec/batch)
step 460, loss=1.42  (1908.8  examples/sec;  0.067  sec/batch)
step 470, loss=1.48  (1887.1  examples/sec;  0.068  sec/batch)
step 480, loss=1.44  (1865.2  examples/sec;  0.069  sec/batch)
step 490, loss=1.47  (1965.5  examples/sec;  0.065  sec/batch)
step 500, loss=1.62  (1778.1  examples/sec;  0.072  sec/batch)
step 510, loss=1.49  (1951.4  examples/sec;  0.066  sec/batch)
step 520, loss=1.41  (1848.1  examples/sec;  0.069  sec/batch)
step 530, loss=1.29  (1861.3  examples/sec;  0.069  sec/batch)
step 540, loss=1.49  (1864.5  examples/sec;  0.069  sec/batch)
step 550, loss=1.50  (1925.5  examples/sec;  0.066  sec/batch)
step 560, loss=1.59  (1965.4  examples/sec;  0.065  sec/batch)
step 570, loss=1.53  (1990.2  examples/sec;  0.064  sec/batch)
step 580, loss=1.39  (1908.6  examples/sec;  0.067  sec/batch)
step 590, loss=1.56  (1961.8  examples/sec;  0.065  sec/batch)
step 600, loss=1.53  (1902.4  examples/sec;  0.067  sec/batch)
step 610, loss=1.36  (1986.2  examples/sec;  0.064  sec/batch)
step 620, loss=1.41  (1975.8  examples/sec;  0.065  sec/batch)
step 630, loss=1.37  (1359.0  examples/sec;  0.094  sec/batch)
step 640, loss=1.42  (2006.3  examples/sec;  0.064  sec/batch)
step 650, loss=1.32  (1926.8  examples/sec;  0.066  sec/batch)
step 660, loss=1.38  (1856.8  examples/sec;  0.069  sec/batch)
step 670, loss=1.31  (1615.1  examples/sec;  0.079  sec/batch)
step 680, loss=1.42  (1897.6  examples/sec;  0.067  sec/batch)
step 690, loss=1.31  (1951.4  examples/sec;  0.066  sec/batch)
step 700, loss=1.45  (1058.2  examples/sec;  0.121  sec/batch)
step 710, loss=1.33  (1863.9  examples/sec;  0.069  sec/batch)
step 720, loss=1.41  (1924.8  examples/sec;  0.067  sec/batch)
step 730, loss=1.39  (1917.7  examples/sec;  0.067  sec/batch)
step 740, loss=1.60  (1858.7  examples/sec;  0.069  sec/batch)
step 750, loss=1.45  (1933.1  examples/sec;  0.066  sec/batch)
step 760, loss=1.30  (1843.2  examples/sec;  0.069  sec/batch)
step 770, loss=1.29  (1922.2  examples/sec;  0.067  sec/batch)
step 780, loss=1.39  (1924.9  examples/sec;  0.066  sec/batch)
step 790, loss=1.45  (1907.3  examples/sec;  0.067  sec/batch)
step 800, loss=1.30  (1913.0  examples/sec;  0.067  sec/batch)
step 810, loss=1.36  (1934.2  examples/sec;  0.066  sec/batch)
step 820, loss=1.34  (1886.5  examples/sec;  0.068  sec/batch)
step 830, loss=1.37  (1841.6  examples/sec;  0.070  sec/batch)
step 840, loss=1.35  (1912.6  examples/sec;  0.067  sec/batch)
step 850, loss=1.24  (1909.8  examples/sec;  0.067  sec/batch)
step 860, loss=1.41  (1917.0  examples/sec;  0.067  sec/batch)
step 870, loss=1.26  (1821.0  examples/sec;  0.070  sec/batch)
step 880, loss=1.29  (1823.2  examples/sec;  0.070  sec/batch)
step 890, loss=1.31  (1913.2  examples/sec;  0.067  sec/batch)
step 900, loss=1.44  (2066.2  examples/sec;  0.062  sec/batch)
step 910, loss=1.47  (1844.2  examples/sec;  0.069  sec/batch)
step 920, loss=1.38  (1914.7  examples/sec;  0.067  sec/batch)
step 930, loss=1.18  (1884.0  examples/sec;  0.068  sec/batch)
step 940, loss=1.33  (1877.6  examples/sec;  0.068  sec/batch)
step 950, loss=1.19  (1826.2  examples/sec;  0.070  sec/batch)
step 960, loss=1.38  (1940.5  examples/sec;  0.066  sec/batch)
step 970, loss=1.37  (1947.3  examples/sec;  0.066  sec/batch)
step 980, loss=1.26  (1921.8  examples/sec;  0.067  sec/batch)
step 990, loss=1.29  (1844.1  examples/sec;  0.069  sec/batch)
step 1000, loss=1.19  (2098.3  examples/sec;  0.061  sec/batch)
step 1010, loss=1.19  (1868.5  examples/sec;  0.069  sec/batch)
step 1020, loss=1.29  (1836.6  examples/sec;  0.070  sec/batch)
step 1030, loss=1.37  (1896.0  examples/sec;  0.068  sec/batch)
step 1040, loss=1.23  (1870.3  examples/sec;  0.068  sec/batch)
step 1050, loss=1.27  (1877.9  examples/sec;  0.068  sec/batch)
step 1060, loss=1.39  (1963.9  examples/sec;  0.065  sec/batch)
step 1070, loss=1.28  (1979.4  examples/sec;  0.065  sec/batch)
step 1080, loss=1.21  (1917.5  examples/sec;  0.067  sec/batch)
step 1090, loss=1.40  (1896.0  examples/sec;  0.068  sec/batch)
step 1100, loss=1.15  (1893.4  examples/sec;  0.068  sec/batch)
step 1110, loss=1.11  (1841.6  examples/sec;  0.070  sec/batch)
step 1120, loss=1.23  (1855.3  examples/sec;  0.069  sec/batch)
step 1130, loss=1.31  (1797.0  examples/sec;  0.071  sec/batch)
step 1140, loss=1.28  (1819.6  examples/sec;  0.070  sec/batch)
step 1150, loss=1.26  (1958.6  examples/sec;  0.065  sec/batch)
step 1160, loss=1.23  (1896.6  examples/sec;  0.067  sec/batch)
step 1170, loss=1.34  (1933.1  examples/sec;  0.066  sec/batch)
step 1180, loss=1.24  (1869.0  examples/sec;  0.068  sec/batch)
step 1190, loss=1.19  (1889.1  examples/sec;  0.068  sec/batch)
step 1200, loss=1.37  (1826.7  examples/sec;  0.070  sec/batch)
step 1210, loss=1.40  (1950.8  examples/sec;  0.066  sec/batch)
step 1220, loss=1.26  (1833.5  examples/sec;  0.070  sec/batch)
step 1230, loss=1.23  (1888.9  examples/sec;  0.068  sec/batch)
step 1240, loss=1.17  (1877.0  examples/sec;  0.068  sec/batch)
step 1250, loss=1.41  (1960.0  examples/sec;  0.065  sec/batch)
step 1260, loss=1.29  (1854.7  examples/sec;  0.069  sec/batch)
step 1270, loss=1.13  (2023.2  examples/sec;  0.063  sec/batch)
step 1280, loss=1.34  (1929.8  examples/sec;  0.066  sec/batch)
step 1290, loss=1.20  (1873.5  examples/sec;  0.068  sec/batch)
step 1300, loss=1.24  (1951.3  examples/sec;  0.066  sec/batch)
step 1310, loss=1.15  (1840.3  examples/sec;  0.070  sec/batch)
step 1320, loss=1.06  (1911.2  examples/sec;  0.067  sec/batch)
step 1330, loss=1.26  (1949.0  examples/sec;  0.066  sec/batch)
step 1340, loss=1.24  (1896.5  examples/sec;  0.067  sec/batch)
step 1350, loss=1.17  (1912.7  examples/sec;  0.067  sec/batch)
step 1360, loss=1.08  (1925.7  examples/sec;  0.066  sec/batch)
step 1370, loss=1.49  (1997.7  examples/sec;  0.064  sec/batch)
step 1380, loss=1.32  (1969.7  examples/sec;  0.065  sec/batch)
step 1390, loss=1.29  (1819.8  examples/sec;  0.070  sec/batch)
step 1400, loss=1.34  (2006.5  examples/sec;  0.064  sec/batch)
step 1410, loss=1.12  (1844.8  examples/sec;  0.069  sec/batch)
step 1420, loss=1.20  (1859.7  examples/sec;  0.069  sec/batch)
step 1430, loss=1.07  (1901.2  examples/sec;  0.067  sec/batch)
step 1440, loss=1.13  (1832.0  examples/sec;  0.070  sec/batch)
step 1450, loss=1.05  (1869.1  examples/sec;  0.068  sec/batch)
step 1460, loss=1.25  (1917.7  examples/sec;  0.067  sec/batch)
step 1470, loss=1.27  (1792.5  examples/sec;  0.071  sec/batch)
step 1480, loss=1.18  (1874.7  examples/sec;  0.068  sec/batch)
step 1490, loss=1.20  (1927.0  examples/sec;  0.066  sec/batch)
step 1500, loss=1.09  (1921.9  examples/sec;  0.067  sec/batch)
step 1510, loss=1.20  (1896.2  examples/sec;  0.068  sec/batch)
step 1520, loss=1.30  (1800.1  examples/sec;  0.071  sec/batch)
step 1530, loss=1.21  (1829.0  examples/sec;  0.070  sec/batch)
step 1540, loss=1.35  (1821.9  examples/sec;  0.070  sec/batch)
step 1550, loss=1.20  (1929.8  examples/sec;  0.066  sec/batch)
step 1560, loss=1.18  (1940.8  examples/sec;  0.066  sec/batch)
step 1570, loss=1.19  (1820.8  examples/sec;  0.070  sec/batch)
step 1580, loss=1.01  (1954.3  examples/sec;  0.065  sec/batch)
step 1590, loss=1.09  (1890.5  examples/sec;  0.068  sec/batch)
step 1600, loss=1.14  (1988.3  examples/sec;  0.064  sec/batch)
step 1610, loss=1.00  (1906.1  examples/sec;  0.067  sec/batch)
step 1620, loss=1.07  (1938.0  examples/sec;  0.066  sec/batch)
step 1630, loss=1.17  (1926.3  examples/sec;  0.066  sec/batch)
step 1640, loss=1.16  (1881.1  examples/sec;  0.068  sec/batch)
step 1650, loss=1.28  (1899.0  examples/sec;  0.067  sec/batch)
step 1660, loss=1.11  (1944.9  examples/sec;  0.066  sec/batch)
step 1670, loss=1.15  (1948.2  examples/sec;  0.066  sec/batch)
step 1680, loss=1.27  (1842.7  examples/sec;  0.069  sec/batch)
step 1690, loss=1.22  (1273.3  examples/sec;  0.101  sec/batch)
step 1700, loss=1.29  (1370.6  examples/sec;  0.093  sec/batch)
step 1710, loss=1.13  (1850.1  examples/sec;  0.069  sec/batch)
step 1720, loss=1.08  (1821.4  examples/sec;  0.070  sec/batch)
step 1730, loss=1.30  (1666.5  examples/sec;  0.077  sec/batch)
step 1740, loss=1.24  (1969.4  examples/sec;  0.065  sec/batch)
step 1750, loss=1.24  (1836.9  examples/sec;  0.070  sec/batch)
step 1760, loss=1.23  (1885.1  examples/sec;  0.068  sec/batch)
step 1770, loss=1.22  (1952.7  examples/sec;  0.066  sec/batch)
step 1780, loss=1.29  (1869.3  examples/sec;  0.068  sec/batch)
step 1790, loss=1.22  (1897.8  examples/sec;  0.067  sec/batch)
step 1800, loss=1.18  (1917.6  examples/sec;  0.067  sec/batch)
step 1810, loss=1.08  (1866.4  examples/sec;  0.069  sec/batch)
step 1820, loss=1.11  (1936.7  examples/sec;  0.066  sec/batch)
step 1830, loss=1.11  (1887.3  examples/sec;  0.068  sec/batch)
step 1840, loss=1.16  (1822.4  examples/sec;  0.070  sec/batch)
step 1850, loss=1.14  (1920.3  examples/sec;  0.067  sec/batch)
step 1860, loss=1.24  (1881.4  examples/sec;  0.068  sec/batch)
step 1870, loss=1.16  (1861.4  examples/sec;  0.069  sec/batch)
step 1880, loss=1.12  (1874.7  examples/sec;  0.068  sec/batch)
step 1890, loss=1.40  (1807.6  examples/sec;  0.071  sec/batch)
step 1900, loss=1.07  (1841.3  examples/sec;  0.070  sec/batch)
step 1910, loss=1.04  (1864.1  examples/sec;  0.069  sec/batch)
step 1920, loss=1.07  (1848.8  examples/sec;  0.069  sec/batch)
step 1930, loss=1.20  (1925.7  examples/sec;  0.066  sec/batch)
step 1940, loss=1.24  (1877.0  examples/sec;  0.068  sec/batch)
step 1950, loss=1.15  (1783.1  examples/sec;  0.072  sec/batch)
step 1960, loss=1.04  (1793.5  examples/sec;  0.071  sec/batch)
step 1970, loss=1.15  (1892.9  examples/sec;  0.068  sec/batch)
step 1980, loss=1.05  (1900.8  examples/sec;  0.067  sec/batch)
step 1990, loss=1.25  (1874.4  examples/sec;  0.068  sec/batch)
step 2000, loss=1.19  (1896.0  examples/sec;  0.068  sec/batch)
step 2010, loss=1.13  (1856.8  examples/sec;  0.069  sec/batch)
step 2020, loss=1.27  (1875.1  examples/sec;  0.068  sec/batch)
step 2030, loss=1.07  (1924.8  examples/sec;  0.067  sec/batch)
step 2040, loss=1.20  (1921.8  examples/sec;  0.067  sec/batch)
step 2050, loss=0.99  (1753.9  examples/sec;  0.073  sec/batch)
step 2060, loss=1.07  (1922.3  examples/sec;  0.067  sec/batch)
step 2070, loss=1.09  (1873.5  examples/sec;  0.068  sec/batch)
step 2080, loss=1.03  (1967.7  examples/sec;  0.065  sec/batch)
step 2090, loss=1.02  (1934.2  examples/sec;  0.066  sec/batch)
step 2100, loss=1.30  (1819.6  examples/sec;  0.070  sec/batch)
step 2110, loss=1.14  (1922.2  examples/sec;  0.067  sec/batch)
step 2120, loss=1.13  (1872.1  examples/sec;  0.068  sec/batch)
step 2130, loss=1.05  (1852.9  examples/sec;  0.069  sec/batch)
step 2140, loss=1.19  (1913.8  examples/sec;  0.067  sec/batch)
step 2150, loss=1.04  (1841.7  examples/sec;  0.070  sec/batch)
step 2160, loss=1.14  (1862.8  examples/sec;  0.069  sec/batch)
step 2170, loss=0.90  (1836.7  examples/sec;  0.070  sec/batch)
step 2180, loss=0.99  (1874.3  examples/sec;  0.068  sec/batch)
step 2190, loss=1.15  (1988.5  examples/sec;  0.064  sec/batch)
step 2200, loss=1.09  (1924.3  examples/sec;  0.067  sec/batch)
step 2210, loss=1.25  (1841.0  examples/sec;  0.070  sec/batch)
step 2220, loss=1.22  (1921.9  examples/sec;  0.067  sec/batch)
step 2230, loss=1.18  (1573.4  examples/sec;  0.081  sec/batch)
step 2240, loss=1.14  (1933.3  examples/sec;  0.066  sec/batch)
step 2250, loss=1.09  (1852.0  examples/sec;  0.069  sec/batch)
step 2260, loss=1.12  (1903.5  examples/sec;  0.067  sec/batch)
step 2270, loss=1.19  (1828.6  examples/sec;  0.070  sec/batch)
step 2280, loss=1.15  (1917.1  examples/sec;  0.067  sec/batch)
step 2290, loss=1.10  (1908.5  examples/sec;  0.067  sec/batch)
step 2300, loss=1.08  (1788.0  examples/sec;  0.072  sec/batch)
step 2310, loss=0.99  (1902.0  examples/sec;  0.067  sec/batch)
step 2320, loss=0.97  (1552.6  examples/sec;  0.082  sec/batch)
step 2330, loss=1.07  (1882.6  examples/sec;  0.068  sec/batch)
step 2340, loss=0.96  (1889.0  examples/sec;  0.068  sec/batch)
step 2350, loss=1.03  (1955.1  examples/sec;  0.065  sec/batch)
step 2360, loss=1.09  (1863.1  examples/sec;  0.069  sec/batch)
step 2370, loss=1.14  (1949.9  examples/sec;  0.066  sec/batch)
step 2380, loss=1.10  (1939.5  examples/sec;  0.066  sec/batch)
step 2390, loss=1.14  (1645.1  examples/sec;  0.078  sec/batch)
step 2400, loss=1.11  (1907.2  examples/sec;  0.067  sec/batch)
step 2410, loss=1.08  (1855.7  examples/sec;  0.069  sec/batch)
step 2420, loss=1.03  (1921.9  examples/sec;  0.067  sec/batch)
step 2430, loss=1.02  (1877.4  examples/sec;  0.068  sec/batch)
step 2440, loss=1.16  (1951.1  examples/sec;  0.066  sec/batch)
step 2450, loss=1.05  (1893.5  examples/sec;  0.068  sec/batch)
step 2460, loss=1.14  (1910.5  examples/sec;  0.067  sec/batch)
step 2470, loss=0.97  (1881.8  examples/sec;  0.068  sec/batch)
step 2480, loss=1.08  (1903.1  examples/sec;  0.067  sec/batch)
step 2490, loss=1.12  (1845.3  examples/sec;  0.069  sec/batch)
step 2500, loss=1.00  (1909.7  examples/sec;  0.067  sec/batch)
step 2510, loss=1.04  (1911.4  examples/sec;  0.067  sec/batch)
step 2520, loss=1.03  (1901.9  examples/sec;  0.067  sec/batch)
step 2530, loss=1.14  (1828.0  examples/sec;  0.070  sec/batch)
step 2540, loss=1.06  (1803.9  examples/sec;  0.071  sec/batch)
step 2550, loss=1.02  (1911.8  examples/sec;  0.067  sec/batch)
step 2560, loss=1.11  (1788.4  examples/sec;  0.072  sec/batch)
step 2570, loss=1.04  (1872.4  examples/sec;  0.068  sec/batch)
step 2580, loss=1.15  (1943.8  examples/sec;  0.066  sec/batch)
step 2590, loss=1.06  (1922.6  examples/sec;  0.067  sec/batch)
step 2600, loss=1.14  (1939.7  examples/sec;  0.066  sec/batch)
step 2610, loss=1.17  (1866.1  examples/sec;  0.069  sec/batch)
step 2620, loss=1.30  (1899.5  examples/sec;  0.067  sec/batch)
step 2630, loss=1.02  (1914.3  examples/sec;  0.067  sec/batch)
step 2640, loss=1.11  (1864.7  examples/sec;  0.069  sec/batch)
step 2650, loss=1.13  (1934.5  examples/sec;  0.066  sec/batch)
step 2660, loss=1.04  (1885.3  examples/sec;  0.068  sec/batch)
step 2670, loss=1.04  (1873.4  examples/sec;  0.068  sec/batch)
step 2680, loss=1.07  (1874.7  examples/sec;  0.068  sec/batch)
step 2690, loss=1.15  (1935.6  examples/sec;  0.066  sec/batch)
step 2700, loss=1.07  (1820.3  examples/sec;  0.070  sec/batch)
step 2710, loss=1.20  (1932.6  examples/sec;  0.066  sec/batch)
step 2720, loss=1.07  (1587.7  examples/sec;  0.081  sec/batch)
step 2730, loss=1.06  (1955.4  examples/sec;  0.065  sec/batch)
step 2740, loss=1.03  (1885.2  examples/sec;  0.068  sec/batch)
step 2750, loss=1.07  (1897.2  examples/sec;  0.067  sec/batch)
step 2760, loss=1.07  (1847.6  examples/sec;  0.069  sec/batch)
step 2770, loss=0.92  (1877.9  examples/sec;  0.068  sec/batch)
step 2780, loss=1.07  (1989.8  examples/sec;  0.064  sec/batch)
step 2790, loss=1.02  (1899.5  examples/sec;  0.067  sec/batch)
step 2800, loss=1.16  (1949.9  examples/sec;  0.066  sec/batch)
step 2810, loss=1.07  (1839.5  examples/sec;  0.070  sec/batch)
step 2820, loss=1.02  (1861.4  examples/sec;  0.069  sec/batch)
step 2830, loss=1.04  (1859.8  examples/sec;  0.069  sec/batch)
step 2840, loss=1.10  (1889.7  examples/sec;  0.068  sec/batch)
step 2850, loss=0.98  (1810.5  examples/sec;  0.071  sec/batch)
step 2860, loss=0.99  (1963.1  examples/sec;  0.065  sec/batch)
step 2870, loss=1.16  (1784.2  examples/sec;  0.072  sec/batch)
step 2880, loss=1.01  (1658.8  examples/sec;  0.077  sec/batch)
step 2890, loss=1.06  (1907.5  examples/sec;  0.067  sec/batch)
step 2900, loss=1.08  (1914.4  examples/sec;  0.067  sec/batch)
step 2910, loss=0.84  (1938.2  examples/sec;  0.066  sec/batch)
step 2920, loss=1.17  (1864.6  examples/sec;  0.069  sec/batch)
step 2930, loss=0.97  (1906.9  examples/sec;  0.067  sec/batch)
step 2940, loss=0.90  (1873.8  examples/sec;  0.068  sec/batch)
step 2950, loss=1.00  (1964.8  examples/sec;  0.065  sec/batch)
step 2960, loss=1.05  (1881.8  examples/sec;  0.068  sec/batch)
step 2970, loss=1.00  (1887.5  examples/sec;  0.068  sec/batch)
step 2980, loss=1.07  (1875.2  examples/sec;  0.068  sec/batch)
step 2990, loss=1.00  (1976.6  examples/sec;  0.065  sec/batch)
step 3000, loss=0.85  (1968.7  examples/sec;  0.065  sec/batch)
step 3010, loss=0.91  (1925.3  examples/sec;  0.066  sec/batch)
step 3020, loss=1.16  (1908.1  examples/sec;  0.067  sec/batch)
step 3030, loss=0.89  (1857.3  examples/sec;  0.069  sec/batch)
step 3040, loss=0.92  (1917.3  examples/sec;  0.067  sec/batch)
step 3050, loss=0.97  (1937.5  examples/sec;  0.066  sec/batch)
step 3060, loss=1.02  (1918.3  examples/sec;  0.067  sec/batch)
step 3070, loss=1.16  (1855.8  examples/sec;  0.069  sec/batch)
step 3080, loss=1.01  (1860.6  examples/sec;  0.069  sec/batch)
step 3090, loss=1.16  (1883.4  examples/sec;  0.068  sec/batch)
step 3100, loss=1.14  (1840.8  examples/sec;  0.070  sec/batch)
step 3110, loss=1.11  (1675.9  examples/sec;  0.076  sec/batch)
step 3120, loss=1.12  (1795.9  examples/sec;  0.071  sec/batch)
step 3130, loss=1.12  (1841.9  examples/sec;  0.069  sec/batch)
step 3140, loss=1.04  (1899.9  examples/sec;  0.067  sec/batch)
step 3150, loss=1.11  (1956.4  examples/sec;  0.065  sec/batch)
step 3160, loss=0.93  (1845.3  examples/sec;  0.069  sec/batch)
step 3170, loss=1.02  (1840.3  examples/sec;  0.070  sec/batch)
step 3180, loss=1.28  (1917.5  examples/sec;  0.067  sec/batch)
step 3190, loss=1.16  (1841.3  examples/sec;  0.070  sec/batch)
step 3200, loss=1.11  (1888.1  examples/sec;  0.068  sec/batch)
step 3210, loss=1.09  (1899.8  examples/sec;  0.067  sec/batch)
step 3220, loss=1.03  (1871.1  examples/sec;  0.068  sec/batch)
step 3230, loss=0.77  (1881.7  examples/sec;  0.068  sec/batch)
step 3240, loss=0.98  (1933.6  examples/sec;  0.066  sec/batch)
step 3250, loss=0.90  (1951.4  examples/sec;  0.066  sec/batch)
step 3260, loss=1.15  (1886.3  examples/sec;  0.068  sec/batch)
step 3270, loss=1.05  (1846.1  examples/sec;  0.069  sec/batch)
step 3280, loss=0.85  (1871.0  examples/sec;  0.068  sec/batch)
step 3290, loss=0.96  (2004.9  examples/sec;  0.064  sec/batch)
step 3300, loss=1.06  (1859.2  examples/sec;  0.069  sec/batch)
step 3310, loss=1.02  (1920.6  examples/sec;  0.067  sec/batch)
step 3320, loss=0.97  (1865.7  examples/sec;  0.069  sec/batch)
step 3330, loss=0.94  (1943.3  examples/sec;  0.066  sec/batch)
step 3340, loss=1.02  (1962.1  examples/sec;  0.065  sec/batch)
step 3350, loss=0.96  (1861.5  examples/sec;  0.069  sec/batch)
step 3360, loss=1.06  (1899.1  examples/sec;  0.067  sec/batch)
step 3370, loss=1.00  (1915.6  examples/sec;  0.067  sec/batch)
step 3380, loss=1.01  (1884.4  examples/sec;  0.068  sec/batch)
step 3390, loss=1.29  (1938.2  examples/sec;  0.066  sec/batch)
step 3400, loss=0.97  (1928.3  examples/sec;  0.066  sec/batch)
step 3410, loss=0.80  (1803.7  examples/sec;  0.071  sec/batch)
step 3420, loss=0.98  (1864.0  examples/sec;  0.069  sec/batch)
step 3430, loss=0.98  (1861.2  examples/sec;  0.069  sec/batch)
step 3440, loss=0.98  (1920.3  examples/sec;  0.067  sec/batch)
step 3450, loss=0.89  (1907.8  examples/sec;  0.067  sec/batch)
step 3460, loss=0.97  (1825.0  examples/sec;  0.070  sec/batch)
step 3470, loss=0.89  (1840.7  examples/sec;  0.070  sec/batch)
step 3480, loss=1.00  (1819.4  examples/sec;  0.070  sec/batch)
step 3490, loss=0.89  (1927.4  examples/sec;  0.066  sec/batch)
step 3500, loss=1.21  (1818.1  examples/sec;  0.070  sec/batch)
step 3510, loss=0.87  (1901.0  examples/sec;  0.067  sec/batch)
step 3520, loss=0.86  (1896.8  examples/sec;  0.067  sec/batch)
step 3530, loss=1.02  (1821.4  examples/sec;  0.070  sec/batch)
step 3540, loss=0.97  (1828.9  examples/sec;  0.070  sec/batch)
step 3550, loss=1.01  (1915.5  examples/sec;  0.067  sec/batch)
step 3560, loss=1.00  (1929.1  examples/sec;  0.066  sec/batch)
step 3570, loss=0.90  (1798.7  examples/sec;  0.071  sec/batch)
step 3580, loss=0.91  (1924.3  examples/sec;  0.067  sec/batch)
step 3590, loss=1.00  (1902.6  examples/sec;  0.067  sec/batch)
step 3600, loss=0.98  (1908.7  examples/sec;  0.067  sec/batch)
step 3610, loss=0.97  (1887.9  examples/sec;  0.068  sec/batch)
step 3620, loss=0.97  (1921.1  examples/sec;  0.067  sec/batch)
step 3630, loss=0.96  (1905.6  examples/sec;  0.067  sec/batch)
step 3640, loss=1.02  (1820.1  examples/sec;  0.070  sec/batch)
step 3650, loss=1.06  (1866.2  examples/sec;  0.069  sec/batch)
step 3660, loss=1.13  (1932.0  examples/sec;  0.066  sec/batch)
step 3670, loss=0.97  (1857.7  examples/sec;  0.069  sec/batch)
step 3680, loss=0.98  (1989.5  examples/sec;  0.064  sec/batch)
step 3690, loss=0.97  (1550.0  examples/sec;  0.083  sec/batch)
step 3700, loss=1.06  (1895.0  examples/sec;  0.068  sec/batch)
step 3710, loss=1.07  (1920.6  examples/sec;  0.067  sec/batch)
step 3720, loss=0.84  (1815.5  examples/sec;  0.071  sec/batch)
step 3730, loss=1.15  (1874.0  examples/sec;  0.068  sec/batch)
step 3740, loss=0.94  (1726.2  examples/sec;  0.074  sec/batch)
step 3750, loss=1.10  (1861.0  examples/sec;  0.069  sec/batch)
step 3760, loss=1.09  (1913.1  examples/sec;  0.067  sec/batch)
step 3770, loss=1.13  (1861.3  examples/sec;  0.069  sec/batch)
step 3780, loss=1.04  (1843.9  examples/sec;  0.069  sec/batch)
step 3790, loss=1.12  (1886.8  examples/sec;  0.068  sec/batch)
step 3800, loss=0.94  (1803.7  examples/sec;  0.071  sec/batch)
step 3810, loss=0.94  (1932.1  examples/sec;  0.066  sec/batch)
step 3820, loss=1.13  (1779.1  examples/sec;  0.072  sec/batch)
step 3830, loss=1.14  (1889.6  examples/sec;  0.068  sec/batch)
step 3840, loss=1.01  (1812.6  examples/sec;  0.071  sec/batch)
step 3850, loss=1.02  (1926.8  examples/sec;  0.066  sec/batch)
step 3860, loss=0.99  (1990.1  examples/sec;  0.064  sec/batch)
step 3870, loss=0.98  (1822.5  examples/sec;  0.070  sec/batch)
step 3880, loss=0.77  (1879.2  examples/sec;  0.068  sec/batch)
step 3890, loss=0.91  (1987.0  examples/sec;  0.064  sec/batch)
step 3900, loss=0.98  (1932.3  examples/sec;  0.066  sec/batch)
step 3910, loss=0.85  (1873.2  examples/sec;  0.068  sec/batch)
step 3920, loss=0.85  (1869.1  examples/sec;  0.068  sec/batch)
step 3930, loss=0.95  (1840.8  examples/sec;  0.070  sec/batch)
step 3940, loss=1.01  (1860.8  examples/sec;  0.069  sec/batch)
step 3950, loss=0.88  (1899.7  examples/sec;  0.067  sec/batch)
step 3960, loss=0.92  (1921.3  examples/sec;  0.067  sec/batch)
step 3970, loss=1.05  (1841.0  examples/sec;  0.070  sec/batch)
step 3980, loss=0.94  (1866.1  examples/sec;  0.069  sec/batch)
step 3990, loss=1.01  (1902.2  examples/sec;  0.067  sec/batch)
step 4000, loss=1.08  (1879.5  examples/sec;  0.068  sec/batch)
step 4010, loss=0.91  (1895.7  examples/sec;  0.068  sec/batch)
step 4020, loss=1.14  (1898.2  examples/sec;  0.067  sec/batch)
step 4030, loss=1.08  (1978.3  examples/sec;  0.065  sec/batch)
step 4040, loss=1.01  (1972.8  examples/sec;  0.065  sec/batch)
step 4050, loss=0.76  (1822.4  examples/sec;  0.070  sec/batch)
step 4060, loss=0.95  (1880.9  examples/sec;  0.068  sec/batch)
step 4070, loss=0.90  (1879.8  examples/sec;  0.068  sec/batch)
step 4080, loss=0.93  (1904.6  examples/sec;  0.067  sec/batch)
step 4090, loss=1.02  (1914.9  examples/sec;  0.067  sec/batch)
step 4100, loss=0.95  (1855.2  examples/sec;  0.069  sec/batch)
step 4110, loss=1.03  (1930.4  examples/sec;  0.066  sec/batch)
step 4120, loss=0.78  (1966.0  examples/sec;  0.065  sec/batch)
step 4130, loss=1.06  (1818.2  examples/sec;  0.070  sec/batch)
step 4140, loss=0.99  (1910.9  examples/sec;  0.067  sec/batch)
step 4150, loss=0.90  (1804.2  examples/sec;  0.071  sec/batch)
step 4160, loss=0.79  (1907.6  examples/sec;  0.067  sec/batch)
step 4170, loss=0.95  (1773.2  examples/sec;  0.072  sec/batch)
step 4180, loss=1.12  (1902.0  examples/sec;  0.067  sec/batch)
step 4190, loss=1.14  (1997.8  examples/sec;  0.064  sec/batch)
step 4200, loss=0.93  (1940.0  examples/sec;  0.066  sec/batch)
step 4210, loss=0.82  (1914.9  examples/sec;  0.067  sec/batch)
step 4220, loss=0.82  (1937.1  examples/sec;  0.066  sec/batch)
step 4230, loss=0.91  (1874.7  examples/sec;  0.068  sec/batch)
step 4240, loss=1.00  (1723.1  examples/sec;  0.074  sec/batch)
step 4250, loss=1.09  (1902.1  examples/sec;  0.067  sec/batch)
step 4260, loss=0.96  (1847.5  examples/sec;  0.069  sec/batch)
step 4270, loss=0.93  (1902.5  examples/sec;  0.067  sec/batch)
step 4280, loss=1.24  (1857.9  examples/sec;  0.069  sec/batch)
step 4290, loss=1.02  (1896.4  examples/sec;  0.067  sec/batch)
step 4300, loss=0.88  (1914.2  examples/sec;  0.067  sec/batch)
step 4310, loss=0.89  (1942.2  examples/sec;  0.066  sec/batch)
step 4320, loss=1.02  (1844.3  examples/sec;  0.069  sec/batch)
step 4330, loss=0.98  (1928.7  examples/sec;  0.066  sec/batch)
step 4340, loss=0.99  (1908.3  examples/sec;  0.067  sec/batch)
step 4350, loss=1.06  (1901.9  examples/sec;  0.067  sec/batch)
step 4360, loss=0.97  (1835.3  examples/sec;  0.070  sec/batch)
step 4370, loss=1.07  (1863.1  examples/sec;  0.069  sec/batch)
step 4380, loss=0.92  (1802.1  examples/sec;  0.071  sec/batch)
step 4390, loss=0.87  (1898.9  examples/sec;  0.067  sec/batch)
step 4400, loss=0.93  (1829.0  examples/sec;  0.070  sec/batch)
step 4410, loss=1.09  (1924.8  examples/sec;  0.066  sec/batch)
step 4420, loss=1.01  (1883.5  examples/sec;  0.068  sec/batch)
step 4430, loss=0.85  (1889.7  examples/sec;  0.068  sec/batch)
step 4440, loss=0.93  (1904.8  examples/sec;  0.067  sec/batch)
step 4450, loss=0.91  (1907.9  examples/sec;  0.067  sec/batch)
step 4460, loss=1.03  (1915.7  examples/sec;  0.067  sec/batch)
step 4470, loss=0.99  (1888.7  examples/sec;  0.068  sec/batch)
step 4480, loss=1.13  (1831.2  examples/sec;  0.070  sec/batch)
step 4490, loss=0.80  (1912.6  examples/sec;  0.067  sec/batch)
step 4500, loss=1.07  (1955.1  examples/sec;  0.065  sec/batch)
step 4510, loss=0.92  (1854.8  examples/sec;  0.069  sec/batch)
step 4520, loss=0.82  (1850.7  examples/sec;  0.069  sec/batch)
step 4530, loss=0.96  (1848.6  examples/sec;  0.069  sec/batch)
step 4540, loss=1.00  (1928.2  examples/sec;  0.066  sec/batch)
step 4550, loss=0.89  (1875.7  examples/sec;  0.068  sec/batch)
step 4560, loss=0.86  (1955.6  examples/sec;  0.065  sec/batch)
step 4570, loss=0.91  (1933.3  examples/sec;  0.066  sec/batch)
step 4580, loss=0.89  (1941.7  examples/sec;  0.066  sec/batch)
step 4590, loss=0.99  (1944.6  examples/sec;  0.066  sec/batch)
step 4600, loss=0.93  (1774.7  examples/sec;  0.072  sec/batch)
step 4610, loss=0.99  (1845.2  examples/sec;  0.069  sec/batch)
step 4620, loss=0.98  (1914.1  examples/sec;  0.067  sec/batch)
step 4630, loss=1.01  (1914.2  examples/sec;  0.067  sec/batch)
step 4640, loss=0.92  (1802.5  examples/sec;  0.071  sec/batch)
step 4650, loss=1.06  (1807.0  examples/sec;  0.071  sec/batch)
step 4660, loss=1.00  (1838.4  examples/sec;  0.070  sec/batch)
step 4670, loss=1.09  (1909.8  examples/sec;  0.067  sec/batch)
step 4680, loss=0.93  (1919.7  examples/sec;  0.067  sec/batch)
step 4690, loss=0.93  (1954.0  examples/sec;  0.066  sec/batch)
step 4700, loss=0.96  (1945.8  examples/sec;  0.066  sec/batch)
step 4710, loss=1.15  (1982.0  examples/sec;  0.065  sec/batch)
step 4720, loss=0.89  (1915.0  examples/sec;  0.067  sec/batch)
step 4730, loss=0.94  (1894.1  examples/sec;  0.068  sec/batch)
step 4740, loss=0.93  (1878.7  examples/sec;  0.068  sec/batch)
step 4750, loss=0.84  (1885.5  examples/sec;  0.068  sec/batch)
step 4760, loss=1.09  (1840.2  examples/sec;  0.070  sec/batch)
step 4770, loss=0.83  (1972.6  examples/sec;  0.065  sec/batch)
step 4780, loss=0.91  (1837.6  examples/sec;  0.070  sec/batch)
step 4790, loss=0.97  (1910.1  examples/sec;  0.067  sec/batch)
step 4800, loss=0.92  (1898.4  examples/sec;  0.067  sec/batch)
step 4810, loss=0.86  (1811.2  examples/sec;  0.071  sec/batch)
step 4820, loss=0.91  (1869.4  examples/sec;  0.068  sec/batch)
step 4830, loss=0.91  (1898.3  examples/sec;  0.067  sec/batch)
step 4840, loss=0.83  (1936.0  examples/sec;  0.066  sec/batch)
step 4850, loss=0.94  (1871.2  examples/sec;  0.068  sec/batch)
step 4860, loss=0.94  (1954.2  examples/sec;  0.066  sec/batch)
step 4870, loss=0.97  (1879.6  examples/sec;  0.068  sec/batch)
step 4880, loss=1.07  (1994.0  examples/sec;  0.064  sec/batch)
step 4890, loss=1.05  (1736.2  examples/sec;  0.074  sec/batch)
step 4900, loss=0.94  (1834.4  examples/sec;  0.070  sec/batch)
step 4910, loss=0.99  (1887.2  examples/sec;  0.068  sec/batch)
step 4920, loss=1.04  (1934.5  examples/sec;  0.066  sec/batch)
step 4930, loss=1.01  (1703.4  examples/sec;  0.075  sec/batch)
step 4940, loss=0.97  (1883.1  examples/sec;  0.068  sec/batch)
step 4950, loss=0.99  (1886.6  examples/sec;  0.068  sec/batch)
step 4960, loss=0.93  (1877.5  examples/sec;  0.068  sec/batch)
step 4970, loss=0.97  (1837.9  examples/sec;  0.070  sec/batch)
step 4980, loss=1.06  (1939.6  examples/sec;  0.066  sec/batch)
step 4990, loss=1.09  (1877.4  examples/sec;  0.068  sec/batch)
step 5000, loss=0.81  (1863.9  examples/sec;  0.069  sec/batch)
step 5010, loss=0.96  (1885.7  examples/sec;  0.068  sec/batch)
step 5020, loss=1.13  (1589.6  examples/sec;  0.081  sec/batch)
step 5030, loss=0.90  (1917.5  examples/sec;  0.067  sec/batch)
step 5040, loss=0.84  (1769.8  examples/sec;  0.072  sec/batch)
step 5050, loss=0.90  (1802.7  examples/sec;  0.071  sec/batch)
step 5060, loss=0.79  (1886.2  examples/sec;  0.068  sec/batch)
step 5070, loss=1.10  (1867.7  examples/sec;  0.069  sec/batch)
step 5080, loss=1.00  (1909.2  examples/sec;  0.067  sec/batch)
step 5090, loss=0.95  (1941.6  examples/sec;  0.066  sec/batch)
step 5100, loss=0.89  (1920.7  examples/sec;  0.067  sec/batch)
step 5110, loss=1.10  (1637.2  examples/sec;  0.078  sec/batch)
step 5120, loss=0.92  (1883.8  examples/sec;  0.068  sec/batch)
step 5130, loss=1.23  (1917.6  examples/sec;  0.067  sec/batch)
step 5140, loss=1.03  (1941.5  examples/sec;  0.066  sec/batch)
step 5150, loss=0.97  (1905.3  examples/sec;  0.067  sec/batch)
step 5160, loss=0.82  (1838.5  examples/sec;  0.070  sec/batch)
step 5170, loss=1.02  (1934.6  examples/sec;  0.066  sec/batch)
step 5180, loss=1.08  (1883.9  examples/sec;  0.068  sec/batch)
step 5190, loss=0.85  (1853.0  examples/sec;  0.069  sec/batch)
step 5200, loss=0.86  (1828.4  examples/sec;  0.070  sec/batch)
step 5210, loss=0.74  (1940.6  examples/sec;  0.066  sec/batch)
step 5220, loss=0.72  (1916.0  examples/sec;  0.067  sec/batch)
step 5230, loss=0.85  (1899.6  examples/sec;  0.067  sec/batch)
step 5240, loss=0.89  (1784.0  examples/sec;  0.072  sec/batch)
step 5250, loss=0.90  (1874.0  examples/sec;  0.068  sec/batch)
step 5260, loss=1.08  (1895.6  examples/sec;  0.068  sec/batch)
step 5270, loss=0.85  (1939.6  examples/sec;  0.066  sec/batch)
step 5280, loss=1.10  (1939.9  examples/sec;  0.066  sec/batch)
step 5290, loss=0.96  (1847.8  examples/sec;  0.069  sec/batch)
step 5300, loss=0.81  (1971.0  examples/sec;  0.065  sec/batch)
step 5310, loss=1.00  (1892.3  examples/sec;  0.068  sec/batch)
step 5320, loss=1.11  (1856.5  examples/sec;  0.069  sec/batch)
step 5330, loss=0.92  (1944.8  examples/sec;  0.066  sec/batch)
step 5340, loss=0.96  (1807.5  examples/sec;  0.071  sec/batch)
step 5350, loss=1.00  (1876.9  examples/sec;  0.068  sec/batch)
step 5360, loss=1.16  (1940.9  examples/sec;  0.066  sec/batch)
step 5370, loss=0.90  (1862.1  examples/sec;  0.069  sec/batch)
step 5380, loss=0.90  (1921.6  examples/sec;  0.067  sec/batch)
step 5390, loss=1.03  (1896.5  examples/sec;  0.067  sec/batch)
step 5400, loss=0.94  (1910.3  examples/sec;  0.067  sec/batch)
step 5410, loss=0.98  (1903.6  examples/sec;  0.067  sec/batch)
step 5420, loss=0.95  (1862.5  examples/sec;  0.069  sec/batch)
step 5430, loss=1.09  (1860.7  examples/sec;  0.069  sec/batch)
step 5440, loss=0.67  (1880.0  examples/sec;  0.068  sec/batch)
step 5450, loss=0.99  (1904.8  examples/sec;  0.067  sec/batch)
step 5460, loss=0.93  (1903.0  examples/sec;  0.067  sec/batch)
step 5470, loss=0.97  (1928.5  examples/sec;  0.066  sec/batch)
step 5480, loss=1.04  (1793.6  examples/sec;  0.071  sec/batch)
step 5490, loss=0.83  (1927.0  examples/sec;  0.066  sec/batch)
step 5500, loss=0.87  (1955.9  examples/sec;  0.065  sec/batch)
step 5510, loss=1.02  (1836.3  examples/sec;  0.070  sec/batch)
step 5520, loss=0.99  (1484.3  examples/sec;  0.086  sec/batch)
step 5530, loss=0.99  (1862.1  examples/sec;  0.069  sec/batch)
step 5540, loss=0.83  (2029.9  examples/sec;  0.063  sec/batch)
step 5550, loss=0.99  (1768.2  examples/sec;  0.072  sec/batch)
step 5560, loss=0.86  (1887.1  examples/sec;  0.068  sec/batch)
step 5570, loss=1.03  (1922.3  examples/sec;  0.067  sec/batch)
step 5580, loss=1.07  (1870.7  examples/sec;  0.068  sec/batch)
step 5590, loss=0.95  (1670.1  examples/sec;  0.077  sec/batch)
step 5600, loss=0.87  (1813.6  examples/sec;  0.071  sec/batch)
step 5610, loss=0.90  (1891.8  examples/sec;  0.068  sec/batch)
step 5620, loss=1.08  (1911.9  examples/sec;  0.067  sec/batch)
step 5630, loss=0.97  (1803.1  examples/sec;  0.071  sec/batch)
step 5640, loss=0.96  (1876.6  examples/sec;  0.068  sec/batch)
step 5650, loss=0.91  (1903.6  examples/sec;  0.067  sec/batch)
step 5660, loss=0.83  (1819.2  examples/sec;  0.070  sec/batch)
step 5670, loss=0.73  (1844.3  examples/sec;  0.069  sec/batch)
step 5680, loss=0.78  (1886.0  examples/sec;  0.068  sec/batch)
step 5690, loss=1.06  (1928.0  examples/sec;  0.066  sec/batch)
step 5700, loss=0.85  (1851.4  examples/sec;  0.069  sec/batch)
step 5710, loss=0.84  (1916.3  examples/sec;  0.067  sec/batch)
step 5720, loss=0.77  (1901.3  examples/sec;  0.067  sec/batch)
step 5730, loss=0.86  (1866.5  examples/sec;  0.069  sec/batch)
step 5740, loss=1.11  (1799.5  examples/sec;  0.071  sec/batch)
step 5750, loss=0.92  (1903.8  examples/sec;  0.067  sec/batch)
step 5760, loss=0.91  (1554.8  examples/sec;  0.082  sec/batch)
step 5770, loss=1.02  (1989.6  examples/sec;  0.064  sec/batch)
step 5780, loss=0.94  (1914.7  examples/sec;  0.067  sec/batch)
step 5790, loss=0.97  (1845.3  examples/sec;  0.069  sec/batch)
step 5800, loss=0.86  (1894.7  examples/sec;  0.068  sec/batch)
step 5810, loss=0.85  (1905.9  examples/sec;  0.067  sec/batch)
step 5820, loss=0.85  (1931.4  examples/sec;  0.066  sec/batch)
step 5830, loss=0.86  (1800.7  examples/sec;  0.071  sec/batch)
step 5840, loss=0.93  (1961.1  examples/sec;  0.065  sec/batch)
step 5850, loss=0.95  (1918.1  examples/sec;  0.067  sec/batch)
step 5860, loss=1.05  (1969.2  examples/sec;  0.065  sec/batch)
step 5870, loss=1.06  (1853.6  examples/sec;  0.069  sec/batch)
step 5880, loss=0.85  (1882.2  examples/sec;  0.068  sec/batch)
step 5890, loss=0.88  (1821.1  examples/sec;  0.070  sec/batch)
step 5900, loss=0.83  (1906.0  examples/sec;  0.067  sec/batch)
step 5910, loss=0.92  (1953.4  examples/sec;  0.066  sec/batch)
step 5920, loss=1.00  (1949.1  examples/sec;  0.066  sec/batch)
step 5930, loss=0.98  (1843.7  examples/sec;  0.069  sec/batch)
step 5940, loss=0.99  (1971.9  examples/sec;  0.065  sec/batch)
step 5950, loss=0.90  (1848.8  examples/sec;  0.069  sec/batch)
step 5960, loss=0.86  (1923.7  examples/sec;  0.067  sec/batch)
step 5970, loss=0.76  (1946.9  examples/sec;  0.066  sec/batch)
step 5980, loss=0.99  (1830.6  examples/sec;  0.070  sec/batch)
step 5990, loss=0.91  (1871.6  examples/sec;  0.068  sec/batch)
step 6000, loss=0.87  (1871.5  examples/sec;  0.068  sec/batch)
step 6010, loss=0.97  (1972.4  examples/sec;  0.065  sec/batch)
step 6020, loss=0.90  (1853.8  examples/sec;  0.069  sec/batch)
step 6030, loss=0.78  (1926.7  examples/sec;  0.066  sec/batch)
step 6040, loss=0.88  (1812.5  examples/sec;  0.071  sec/batch)
step 6050, loss=0.93  (1836.6  examples/sec;  0.070  sec/batch)
step 6060, loss=0.90  (1790.1  examples/sec;  0.072  sec/batch)
step 6070, loss=0.86  (1781.2  examples/sec;  0.072  sec/batch)
step 6080, loss=0.86  (1928.7  examples/sec;  0.066  sec/batch)
step 6090, loss=0.95  (1869.6  examples/sec;  0.068  sec/batch)
step 6100, loss=1.11  (1860.1  examples/sec;  0.069  sec/batch)
step 6110, loss=0.88  (1932.9  examples/sec;  0.066  sec/batch)
step 6120, loss=0.88  (1995.3  examples/sec;  0.064  sec/batch)
step 6130, loss=0.97  (1848.8  examples/sec;  0.069  sec/batch)
step 6140, loss=0.80  (1810.1  examples/sec;  0.071  sec/batch)
step 6150, loss=0.94  (1849.6  examples/sec;  0.069  sec/batch)
step 6160, loss=0.96  (1588.3  examples/sec;  0.081  sec/batch)
step 6170, loss=0.89  (1858.6  examples/sec;  0.069  sec/batch)
step 6180, loss=1.09  (1812.9  examples/sec;  0.071  sec/batch)
step 6190, loss=0.86  (1884.9  examples/sec;  0.068  sec/batch)
step 6200, loss=0.94  (1835.5  examples/sec;  0.070  sec/batch)
step 6210, loss=0.90  (1869.6  examples/sec;  0.068  sec/batch)
step 6220, loss=1.07  (1908.0  examples/sec;  0.067  sec/batch)
step 6230, loss=0.94  (1881.7  examples/sec;  0.068  sec/batch)
step 6240, loss=0.86  (1930.2  examples/sec;  0.066  sec/batch)
step 6250, loss=0.68  (1826.9  examples/sec;  0.070  sec/batch)
step 6260, loss=0.94  (1870.8  examples/sec;  0.068  sec/batch)
step 6270, loss=0.97  (1888.6  examples/sec;  0.068  sec/batch)
step 6280, loss=0.92  (1882.7  examples/sec;  0.068  sec/batch)
step 6290, loss=1.00  (1855.6  examples/sec;  0.069  sec/batch)
step 6300, loss=1.03  (1672.6  examples/sec;  0.077  sec/batch)
step 6310, loss=0.90  (1927.1  examples/sec;  0.066  sec/batch)
step 6320, loss=0.93  (1910.1  examples/sec;  0.067  sec/batch)
step 6330, loss=0.77  (1839.4  examples/sec;  0.070  sec/batch)
step 6340, loss=0.73  (1862.2  examples/sec;  0.069  sec/batch)
step 6350, loss=0.88  (1843.5  examples/sec;  0.069  sec/batch)
step 6360, loss=1.15  (1850.5  examples/sec;  0.069  sec/batch)
step 6370, loss=1.09  (1900.2  examples/sec;  0.067  sec/batch)
step 6380, loss=0.78  (1942.7  examples/sec;  0.066  sec/batch)
step 6390, loss=0.81  (1866.0  examples/sec;  0.069  sec/batch)
step 6400, loss=0.86  (1812.2  examples/sec;  0.071  sec/batch)
step 6410, loss=0.83  (1876.2  examples/sec;  0.068  sec/batch)
step 6420, loss=0.83  (1877.4  examples/sec;  0.068  sec/batch)
step 6430, loss=0.91  (1764.4  examples/sec;  0.073  sec/batch)
step 6440, loss=0.97  (1744.8  examples/sec;  0.073  sec/batch)
step 6450, loss=1.02  (1805.4  examples/sec;  0.071  sec/batch)
step 6460, loss=0.93  (1843.1  examples/sec;  0.069  sec/batch)
step 6470, loss=0.89  (1871.5  examples/sec;  0.068  sec/batch)
step 6480, loss=0.78  (1864.4  examples/sec;  0.069  sec/batch)
step 6490, loss=0.79  (1881.4  examples/sec;  0.068  sec/batch)
step 6500, loss=0.84  (1563.0  examples/sec;  0.082  sec/batch)
step 6510, loss=0.88  (1865.6  examples/sec;  0.069  sec/batch)
step 6520, loss=0.98  (1844.4  examples/sec;  0.069  sec/batch)
step 6530, loss=0.82  (1926.1  examples/sec;  0.066  sec/batch)
step 6540, loss=0.92  (1912.4  examples/sec;  0.067  sec/batch)
step 6550, loss=0.76  (1890.5  examples/sec;  0.068  sec/batch)
step 6560, loss=0.99  (1875.3  examples/sec;  0.068  sec/batch)
step 6570, loss=0.78  (1873.8  examples/sec;  0.068  sec/batch)
step 6580, loss=0.97  (1872.8  examples/sec;  0.068  sec/batch)
step 6590, loss=0.97  (1908.2  examples/sec;  0.067  sec/batch)
step 6600, loss=0.84  (1869.4  examples/sec;  0.068  sec/batch)
step 6610, loss=0.80  (1918.0  examples/sec;  0.067  sec/batch)
step 6620, loss=0.99  (1850.0  examples/sec;  0.069  sec/batch)
step 6630, loss=0.88  (1900.9  examples/sec;  0.067  sec/batch)
step 6640, loss=0.92  (1841.9  examples/sec;  0.069  sec/batch)
step 6650, loss=1.00  (1853.5  examples/sec;  0.069  sec/batch)
step 6660, loss=1.13  (1822.3  examples/sec;  0.070  sec/batch)
step 6670, loss=0.84  (1997.8  examples/sec;  0.064  sec/batch)
step 6680, loss=0.93  (1906.7  examples/sec;  0.067  sec/batch)
step 6690, loss=0.96  (1786.9  examples/sec;  0.072  sec/batch)
step 6700, loss=0.79  (1959.3  examples/sec;  0.065  sec/batch)
step 6710, loss=0.83  (1859.3  examples/sec;  0.069  sec/batch)
step 6720, loss=0.89  (1966.6  examples/sec;  0.065  sec/batch)
step 6730, loss=0.79  (1809.7  examples/sec;  0.071  sec/batch)
step 6740, loss=0.83  (2001.0  examples/sec;  0.064  sec/batch)
step 6750, loss=0.87  (1788.1  examples/sec;  0.072  sec/batch)
step 6760, loss=0.81  (1885.5  examples/sec;  0.068  sec/batch)
step 6770, loss=0.89  (1922.5  examples/sec;  0.067  sec/batch)
step 6780, loss=0.99  (1875.2  examples/sec;  0.068  sec/batch)
step 6790, loss=0.94  (1968.4  examples/sec;  0.065  sec/batch)
step 6800, loss=0.97  (1866.6  examples/sec;  0.069  sec/batch)
step 6810, loss=1.00  (1911.3  examples/sec;  0.067  sec/batch)
step 6820, loss=0.83  (1823.6  examples/sec;  0.070  sec/batch)
step 6830, loss=0.96  (1859.4  examples/sec;  0.069  sec/batch)
step 6840, loss=0.94  (1958.2  examples/sec;  0.065  sec/batch)
step 6850, loss=0.89  (1862.6  examples/sec;  0.069  sec/batch)
step 6860, loss=1.03  (1953.6  examples/sec;  0.066  sec/batch)
step 6870, loss=0.87  (1825.7  examples/sec;  0.070  sec/batch)
step 6880, loss=0.96  (1881.3  examples/sec;  0.068  sec/batch)
step 6890, loss=0.84  (1919.1  examples/sec;  0.067  sec/batch)
step 6900, loss=0.91  (1867.4  examples/sec;  0.069  sec/batch)
step 6910, loss=0.79  (1914.0  examples/sec;  0.067  sec/batch)
step 6920, loss=0.70  (1902.4  examples/sec;  0.067  sec/batch)
step 6930, loss=0.91  (1919.7  examples/sec;  0.067  sec/batch)
step 6940, loss=0.95  (1945.2  examples/sec;  0.066  sec/batch)
step 6950, loss=0.96  (1803.5  examples/sec;  0.071  sec/batch)
step 6960, loss=0.91  (1861.6  examples/sec;  0.069  sec/batch)
step 6970, loss=0.75  (1877.2  examples/sec;  0.068  sec/batch)
step 6980, loss=0.74  (1905.2  examples/sec;  0.067  sec/batch)
step 6990, loss=0.98  (1881.3  examples/sec;  0.068  sec/batch)
step 7000, loss=0.84  (2004.7  examples/sec;  0.064  sec/batch)
step 7010, loss=0.97  (1900.1  examples/sec;  0.067  sec/batch)
step 7020, loss=1.06  (1899.3  examples/sec;  0.067  sec/batch)
step 7030, loss=0.81  (1846.8  examples/sec;  0.069  sec/batch)
step 7040, loss=0.84  (1704.4  examples/sec;  0.075  sec/batch)
step 7050, loss=0.85  (1867.1  examples/sec;  0.069  sec/batch)
step 7060, loss=0.80  (1777.6  examples/sec;  0.072  sec/batch)
step 7070, loss=0.95  (1924.9  examples/sec;  0.066  sec/batch)
step 7080, loss=0.90  (1884.0  examples/sec;  0.068  sec/batch)
step 7090, loss=0.86  (1915.2  examples/sec;  0.067  sec/batch)
step 7100, loss=0.93  (1903.4  examples/sec;  0.067  sec/batch)
step 7110, loss=0.74  (1868.2  examples/sec;  0.069  sec/batch)
step 7120, loss=1.02  (1896.2  examples/sec;  0.068  sec/batch)
step 7130, loss=0.83  (1903.2  examples/sec;  0.067  sec/batch)
step 7140, loss=0.77  (1864.1  examples/sec;  0.069  sec/batch)
step 7150, loss=0.89  (1854.5  examples/sec;  0.069  sec/batch)
step 7160, loss=0.88  (1734.8  examples/sec;  0.074  sec/batch)
step 7170, loss=0.87  (1914.9  examples/sec;  0.067  sec/batch)
step 7180, loss=0.76  (1934.3  examples/sec;  0.066  sec/batch)
step 7190, loss=1.05  (1871.7  examples/sec;  0.068  sec/batch)
step 7200, loss=1.00  (1903.6  examples/sec;  0.067  sec/batch)
step 7210, loss=0.76  (1921.0  examples/sec;  0.067  sec/batch)
step 7220, loss=1.00  (1986.2  examples/sec;  0.064  sec/batch)
step 7230, loss=1.06  (1878.6  examples/sec;  0.068  sec/batch)
step 7240, loss=0.96  (1957.6  examples/sec;  0.065  sec/batch)
step 7250, loss=0.75  (1853.2  examples/sec;  0.069  sec/batch)
step 7260, loss=0.94  (1881.3  examples/sec;  0.068  sec/batch)
step 7270, loss=0.84  (1891.2  examples/sec;  0.068  sec/batch)
step 7280, loss=0.88  (1899.7  examples/sec;  0.067  sec/batch)
step 7290, loss=1.09  (1899.7  examples/sec;  0.067  sec/batch)
step 7300, loss=0.79  (1868.5  examples/sec;  0.069  sec/batch)
step 7310, loss=0.75  (1936.1  examples/sec;  0.066  sec/batch)
step 7320, loss=0.96  (1642.6  examples/sec;  0.078  sec/batch)
step 7330, loss=0.85  (1945.2  examples/sec;  0.066  sec/batch)
step 7340, loss=0.77  (1894.0  examples/sec;  0.068  sec/batch)
step 7350, loss=0.83  (1876.1  examples/sec;  0.068  sec/batch)
step 7360, loss=0.86  (1854.7  examples/sec;  0.069  sec/batch)
step 7370, loss=0.85  (1883.2  examples/sec;  0.068  sec/batch)
step 7380, loss=0.62  (1861.7  examples/sec;  0.069  sec/batch)
step 7390, loss=0.97  (1858.3  examples/sec;  0.069  sec/batch)
step 7400, loss=1.02  (1907.2  examples/sec;  0.067  sec/batch)
step 7410, loss=0.80  (1895.3  examples/sec;  0.068  sec/batch)
step 7420, loss=0.82  (1922.8  examples/sec;  0.067  sec/batch)
step 7430, loss=0.99  (1885.0  examples/sec;  0.068  sec/batch)
step 7440, loss=0.88  (1846.6  examples/sec;  0.069  sec/batch)
step 7450, loss=0.80  (1923.4  examples/sec;  0.067  sec/batch)
step 7460, loss=0.92  (1786.2  examples/sec;  0.072  sec/batch)
step 7470, loss=0.86  (1922.2  examples/sec;  0.067  sec/batch)
step 7480, loss=0.90  (1928.8  examples/sec;  0.066  sec/batch)
step 7490, loss=0.94  (1827.1  examples/sec;  0.070  sec/batch)
step 7500, loss=0.87  (1866.0  examples/sec;  0.069  sec/batch)
step 7510, loss=0.87  (1978.6  examples/sec;  0.065  sec/batch)
step 7520, loss=0.78  (1894.5  examples/sec;  0.068  sec/batch)
step 7530, loss=0.94  (1912.0  examples/sec;  0.067  sec/batch)
step 7540, loss=0.93  (1913.8  examples/sec;  0.067  sec/batch)
step 7550, loss=0.74  (1942.6  examples/sec;  0.066  sec/batch)
step 7560, loss=0.91  (1894.0  examples/sec;  0.068  sec/batch)
step 7570, loss=0.76  (1856.6  examples/sec;  0.069  sec/batch)
step 7580, loss=0.96  (1913.8  examples/sec;  0.067  sec/batch)
step 7590, loss=0.82  (1938.0  examples/sec;  0.066  sec/batch)
step 7600, loss=0.83  (1875.2  examples/sec;  0.068  sec/batch)
step 7610, loss=0.87  (1866.9  examples/sec;  0.069  sec/batch)
step 7620, loss=0.94  (1812.1  examples/sec;  0.071  sec/batch)
step 7630, loss=1.02  (1881.4  examples/sec;  0.068  sec/batch)
step 7640, loss=0.92  (1692.5  examples/sec;  0.076  sec/batch)
step 7650, loss=0.89  (1877.3  examples/sec;  0.068  sec/batch)
step 7660, loss=0.81  (1856.2  examples/sec;  0.069  sec/batch)
step 7670, loss=0.78  (1904.7  examples/sec;  0.067  sec/batch)
step 7680, loss=0.68  (1902.0  examples/sec;  0.067  sec/batch)
step 7690, loss=0.77  (1847.4  examples/sec;  0.069  sec/batch)
step 7700, loss=0.96  (1850.3  examples/sec;  0.069  sec/batch)
step 7710, loss=0.83  (1906.6  examples/sec;  0.067  sec/batch)
step 7720, loss=0.86  (1845.9  examples/sec;  0.069  sec/batch)
step 7730, loss=0.99  (1914.2  examples/sec;  0.067  sec/batch)
step 7740, loss=0.73  (1683.0  examples/sec;  0.076  sec/batch)
step 7750, loss=0.82  (1847.9  examples/sec;  0.069  sec/batch)
step 7760, loss=0.88  (1915.2  examples/sec;  0.067  sec/batch)
step 7770, loss=0.92  (1930.5  examples/sec;  0.066  sec/batch)
step 7780, loss=0.66  (1949.4  examples/sec;  0.066  sec/batch)
step 7790, loss=0.86  (1996.3  examples/sec;  0.064  sec/batch)
step 7800, loss=0.85  (1888.7  examples/sec;  0.068  sec/batch)
step 7810, loss=0.89  (1890.9  examples/sec;  0.068  sec/batch)
step 7820, loss=0.75  (1859.5  examples/sec;  0.069  sec/batch)
step 7830, loss=0.94  (1959.7  examples/sec;  0.065  sec/batch)
step 7840, loss=0.91  (1943.1  examples/sec;  0.066  sec/batch)
step 7850, loss=0.88  (1953.8  examples/sec;  0.066  sec/batch)
step 7860, loss=0.76  (1877.1  examples/sec;  0.068  sec/batch)
step 7870, loss=0.77  (1884.9  examples/sec;  0.068  sec/batch)
step 7880, loss=0.93  (1775.3  examples/sec;  0.072  sec/batch)
step 7890, loss=0.86  (1869.7  examples/sec;  0.068  sec/batch)
step 7900, loss=1.09  (1855.3  examples/sec;  0.069  sec/batch)
step 7910, loss=0.92  (1939.4  examples/sec;  0.066  sec/batch)
step 7920, loss=0.87  (1870.0  examples/sec;  0.068  sec/batch)
step 7930, loss=0.87  (1794.0  examples/sec;  0.071  sec/batch)
step 7940, loss=0.86  (1765.2  examples/sec;  0.073  sec/batch)
step 7950, loss=0.97  (1960.3  examples/sec;  0.065  sec/batch)
step 7960, loss=0.97  (1893.2  examples/sec;  0.068  sec/batch)
step 7970, loss=0.84  (1916.6  examples/sec;  0.067  sec/batch)
step 7980, loss=0.85  (1997.5  examples/sec;  0.064  sec/batch)
step 7990, loss=1.00  (1878.2  examples/sec;  0.068  sec/batch)
step 8000, loss=0.78  (1918.3  examples/sec;  0.067  sec/batch)
step 8010, loss=0.80  (1896.4  examples/sec;  0.067  sec/batch)
step 8020, loss=0.81  (1833.1  examples/sec;  0.070  sec/batch)
step 8030, loss=0.85  (1868.9  examples/sec;  0.068  sec/batch)
step 8040, loss=0.90  (1873.1  examples/sec;  0.068  sec/batch)
step 8050, loss=0.79  (1929.0  examples/sec;  0.066  sec/batch)
step 8060, loss=0.93  (1889.5  examples/sec;  0.068  sec/batch)
step 8070, loss=0.65  (1916.3  examples/sec;  0.067  sec/batch)
step 8080, loss=0.92  (1943.0  examples/sec;  0.066  sec/batch)
step 8090, loss=0.92  (1804.7  examples/sec;  0.071  sec/batch)
step 8100, loss=0.90  (1824.0  examples/sec;  0.070  sec/batch)
step 8110, loss=1.03  (1872.5  examples/sec;  0.068  sec/batch)
step 8120, loss=0.90  (1772.0  examples/sec;  0.072  sec/batch)
step 8130, loss=0.95  (1894.2  examples/sec;  0.068  sec/batch)
step 8140, loss=0.99  (1854.5  examples/sec;  0.069  sec/batch)
step 8150, loss=0.82  (2052.3  examples/sec;  0.062  sec/batch)
step 8160, loss=0.75  (1685.9  examples/sec;  0.076  sec/batch)
step 8170, loss=0.88  (1838.6  examples/sec;  0.070  sec/batch)
step 8180, loss=0.85  (1892.1  examples/sec;  0.068  sec/batch)
step 8190, loss=0.82  (1920.3  examples/sec;  0.067  sec/batch)
step 8200, loss=0.78  (1808.2  examples/sec;  0.071  sec/batch)
step 8210, loss=1.02  (1835.4  examples/sec;  0.070  sec/batch)
step 8220, loss=0.80  (1843.4  examples/sec;  0.069  sec/batch)
step 8230, loss=0.94  (1890.8  examples/sec;  0.068  sec/batch)
step 8240, loss=0.84  (1878.4  examples/sec;  0.068  sec/batch)
step 8250, loss=0.90  (1873.2  examples/sec;  0.068  sec/batch)
step 8260, loss=0.73  (1803.6  examples/sec;  0.071  sec/batch)
step 8270, loss=0.71  (1633.6  examples/sec;  0.078  sec/batch)
step 8280, loss=0.98  (1862.2  examples/sec;  0.069  sec/batch)
step 8290, loss=0.96  (1905.8  examples/sec;  0.067  sec/batch)
step 8300, loss=0.86  (1833.9  examples/sec;  0.070  sec/batch)
step 8310, loss=0.89  (1868.9  examples/sec;  0.068  sec/batch)
step 8320, loss=0.71  (1888.0  examples/sec;  0.068  sec/batch)
step 8330, loss=0.91  (1775.7  examples/sec;  0.072  sec/batch)
step 8340, loss=0.92  (1833.3  examples/sec;  0.070  sec/batch)
step 8350, loss=0.72  (1850.8  examples/sec;  0.069  sec/batch)
step 8360, loss=0.96  (1910.8  examples/sec;  0.067  sec/batch)
step 8370, loss=0.69  (1781.9  examples/sec;  0.072  sec/batch)
step 8380, loss=0.87  (1851.2  examples/sec;  0.069  sec/batch)
step 8390, loss=0.93  (1853.0  examples/sec;  0.069  sec/batch)
step 8400, loss=0.85  (1884.4  examples/sec;  0.068  sec/batch)
step 8410, loss=0.95  (1904.1  examples/sec;  0.067  sec/batch)
step 8420, loss=0.97  (1914.9  examples/sec;  0.067  sec/batch)
step 8430, loss=0.88  (1883.3  examples/sec;  0.068  sec/batch)
step 8440, loss=0.90  (1873.6  examples/sec;  0.068  sec/batch)
step 8450, loss=1.01  (1878.6  examples/sec;  0.068  sec/batch)
step 8460, loss=0.90  (1913.1  examples/sec;  0.067  sec/batch)
step 8470, loss=0.77  (1908.5  examples/sec;  0.067  sec/batch)
step 8480, loss=0.93  (1839.0  examples/sec;  0.070  sec/batch)
step 8490, loss=0.86  (1868.2  examples/sec;  0.069  sec/batch)
step 8500, loss=0.86  (1931.5  examples/sec;  0.066  sec/batch)
step 8510, loss=0.92  (1848.0  examples/sec;  0.069  sec/batch)
step 8520, loss=0.92  (1844.7  examples/sec;  0.069  sec/batch)
step 8530, loss=0.93  (1968.5  examples/sec;  0.065  sec/batch)
step 8540, loss=0.82  (1921.0  examples/sec;  0.067  sec/batch)
step 8550, loss=0.81  (1807.7  examples/sec;  0.071  sec/batch)
step 8560, loss=0.90  (1872.2  examples/sec;  0.068  sec/batch)
step 8570, loss=0.94  (1870.9  examples/sec;  0.068  sec/batch)
step 8580, loss=1.23  (1846.2  examples/sec;  0.069  sec/batch)
step 8590, loss=0.90  (1737.2  examples/sec;  0.074  sec/batch)
step 8600, loss=0.76  (1981.2  examples/sec;  0.065  sec/batch)
step 8610, loss=0.97  (2022.8  examples/sec;  0.063  sec/batch)
step 8620, loss=0.92  (1863.2  examples/sec;  0.069  sec/batch)
step 8630, loss=0.99  (1900.3  examples/sec;  0.067  sec/batch)
step 8640, loss=0.69  (1897.0  examples/sec;  0.067  sec/batch)
step 8650, loss=0.90  (2002.7  examples/sec;  0.064  sec/batch)
step 8660, loss=0.92  (1841.6  examples/sec;  0.070  sec/batch)
step 8670, loss=0.90  (1856.6  examples/sec;  0.069  sec/batch)
step 8680, loss=0.87  (2019.2  examples/sec;  0.063  sec/batch)
step 8690, loss=0.95  (1872.8  examples/sec;  0.068  sec/batch)
step 8700, loss=0.96  (1932.0  examples/sec;  0.066  sec/batch)
step 8710, loss=0.79  (1902.5  examples/sec;  0.067  sec/batch)
step 8720, loss=0.85  (1909.4  examples/sec;  0.067  sec/batch)
step 8730, loss=1.12  (1915.0  examples/sec;  0.067  sec/batch)
step 8740, loss=0.88  (1846.6  examples/sec;  0.069  sec/batch)
step 8750, loss=0.97  (1804.2  examples/sec;  0.071  sec/batch)
step 8760, loss=0.92  (1833.8  examples/sec;  0.070  sec/batch)
step 8770, loss=0.82  (1948.4  examples/sec;  0.066  sec/batch)
step 8780, loss=1.03  (1977.7  examples/sec;  0.065  sec/batch)
step 8790, loss=0.80  (1954.9  examples/sec;  0.065  sec/batch)
step 8800, loss=0.82  (1875.5  examples/sec;  0.068  sec/batch)
step 8810, loss=0.86  (1985.5  examples/sec;  0.064  sec/batch)
step 8820, loss=0.70  (1908.6  examples/sec;  0.067  sec/batch)
step 8830, loss=0.78  (1952.0  examples/sec;  0.066  sec/batch)
step 8840, loss=0.84  (1807.1  examples/sec;  0.071  sec/batch)
step 8850, loss=0.83  (1831.4  examples/sec;  0.070  sec/batch)
step 8860, loss=0.94  (1823.7  examples/sec;  0.070  sec/batch)
step 8870, loss=0.99  (1822.8  examples/sec;  0.070  sec/batch)
step 8880, loss=0.93  (1909.0  examples/sec;  0.067  sec/batch)
step 8890, loss=0.85  (1878.8  examples/sec;  0.068  sec/batch)
step 8900, loss=0.67  (1928.2  examples/sec;  0.066  sec/batch)
step 8910, loss=0.96  (1821.0  examples/sec;  0.070  sec/batch)
step 8920, loss=0.83  (1796.2  examples/sec;  0.071  sec/batch)
step 8930, loss=0.79  (1892.7  examples/sec;  0.068  sec/batch)
step 8940, loss=0.91  (1787.7  examples/sec;  0.072  sec/batch)
step 8950, loss=0.85  (1898.9  examples/sec;  0.067  sec/batch)
step 8960, loss=0.85  (1897.9  examples/sec;  0.067  sec/batch)
step 8970, loss=1.00  (1815.3  examples/sec;  0.071  sec/batch)
step 8980, loss=0.66  (1871.6  examples/sec;  0.068  sec/batch)
step 8990, loss=0.87  (1920.8  examples/sec;  0.067  sec/batch)
step 9000, loss=1.04  (1875.1  examples/sec;  0.068  sec/batch)
step 9010, loss=0.91  (1846.9  examples/sec;  0.069  sec/batch)
step 9020, loss=0.81  (1922.1  examples/sec;  0.067  sec/batch)
step 9030, loss=0.73  (1832.3  examples/sec;  0.070  sec/batch)
step 9040, loss=0.79  (1894.8  examples/sec;  0.068  sec/batch)
step 9050, loss=0.82  (1977.8  examples/sec;  0.065  sec/batch)
step 9060, loss=0.91  (1883.1  examples/sec;  0.068  sec/batch)
step 9070, loss=0.99  (1875.9  examples/sec;  0.068  sec/batch)
step 9080, loss=0.88  (1887.6  examples/sec;  0.068  sec/batch)
step 9090, loss=0.90  (1764.3  examples/sec;  0.073  sec/batch)
step 9100, loss=0.87  (1866.9  examples/sec;  0.069  sec/batch)
step 9110, loss=0.93  (1907.1  examples/sec;  0.067  sec/batch)
step 9120, loss=0.93  (1895.8  examples/sec;  0.068  sec/batch)
step 9130, loss=0.93  (1860.4  examples/sec;  0.069  sec/batch)
step 9140, loss=1.05  (1888.1  examples/sec;  0.068  sec/batch)
step 9150, loss=0.99  (1952.7  examples/sec;  0.066  sec/batch)
step 9160, loss=0.93  (1968.2  examples/sec;  0.065  sec/batch)
step 9170, loss=0.94  (1838.9  examples/sec;  0.070  sec/batch)
step 9180, loss=0.86  (1862.3  examples/sec;  0.069  sec/batch)
step 9190, loss=0.90  (1845.1  examples/sec;  0.069  sec/batch)
step 9200, loss=0.83  (1866.7  examples/sec;  0.069  sec/batch)
step 9210, loss=0.81  (1855.6  examples/sec;  0.069  sec/batch)
step 9220, loss=0.72  (1805.7  examples/sec;  0.071  sec/batch)
step 9230, loss=1.09  (1877.5  examples/sec;  0.068  sec/batch)
step 9240, loss=0.85  (1860.5  examples/sec;  0.069  sec/batch)
step 9250, loss=0.78  (1897.6  examples/sec;  0.067  sec/batch)
step 9260, loss=0.73  (1910.7  examples/sec;  0.067  sec/batch)
step 9270, loss=0.74  (1875.9  examples/sec;  0.068  sec/batch)
step 9280, loss=0.81  (1901.1  examples/sec;  0.067  sec/batch)
step 9290, loss=0.96  (1799.5  examples/sec;  0.071  sec/batch)
step 9300, loss=0.95  (1890.6  examples/sec;  0.068  sec/batch)
step 9310, loss=0.96  (1822.7  examples/sec;  0.070  sec/batch)
step 9320, loss=0.95  (1657.9  examples/sec;  0.077  sec/batch)
step 9330, loss=0.80  (1860.8  examples/sec;  0.069  sec/batch)
step 9340, loss=0.87  (1920.3  examples/sec;  0.067  sec/batch)
step 9350, loss=0.88  (1738.4  examples/sec;  0.074  sec/batch)
step 9360, loss=0.85  (1896.2  examples/sec;  0.068  sec/batch)
step 9370, loss=0.84  (1919.7  examples/sec;  0.067  sec/batch)
step 9380, loss=0.80  (1905.2  examples/sec;  0.067  sec/batch)
step 9390, loss=0.87  (1856.0  examples/sec;  0.069  sec/batch)
step 9400, loss=0.97  (1902.9  examples/sec;  0.067  sec/batch)
step 9410, loss=0.80  (1940.1  examples/sec;  0.066  sec/batch)
step 9420, loss=0.84  (1870.7  examples/sec;  0.068  sec/batch)
step 9430, loss=0.74  (1974.2  examples/sec;  0.065  sec/batch)
step 9440, loss=0.68  (1947.7  examples/sec;  0.066  sec/batch)
step 9450, loss=1.00  (1924.6  examples/sec;  0.067  sec/batch)
step 9460, loss=0.72  (1876.0  examples/sec;  0.068  sec/batch)
step 9470, loss=0.83  (1878.8  examples/sec;  0.068  sec/batch)
step 9480, loss=0.77  (1871.9  examples/sec;  0.068  sec/batch)
step 9490, loss=0.77  (1944.7  examples/sec;  0.066  sec/batch)
step 9500, loss=0.74  (1597.4  examples/sec;  0.080  sec/batch)
step 9510, loss=0.73  (1920.0  examples/sec;  0.067  sec/batch)
step 9520, loss=0.79  (1930.8  examples/sec;  0.066  sec/batch)
step 9530, loss=0.67  (1789.3  examples/sec;  0.072  sec/batch)
step 9540, loss=0.75  (1858.4  examples/sec;  0.069  sec/batch)
step 9550, loss=0.78  (1929.5  examples/sec;  0.066  sec/batch)
step 9560, loss=0.98  (1848.9  examples/sec;  0.069  sec/batch)
step 9570, loss=0.77  (1904.0  examples/sec;  0.067  sec/batch)
step 9580, loss=0.79  (1750.7  examples/sec;  0.073  sec/batch)
step 9590, loss=0.81  (1717.4  examples/sec;  0.075  sec/batch)
step 9600, loss=0.80  (1857.0  examples/sec;  0.069  sec/batch)
step 9610, loss=0.90  (1898.5  examples/sec;  0.067  sec/batch)
step 9620, loss=0.84  (1812.0  examples/sec;  0.071  sec/batch)
step 9630, loss=0.82  (1907.6  examples/sec;  0.067  sec/batch)
step 9640, loss=1.01  (1872.3  examples/sec;  0.068  sec/batch)
step 9650, loss=0.92  (1913.2  examples/sec;  0.067  sec/batch)
step 9660, loss=0.82  (1893.4  examples/sec;  0.068  sec/batch)
step 9670, loss=0.73  (1882.7  examples/sec;  0.068  sec/batch)
step 9680, loss=0.81  (1916.8  examples/sec;  0.067  sec/batch)
step 9690, loss=0.81  (1904.9  examples/sec;  0.067  sec/batch)
step 9700, loss=0.76  (1911.6  examples/sec;  0.067  sec/batch)
step 9710, loss=0.79  (1940.4  examples/sec;  0.066  sec/batch)
step 9720, loss=0.91  (1942.0  examples/sec;  0.066  sec/batch)
step 9730, loss=0.99  (1874.6  examples/sec;  0.068  sec/batch)
step 9740, loss=0.80  (1923.9  examples/sec;  0.067  sec/batch)
step 9750, loss=0.84  (1866.3  examples/sec;  0.069  sec/batch)
step 9760, loss=0.92  (1909.8  examples/sec;  0.067  sec/batch)
step 9770, loss=0.86  (1756.1  examples/sec;  0.073  sec/batch)
step 9780, loss=0.84  (1836.9  examples/sec;  0.070  sec/batch)
step 9790, loss=0.84  (1793.3  examples/sec;  0.071  sec/batch)
step 9800, loss=0.91  (1921.7  examples/sec;  0.067  sec/batch)
step 9810, loss=0.85  (1927.6  examples/sec;  0.066  sec/batch)
step 9820, loss=0.89  (1926.3  examples/sec;  0.066  sec/batch)
step 9830, loss=0.81  (1837.3  examples/sec;  0.070  sec/batch)
step 9840, loss=0.81  (1825.8  examples/sec;  0.070  sec/batch)
step 9850, loss=0.88  (1953.5  examples/sec;  0.066  sec/batch)
step 9860, loss=0.85  (1837.6  examples/sec;  0.070  sec/batch)
step 9870, loss=0.94  (1853.1  examples/sec;  0.069  sec/batch)
step 9880, loss=0.75  (1853.0  examples/sec;  0.069  sec/batch)
step 9890, loss=0.89  (1821.3  examples/sec;  0.070  sec/batch)
step 9900, loss=0.89  (1950.0  examples/sec;  0.066  sec/batch)
step 9910, loss=0.76  (1945.1  examples/sec;  0.066  sec/batch)
step 9920, loss=0.82  (1999.2  examples/sec;  0.064  sec/batch)
step 9930, loss=0.68  (1760.9  examples/sec;  0.073  sec/batch)
step 9940, loss=0.75  (2001.9  examples/sec;  0.064  sec/batch)
step 9950, loss=0.93  (1779.0  examples/sec;  0.072  sec/batch)
step 9960, loss=0.68  (1736.0  examples/sec;  0.074  sec/batch)
step 9970, loss=1.04  (1945.9  examples/sec;  0.066  sec/batch)
step 9980, loss=0.75  (1909.6  examples/sec;  0.067  sec/batch)
step 9990, loss=0.88  (1967.3  examples/sec;  0.065  sec/batch)
step 10000, loss=0.83  (1914.6  examples/sec;  0.067  sec/batch)
step 10010, loss=0.92  (1791.0  examples/sec;  0.071  sec/batch)
step 10020, loss=0.78  (1927.2  examples/sec;  0.066  sec/batch)
step 10030, loss=0.93  (1922.0  examples/sec;  0.067  sec/batch)
step 10040, loss=0.95  (1915.1  examples/sec;  0.067  sec/batch)
step 10050, loss=0.85  (1822.3  examples/sec;  0.070  sec/batch)
step 10060, loss=0.84  (1859.1  examples/sec;  0.069  sec/batch)
step 10070, loss=0.88  (1909.0  examples/sec;  0.067  sec/batch)
step 10080, loss=0.86  (1856.3  examples/sec;  0.069  sec/batch)
step 10090, loss=0.64  (1857.1  examples/sec;  0.069  sec/batch)
step 10100, loss=0.71  (1844.6  examples/sec;  0.069  sec/batch)
step 10110, loss=0.78  (1744.2  examples/sec;  0.073  sec/batch)
step 10120, loss=0.89  (1923.8  examples/sec;  0.067  sec/batch)
step 10130, loss=0.80  (1780.7  examples/sec;  0.072  sec/batch)
step 10140, loss=0.67  (1649.1  examples/sec;  0.078  sec/batch)
step 10150, loss=0.91  (1816.2  examples/sec;  0.070  sec/batch)
step 10160, loss=0.83  (1906.5  examples/sec;  0.067  sec/batch)
step 10170, loss=0.84  (1888.7  examples/sec;  0.068  sec/batch)
step 10180, loss=0.82  (1665.8  examples/sec;  0.077  sec/batch)
step 10190, loss=0.85  (1900.4  examples/sec;  0.067  sec/batch)
step 10200, loss=0.81  (1812.7  examples/sec;  0.071  sec/batch)
step 10210, loss=0.81  (1916.6  examples/sec;  0.067  sec/batch)
step 10220, loss=0.84  (1883.8  examples/sec;  0.068  sec/batch)
step 10230, loss=0.81  (1912.3  examples/sec;  0.067  sec/batch)
step 10240, loss=1.02  (1897.5  examples/sec;  0.067  sec/batch)
step 10250, loss=0.88  (1916.4  examples/sec;  0.067  sec/batch)
step 10260, loss=0.85  (1805.2  examples/sec;  0.071  sec/batch)
step 10270, loss=0.82  (1865.0  examples/sec;  0.069  sec/batch)
step 10280, loss=0.74  (1930.3  examples/sec;  0.066  sec/batch)
step 10290, loss=0.83  (1903.3  examples/sec;  0.067  sec/batch)
step 10300, loss=0.84  (1875.6  examples/sec;  0.068  sec/batch)
step 10310, loss=1.02  (1856.5  examples/sec;  0.069  sec/batch)
step 10320, loss=0.92  (1872.1  examples/sec;  0.068  sec/batch)
step 10330, loss=0.91  (1840.6  examples/sec;  0.070  sec/batch)
step 10340, loss=0.85  (1910.8  examples/sec;  0.067  sec/batch)
step 10350, loss=0.75  (1821.1  examples/sec;  0.070  sec/batch)
step 10360, loss=0.74  (1975.6  examples/sec;  0.065  sec/batch)
step 10370, loss=0.82  (1839.7  examples/sec;  0.070  sec/batch)
step 10380, loss=0.91  (1940.5  examples/sec;  0.066  sec/batch)
step 10390, loss=0.89  (1921.2  examples/sec;  0.067  sec/batch)
step 10400, loss=0.67  (1851.0  examples/sec;  0.069  sec/batch)
step 10410, loss=0.76  (1888.1  examples/sec;  0.068  sec/batch)
step 10420, loss=0.79  (1874.5  examples/sec;  0.068  sec/batch)
step 10430, loss=0.74  (1836.8  examples/sec;  0.070  sec/batch)
step 10440, loss=1.01  (1867.7  examples/sec;  0.069  sec/batch)
step 10450, loss=0.88  (1813.7  examples/sec;  0.071  sec/batch)
step 10460, loss=0.96  (1977.4  examples/sec;  0.065  sec/batch)
step 10470, loss=0.87  (1845.9  examples/sec;  0.069  sec/batch)
step 10480, loss=0.82  (1785.1  examples/sec;  0.072  sec/batch)
step 10490, loss=0.75  (1849.1  examples/sec;  0.069  sec/batch)
step 10500, loss=0.69  (1917.0  examples/sec;  0.067  sec/batch)
step 10510, loss=0.99  (1877.3  examples/sec;  0.068  sec/batch)
step 10520, loss=0.82  (1868.6  examples/sec;  0.069  sec/batch)
step 10530, loss=0.87  (1856.7  examples/sec;  0.069  sec/batch)
step 10540, loss=0.87  (1938.9  examples/sec;  0.066  sec/batch)
step 10550, loss=0.88  (1890.1  examples/sec;  0.068  sec/batch)
step 10560, loss=0.70  (1877.3  examples/sec;  0.068  sec/batch)
step 10570, loss=0.78  (1881.5  examples/sec;  0.068  sec/batch)
step 10580, loss=0.73  (1805.6  examples/sec;  0.071  sec/batch)
step 10590, loss=0.82  (1889.3  examples/sec;  0.068  sec/batch)
step 10600, loss=0.68  (1951.5  examples/sec;  0.066  sec/batch)
step 10610, loss=0.89  (1909.4  examples/sec;  0.067  sec/batch)
step 10620, loss=0.80  (1868.8  examples/sec;  0.068  sec/batch)
step 10630, loss=0.78  (1913.8  examples/sec;  0.067  sec/batch)
step 10640, loss=0.71  (2012.4  examples/sec;  0.064  sec/batch)
step 10650, loss=0.82  (1908.4  examples/sec;  0.067  sec/batch)
step 10660, loss=0.71  (1866.8  examples/sec;  0.069  sec/batch)
step 10670, loss=0.84  (2053.2  examples/sec;  0.062  sec/batch)
step 10680, loss=0.73  (1964.9  examples/sec;  0.065  sec/batch)
step 10690, loss=0.76  (1794.6  examples/sec;  0.071  sec/batch)
step 10700, loss=0.86  (1931.7  examples/sec;  0.066  sec/batch)
step 10710, loss=0.83  (1939.1  examples/sec;  0.066  sec/batch)
step 10720, loss=0.79  (1831.6  examples/sec;  0.070  sec/batch)
step 10730, loss=0.68  (1854.9  examples/sec;  0.069  sec/batch)
step 10740, loss=0.73  (1657.0  examples/sec;  0.077  sec/batch)
step 10750, loss=0.73  (1877.5  examples/sec;  0.068  sec/batch)
step 10760, loss=0.76  (1904.8  examples/sec;  0.067  sec/batch)
step 10770, loss=0.87  (1795.6  examples/sec;  0.071  sec/batch)
step 10780, loss=0.80  (1847.5  examples/sec;  0.069  sec/batch)
step 10790, loss=0.91  (1934.1  examples/sec;  0.066  sec/batch)
step 10800, loss=0.78  (1996.3  examples/sec;  0.064  sec/batch)
step 10810, loss=0.61  (1845.1  examples/sec;  0.069  sec/batch)
step 10820, loss=0.83  (1900.5  examples/sec;  0.067  sec/batch)
step 10830, loss=0.75  (1847.0  examples/sec;  0.069  sec/batch)
step 10840, loss=0.74  (1829.9  examples/sec;  0.070  sec/batch)
step 10850, loss=0.91  (1906.7  examples/sec;  0.067  sec/batch)
step 10860, loss=0.88  (1884.0  examples/sec;  0.068  sec/batch)
step 10870, loss=0.67  (1918.2  examples/sec;  0.067  sec/batch)
step 10880, loss=0.93  (1802.1  examples/sec;  0.071  sec/batch)
step 10890, loss=0.87  (1931.3  examples/sec;  0.066  sec/batch)
step 10900, loss=0.84  (1811.5  examples/sec;  0.071  sec/batch)
step 10910, loss=0.75  (1904.6  examples/sec;  0.067  sec/batch)
step 10920, loss=0.78  (1797.9  examples/sec;  0.071  sec/batch)
step 10930, loss=0.94  (1805.5  examples/sec;  0.071  sec/batch)
step 10940, loss=0.83  (1942.0  examples/sec;  0.066  sec/batch)
step 10950, loss=0.84  (1863.2  examples/sec;  0.069  sec/batch)
step 10960, loss=0.91  (1836.3  examples/sec;  0.070  sec/batch)
step 10970, loss=0.89  (1909.6  examples/sec;  0.067  sec/batch)
step 10980, loss=0.73  (1891.2  examples/sec;  0.068  sec/batch)
step 10990, loss=0.80  (1939.3  examples/sec;  0.066  sec/batch)
step 11000, loss=0.97  (1925.3  examples/sec;  0.066  sec/batch)
step 11010, loss=0.69  (1921.3  examples/sec;  0.067  sec/batch)
step 11020, loss=0.83  (1954.1  examples/sec;  0.066  sec/batch)
step 11030, loss=0.81  (1900.6  examples/sec;  0.067  sec/batch)
step 11040, loss=0.78  (1892.1  examples/sec;  0.068  sec/batch)
step 11050, loss=0.91  (1947.4  examples/sec;  0.066  sec/batch)
step 11060, loss=0.69  (1787.8  examples/sec;  0.072  sec/batch)
step 11070, loss=0.89  (1908.0  examples/sec;  0.067  sec/batch)
step 11080, loss=0.68  (1958.1  examples/sec;  0.065  sec/batch)
step 11090, loss=0.97  (1851.7  examples/sec;  0.069  sec/batch)
step 11100, loss=0.85  (1911.1  examples/sec;  0.067  sec/batch)
step 11110, loss=0.75  (1908.3  examples/sec;  0.067  sec/batch)
step 11120, loss=0.90  (1848.6  examples/sec;  0.069  sec/batch)
step 11130, loss=0.85  (1932.4  examples/sec;  0.066  sec/batch)
step 11140, loss=0.74  (1814.1  examples/sec;  0.071  sec/batch)
step 11150, loss=0.77  (1849.9  examples/sec;  0.069  sec/batch)
step 11160, loss=0.97  (1762.2  examples/sec;  0.073  sec/batch)
step 11170, loss=0.73  (1930.8  examples/sec;  0.066  sec/batch)
step 11180, loss=0.77  (1238.5  examples/sec;  0.103  sec/batch)
step 11190, loss=0.79  (1397.2  examples/sec;  0.092  sec/batch)
step 11200, loss=0.82  (1506.2  examples/sec;  0.085  sec/batch)
step 11210, loss=0.84  (1612.1  examples/sec;  0.079  sec/batch)
step 11220, loss=0.81  (1291.7  examples/sec;  0.099  sec/batch)
step 11230, loss=1.01  (1972.2  examples/sec;  0.065  sec/batch)
step 11240, loss=0.95  (1885.4  examples/sec;  0.068  sec/batch)
step 11250, loss=0.82  (1973.1  examples/sec;  0.065  sec/batch)
step 11260, loss=0.87  (1887.9  examples/sec;  0.068  sec/batch)
step 11270, loss=0.85  (1876.3  examples/sec;  0.068  sec/batch)
step 11280, loss=0.76  (1878.6  examples/sec;  0.068  sec/batch)
step 11290, loss=0.76  (1993.1  examples/sec;  0.064  sec/batch)
step 11300, loss=0.81  (1863.1  examples/sec;  0.069  sec/batch)
step 11310, loss=0.90  (1873.0  examples/sec;  0.068  sec/batch)
step 11320, loss=0.72  (1944.2  examples/sec;  0.066  sec/batch)
step 11330, loss=0.84  (1849.1  examples/sec;  0.069  sec/batch)
step 11340, loss=1.00  (1842.4  examples/sec;  0.069  sec/batch)
step 11350, loss=0.89  (2060.6  examples/sec;  0.062  sec/batch)
step 11360, loss=0.75  (1800.5  examples/sec;  0.071  sec/batch)
step 11370, loss=0.84  (1885.6  examples/sec;  0.068  sec/batch)
step 11380, loss=0.81  (1903.0  examples/sec;  0.067  sec/batch)
step 11390, loss=0.70  (1878.1  examples/sec;  0.068  sec/batch)
step 11400, loss=0.79  (1892.1  examples/sec;  0.068  sec/batch)
step 11410, loss=0.99  (1906.3  examples/sec;  0.067  sec/batch)
step 11420, loss=0.82  (1948.8  examples/sec;  0.066  sec/batch)
step 11430, loss=0.88  (1882.4  examples/sec;  0.068  sec/batch)
step 11440, loss=0.85  (1921.6  examples/sec;  0.067  sec/batch)
step 11450, loss=0.63  (1875.4  examples/sec;  0.068  sec/batch)
step 11460, loss=0.86  (1900.8  examples/sec;  0.067  sec/batch)
step 11470, loss=0.76  (1841.5  examples/sec;  0.070  sec/batch)
step 11480, loss=0.88  (1850.0  examples/sec;  0.069  sec/batch)
step 11490, loss=0.73  (1837.2  examples/sec;  0.070  sec/batch)
step 11500, loss=0.93  (1822.0  examples/sec;  0.070  sec/batch)
step 11510, loss=0.79  (1861.3  examples/sec;  0.069  sec/batch)
step 11520, loss=0.70  (1784.2  examples/sec;  0.072  sec/batch)
step 11530, loss=0.81  (1818.9  examples/sec;  0.070  sec/batch)
step 11540, loss=0.80  (1889.4  examples/sec;  0.068  sec/batch)
step 11550, loss=0.92  (1892.8  examples/sec;  0.068  sec/batch)
step 11560, loss=0.98  (1877.0  examples/sec;  0.068  sec/batch)
step 11570, loss=0.76  (1410.3  examples/sec;  0.091  sec/batch)
step 11580, loss=0.74  (1961.8  examples/sec;  0.065  sec/batch)
step 11590, loss=0.79  (1944.3  examples/sec;  0.066  sec/batch)
step 11600, loss=0.76  (1736.3  examples/sec;  0.074  sec/batch)
step 11610, loss=0.74  (1922.4  examples/sec;  0.067  sec/batch)
step 11620, loss=0.71  (1890.6  examples/sec;  0.068  sec/batch)
step 11630, loss=0.78  (1846.5  examples/sec;  0.069  sec/batch)
step 11640, loss=0.73  (1838.1  examples/sec;  0.070  sec/batch)
step 11650, loss=0.77  (1866.0  examples/sec;  0.069  sec/batch)
step 11660, loss=0.98  (1966.1  examples/sec;  0.065  sec/batch)
step 11670, loss=0.74  (1840.4  examples/sec;  0.070  sec/batch)
step 11680, loss=0.96  (1899.8  examples/sec;  0.067  sec/batch)
step 11690, loss=0.93  (1981.0  examples/sec;  0.065  sec/batch)
step 11700, loss=0.91  (1883.7  examples/sec;  0.068  sec/batch)
step 11710, loss=0.73  (1842.3  examples/sec;  0.069  sec/batch)
step 11720, loss=1.03  (1851.7  examples/sec;  0.069  sec/batch)
step 11730, loss=0.87  (1774.6  examples/sec;  0.072  sec/batch)
step 11740, loss=0.87  (1911.9  examples/sec;  0.067  sec/batch)
step 11750, loss=0.86  (1866.7  examples/sec;  0.069  sec/batch)
step 11760, loss=0.86  (1856.0  examples/sec;  0.069  sec/batch)
step 11770, loss=0.86  (1819.6  examples/sec;  0.070  sec/batch)
step 11780, loss=0.82  (1859.0  examples/sec;  0.069  sec/batch)
step 11790, loss=0.74  (1998.7  examples/sec;  0.064  sec/batch)
step 11800, loss=0.88  (1963.3  examples/sec;  0.065  sec/batch)
step 11810, loss=0.89  (1837.6  examples/sec;  0.070  sec/batch)
step 11820, loss=0.77  (1994.2  examples/sec;  0.064  sec/batch)
step 11830, loss=0.92  (1866.1  examples/sec;  0.069  sec/batch)
step 11840, loss=0.91  (1892.3  examples/sec;  0.068  sec/batch)
step 11850, loss=0.81  (1860.3  examples/sec;  0.069  sec/batch)
step 11860, loss=1.00  (1790.3  examples/sec;  0.071  sec/batch)
step 11870, loss=0.79  (1843.1  examples/sec;  0.069  sec/batch)
step 11880, loss=0.69  (1888.9  examples/sec;  0.068  sec/batch)
step 11890, loss=0.95  (1924.4  examples/sec;  0.067  sec/batch)
step 11900, loss=0.73  (1830.6  examples/sec;  0.070  sec/batch)
step 11910, loss=0.89  (1848.9  examples/sec;  0.069  sec/batch)
step 11920, loss=0.93  (1858.3  examples/sec;  0.069  sec/batch)
step 11930, loss=0.90  (1917.0  examples/sec;  0.067  sec/batch)
step 11940, loss=0.81  (1893.9  examples/sec;  0.068  sec/batch)
step 11950, loss=0.81  (1910.2  examples/sec;  0.067  sec/batch)
step 11960, loss=0.80  (1895.4  examples/sec;  0.068  sec/batch)
step 11970, loss=0.84  (1881.8  examples/sec;  0.068  sec/batch)
step 11980, loss=0.80  (1862.6  examples/sec;  0.069  sec/batch)
step 11990, loss=0.80  (1905.4  examples/sec;  0.067  sec/batch)
step 12000, loss=0.83  (1842.8  examples/sec;  0.069  sec/batch)
step 12010, loss=0.70  (1851.5  examples/sec;  0.069  sec/batch)
step 12020, loss=0.74  (1874.2  examples/sec;  0.068  sec/batch)
step 12030, loss=0.73  (1848.4  examples/sec;  0.069  sec/batch)
step 12040, loss=0.83  (1801.0  examples/sec;  0.071  sec/batch)
step 12050, loss=0.61  (1616.1  examples/sec;  0.079  sec/batch)
step 12060, loss=0.69  (1946.1  examples/sec;  0.066  sec/batch)
step 12070, loss=0.96  (1914.5  examples/sec;  0.067  sec/batch)
step 12080, loss=0.75  (1943.6  examples/sec;  0.066  sec/batch)
step 12090, loss=0.78  (1925.5  examples/sec;  0.066  sec/batch)
step 12100, loss=0.95  (1899.5  examples/sec;  0.067  sec/batch)
step 12110, loss=0.91  (1833.8  examples/sec;  0.070  sec/batch)
step 12120, loss=0.69  (1849.6  examples/sec;  0.069  sec/batch)
step 12130, loss=0.85  (1881.1  examples/sec;  0.068  sec/batch)
step 12140, loss=0.62  (1882.7  examples/sec;  0.068  sec/batch)
step 12150, loss=0.75  (2004.1  examples/sec;  0.064  sec/batch)
step 12160, loss=0.73  (1877.5  examples/sec;  0.068  sec/batch)
step 12170, loss=0.87  (1894.0  examples/sec;  0.068  sec/batch)
step 12180, loss=0.79  (1901.6  examples/sec;  0.067  sec/batch)
step 12190, loss=0.87  (1975.4  examples/sec;  0.065  sec/batch)
step 12200, loss=0.82  (1919.7  examples/sec;  0.067  sec/batch)
step 12210, loss=0.87  (1869.5  examples/sec;  0.068  sec/batch)
step 12220, loss=0.77  (1484.1  examples/sec;  0.086  sec/batch)
step 12230, loss=0.87  (1824.1  examples/sec;  0.070  sec/batch)
step 12240, loss=0.80  (1969.8  examples/sec;  0.065  sec/batch)
step 12250, loss=0.66  (1872.3  examples/sec;  0.068  sec/batch)
step 12260, loss=0.97  (1897.4  examples/sec;  0.067  sec/batch)
step 12270, loss=0.82  (1905.0  examples/sec;  0.067  sec/batch)
step 12280, loss=0.75  (1917.7  examples/sec;  0.067  sec/batch)
step 12290, loss=0.82  (1841.2  examples/sec;  0.070  sec/batch)
step 12300, loss=0.75  (1901.8  examples/sec;  0.067  sec/batch)
step 12310, loss=0.77  (1952.6  examples/sec;  0.066  sec/batch)
step 12320, loss=0.76  (1856.5  examples/sec;  0.069  sec/batch)
step 12330, loss=0.74  (1905.7  examples/sec;  0.067  sec/batch)
step 12340, loss=0.81  (1800.6  examples/sec;  0.071  sec/batch)
step 12350, loss=0.73  (1816.8  examples/sec;  0.070  sec/batch)
step 12360, loss=0.79  (1939.6  examples/sec;  0.066  sec/batch)
step 12370, loss=0.78  (1847.3  examples/sec;  0.069  sec/batch)
step 12380, loss=0.88  (1887.5  examples/sec;  0.068  sec/batch)
step 12390, loss=1.03  (1924.5  examples/sec;  0.067  sec/batch)
step 12400, loss=0.85  (1830.1  examples/sec;  0.070  sec/batch)
step 12410, loss=0.87  (1860.2  examples/sec;  0.069  sec/batch)
step 12420, loss=0.76  (1919.7  examples/sec;  0.067  sec/batch)
step 12430, loss=0.67  (1771.2  examples/sec;  0.072  sec/batch)
step 12440, loss=0.77  (1871.4  examples/sec;  0.068  sec/batch)
step 12450, loss=0.84  (1927.1  examples/sec;  0.066  sec/batch)
step 12460, loss=0.68  (1730.5  examples/sec;  0.074  sec/batch)
step 12470, loss=0.65  (1825.8  examples/sec;  0.070  sec/batch)
step 12480, loss=1.00  (1855.2  examples/sec;  0.069  sec/batch)
step 12490, loss=0.84  (1798.2  examples/sec;  0.071  sec/batch)
step 12500, loss=0.80  (1805.9  examples/sec;  0.071  sec/batch)
step 12510, loss=0.83  (1845.0  examples/sec;  0.069  sec/batch)
step 12520, loss=0.78  (1856.4  examples/sec;  0.069  sec/batch)
step 12530, loss=0.77  (1884.6  examples/sec;  0.068  sec/batch)
step 12540, loss=0.88  (1573.9  examples/sec;  0.081  sec/batch)
step 12550, loss=0.94  (1815.6  examples/sec;  0.070  sec/batch)
step 12560, loss=0.81  (1894.0  examples/sec;  0.068  sec/batch)
step 12570, loss=0.71  (1894.8  examples/sec;  0.068  sec/batch)
step 12580, loss=0.82  (1953.4  examples/sec;  0.066  sec/batch)
step 12590, loss=0.78  (1834.2  examples/sec;  0.070  sec/batch)
step 12600, loss=0.74  (1877.3  examples/sec;  0.068  sec/batch)
step 12610, loss=0.95  (1937.5  examples/sec;  0.066  sec/batch)
step 12620, loss=0.95  (1915.1  examples/sec;  0.067  sec/batch)
step 12630, loss=0.98  (1835.0  examples/sec;  0.070  sec/batch)
step 12640, loss=0.95  (1939.1  examples/sec;  0.066  sec/batch)
step 12650, loss=0.89  (1904.0  examples/sec;  0.067  sec/batch)
step 12660, loss=1.02  (1878.0  examples/sec;  0.068  sec/batch)
step 12670, loss=0.86  (1907.3  examples/sec;  0.067  sec/batch)
step 12680, loss=0.71  (1882.1  examples/sec;  0.068  sec/batch)
step 12690, loss=0.85  (1830.8  examples/sec;  0.070  sec/batch)
step 12700, loss=0.76  (1923.4  examples/sec;  0.067  sec/batch)
step 12710, loss=0.94  (1922.9  examples/sec;  0.067  sec/batch)
step 12720, loss=0.69  (1898.4  examples/sec;  0.067  sec/batch)
step 12730, loss=0.73  (1859.0  examples/sec;  0.069  sec/batch)
step 12740, loss=0.97  (1950.0  examples/sec;  0.066  sec/batch)
step 12750, loss=0.76  (1952.3  examples/sec;  0.066  sec/batch)
step 12760, loss=0.94  (1806.9  examples/sec;  0.071  sec/batch)
step 12770, loss=0.65  (1989.6  examples/sec;  0.064  sec/batch)
step 12780, loss=0.75  (1936.6  examples/sec;  0.066  sec/batch)
step 12790, loss=0.75  (1739.9  examples/sec;  0.074  sec/batch)
step 12800, loss=0.83  (1809.3  examples/sec;  0.071  sec/batch)
step 12810, loss=0.95  (1870.4  examples/sec;  0.068  sec/batch)
step 12820, loss=0.88  (1969.6  examples/sec;  0.065  sec/batch)
step 12830, loss=0.74  (1897.9  examples/sec;  0.067  sec/batch)
step 12840, loss=0.71  (1865.9  examples/sec;  0.069  sec/batch)
step 12850, loss=0.95  (1717.4  examples/sec;  0.075  sec/batch)
step 12860, loss=0.82  (1907.7  examples/sec;  0.067  sec/batch)
step 12870, loss=0.88  (1808.4  examples/sec;  0.071  sec/batch)
step 12880, loss=0.85  (1858.1  examples/sec;  0.069  sec/batch)
step 12890, loss=0.62  (1962.7  examples/sec;  0.065  sec/batch)
step 12900, loss=1.01  (1982.7  examples/sec;  0.065  sec/batch)
step 12910, loss=0.87  (1856.0  examples/sec;  0.069  sec/batch)
step 12920, loss=0.87  (1862.7  examples/sec;  0.069  sec/batch)
step 12930, loss=0.85  (1912.0  examples/sec;  0.067  sec/batch)
step 12940, loss=0.63  (1866.8  examples/sec;  0.069  sec/batch)
step 12950, loss=0.96  (1857.5  examples/sec;  0.069  sec/batch)
step 12960, loss=0.71  (1839.0  examples/sec;  0.070  sec/batch)
step 12970, loss=0.59  (1869.0  examples/sec;  0.068  sec/batch)
step 12980, loss=0.78  (1874.3  examples/sec;  0.068  sec/batch)
step 12990, loss=0.89  (1859.7  examples/sec;  0.069  sec/batch)
step 13000, loss=0.93  (1861.2  examples/sec;  0.069  sec/batch)
step 13010, loss=0.82  (1874.6  examples/sec;  0.068  sec/batch)
step 13020, loss=0.63  (1839.2  examples/sec;  0.070  sec/batch)
step 13030, loss=0.86  (1877.5  examples/sec;  0.068  sec/batch)
step 13040, loss=0.76  (1871.2  examples/sec;  0.068  sec/batch)
step 13050, loss=0.72  (1877.8  examples/sec;  0.068  sec/batch)
step 13060, loss=0.85  (1907.2  examples/sec;  0.067  sec/batch)
step 13070, loss=0.91  (1896.1  examples/sec;  0.068  sec/batch)
step 13080, loss=1.06  (1987.3  examples/sec;  0.064  sec/batch)
step 13090, loss=0.86  (1868.0  examples/sec;  0.069  sec/batch)
step 13100, loss=0.82  (1920.2  examples/sec;  0.067  sec/batch)
step 13110, loss=0.72  (1843.9  examples/sec;  0.069  sec/batch)
step 13120, loss=0.88  (1944.4  examples/sec;  0.066  sec/batch)
step 13130, loss=0.78  (1848.9  examples/sec;  0.069  sec/batch)
step 13140, loss=0.69  (1748.9  examples/sec;  0.073  sec/batch)
step 13150, loss=0.72  (1870.7  examples/sec;  0.068  sec/batch)
step 13160, loss=0.71  (1939.2  examples/sec;  0.066  sec/batch)
step 13170, loss=0.86  (1827.0  examples/sec;  0.070  sec/batch)
step 13180, loss=0.97  (1855.0  examples/sec;  0.069  sec/batch)
step 13190, loss=0.54  (1868.2  examples/sec;  0.069  sec/batch)
step 13200, loss=0.87  (1885.5  examples/sec;  0.068  sec/batch)
step 13210, loss=0.82  (1778.4  examples/sec;  0.072  sec/batch)
step 13220, loss=0.89  (1881.5  examples/sec;  0.068  sec/batch)
step 13230, loss=0.87  (1867.5  examples/sec;  0.069  sec/batch)
step 13240, loss=0.73  (1921.5  examples/sec;  0.067  sec/batch)
step 13250, loss=0.79  (1847.9  examples/sec;  0.069  sec/batch)
step 13260, loss=0.95  (1907.1  examples/sec;  0.067  sec/batch)
step 13270, loss=0.68  (1893.0  examples/sec;  0.068  sec/batch)
step 13280, loss=0.60  (1922.9  examples/sec;  0.067  sec/batch)
step 13290, loss=0.84  (1782.4  examples/sec;  0.072  sec/batch)
step 13300, loss=0.60  (1924.5  examples/sec;  0.067  sec/batch)
step 13310, loss=0.73  (1914.3  examples/sec;  0.067  sec/batch)
step 13320, loss=0.72  (1861.8  examples/sec;  0.069  sec/batch)
step 13330, loss=0.77  (1871.4  examples/sec;  0.068  sec/batch)
step 13340, loss=0.71  (1876.1  examples/sec;  0.068  sec/batch)
step 13350, loss=0.88  (1886.1  examples/sec;  0.068  sec/batch)
step 13360, loss=0.86  (1832.9  examples/sec;  0.070  sec/batch)
step 13370, loss=0.87  (1816.9  examples/sec;  0.070  sec/batch)
step 13380, loss=0.74  (1767.3  examples/sec;  0.072  sec/batch)
step 13390, loss=0.62  (1962.6  examples/sec;  0.065  sec/batch)
step 13400, loss=0.96  (1922.3  examples/sec;  0.067  sec/batch)
step 13410, loss=0.82  (1959.4  examples/sec;  0.065  sec/batch)
step 13420, loss=0.93  (1927.6  examples/sec;  0.066  sec/batch)
step 13430, loss=0.77  (1898.5  examples/sec;  0.067  sec/batch)
step 13440, loss=0.50  (1872.8  examples/sec;  0.068  sec/batch)
step 13450, loss=0.75  (1818.9  examples/sec;  0.070  sec/batch)
step 13460, loss=0.76  (1880.9  examples/sec;  0.068  sec/batch)
step 13470, loss=0.75  (1827.9  examples/sec;  0.070  sec/batch)
step 13480, loss=0.85  (1575.6  examples/sec;  0.081  sec/batch)
step 13490, loss=0.84  (1889.5  examples/sec;  0.068  sec/batch)
step 13500, loss=0.80  (1864.2  examples/sec;  0.069  sec/batch)
step 13510, loss=0.69  (1892.3  examples/sec;  0.068  sec/batch)
step 13520, loss=0.66  (1875.3  examples/sec;  0.068  sec/batch)
step 13530, loss=0.90  (1937.9  examples/sec;  0.066  sec/batch)
step 13540, loss=0.79  (1837.5  examples/sec;  0.070  sec/batch)
step 13550, loss=0.65  (1911.6  examples/sec;  0.067  sec/batch)
step 13560, loss=0.93  (1830.1  examples/sec;  0.070  sec/batch)
step 13570, loss=0.76  (1986.5  examples/sec;  0.064  sec/batch)
step 13580, loss=0.82  (1853.6  examples/sec;  0.069  sec/batch)
step 13590, loss=0.78  (1872.6  examples/sec;  0.068  sec/batch)
step 13600, loss=0.84  (1845.3  examples/sec;  0.069  sec/batch)
step 13610, loss=0.95  (1870.3  examples/sec;  0.068  sec/batch)
step 13620, loss=0.75  (1892.0  examples/sec;  0.068  sec/batch)
step 13630, loss=0.73  (1870.8  examples/sec;  0.068  sec/batch)
step 13640, loss=0.82  (1878.2  examples/sec;  0.068  sec/batch)
step 13650, loss=0.71  (1884.4  examples/sec;  0.068  sec/batch)
step 13660, loss=0.77  (1853.1  examples/sec;  0.069  sec/batch)
step 13670, loss=0.79  (1861.0  examples/sec;  0.069  sec/batch)
step 13680, loss=0.67  (1883.1  examples/sec;  0.068  sec/batch)
step 13690, loss=0.93  (1907.3  examples/sec;  0.067  sec/batch)
step 13700, loss=0.76  (1809.9  examples/sec;  0.071  sec/batch)
step 13710, loss=0.81  (1852.8  examples/sec;  0.069  sec/batch)
step 13720, loss=0.77  (1927.6  examples/sec;  0.066  sec/batch)
step 13730, loss=0.80  (1873.4  examples/sec;  0.068  sec/batch)
step 13740, loss=0.82  (1922.5  examples/sec;  0.067  sec/batch)
step 13750, loss=0.63  (1861.4  examples/sec;  0.069  sec/batch)
step 13760, loss=0.66  (1915.5  examples/sec;  0.067  sec/batch)
step 13770, loss=0.70  (1949.2  examples/sec;  0.066  sec/batch)
step 13780, loss=0.84  (1910.3  examples/sec;  0.067  sec/batch)
step 13790, loss=0.70  (1922.6  examples/sec;  0.067  sec/batch)
step 13800, loss=0.79  (1925.3  examples/sec;  0.066  sec/batch)
step 13810, loss=0.70  (1943.2  examples/sec;  0.066  sec/batch)
step 13820, loss=0.71  (1888.2  examples/sec;  0.068  sec/batch)
step 13830, loss=0.75  (1868.9  examples/sec;  0.068  sec/batch)
step 13840, loss=0.60  (1826.5  examples/sec;  0.070  sec/batch)
step 13850, loss=0.95  (1904.2  examples/sec;  0.067  sec/batch)
step 13860, loss=0.80  (2015.3  examples/sec;  0.064  sec/batch)
step 13870, loss=0.71  (1800.3  examples/sec;  0.071  sec/batch)
step 13880, loss=0.67  (1915.2  examples/sec;  0.067  sec/batch)
step 13890, loss=0.94  (1894.8  examples/sec;  0.068  sec/batch)
step 13900, loss=0.87  (1916.5  examples/sec;  0.067  sec/batch)
step 13910, loss=0.79  (1953.7  examples/sec;  0.066  sec/batch)
step 13920, loss=0.83  (1847.6  examples/sec;  0.069  sec/batch)
step 13930, loss=0.77  (1872.1  examples/sec;  0.068  sec/batch)
step 13940, loss=0.78  (1818.7  examples/sec;  0.070  sec/batch)
step 13950, loss=0.76  (1864.3  examples/sec;  0.069  sec/batch)
step 13960, loss=0.88  (1886.5  examples/sec;  0.068  sec/batch)
step 13970, loss=0.91  (1831.6  examples/sec;  0.070  sec/batch)
step 13980, loss=0.66  (1923.5  examples/sec;  0.067  sec/batch)
step 13990, loss=0.82  (1895.0  examples/sec;  0.068  sec/batch)
step 14000, loss=0.87  (1937.9  examples/sec;  0.066  sec/batch)
step 14010, loss=0.76  (1849.8  examples/sec;  0.069  sec/batch)
step 14020, loss=0.74  (1852.5  examples/sec;  0.069  sec/batch)
step 14030, loss=0.88  (1901.9  examples/sec;  0.067  sec/batch)
step 14040, loss=0.76  (1774.2  examples/sec;  0.072  sec/batch)
step 14050, loss=0.69  (1889.5  examples/sec;  0.068  sec/batch)
step 14060, loss=0.75  (1861.4  examples/sec;  0.069  sec/batch)
step 14070, loss=0.86  (1882.5  examples/sec;  0.068  sec/batch)
step 14080, loss=0.79  (1915.1  examples/sec;  0.067  sec/batch)
step 14090, loss=0.86  (2038.8  examples/sec;  0.063  sec/batch)
step 14100, loss=0.88  (1816.9  examples/sec;  0.070  sec/batch)
step 14110, loss=0.84  (1844.7  examples/sec;  0.069  sec/batch)
step 14120, loss=0.86  (1686.5  examples/sec;  0.076  sec/batch)
step 14130, loss=0.64  (1935.5  examples/sec;  0.066  sec/batch)
step 14140, loss=0.93  (1900.9  examples/sec;  0.067  sec/batch)
step 14150, loss=0.82  (1850.5  examples/sec;  0.069  sec/batch)
step 14160, loss=0.77  (1876.0  examples/sec;  0.068  sec/batch)
step 14170, loss=0.68  (1809.9  examples/sec;  0.071  sec/batch)
step 14180, loss=0.64  (1863.0  examples/sec;  0.069  sec/batch)
step 14190, loss=0.60  (1932.6  examples/sec;  0.066  sec/batch)
step 14200, loss=0.66  (1922.4  examples/sec;  0.067  sec/batch)
step 14210, loss=0.73  (1902.5  examples/sec;  0.067  sec/batch)
step 14220, loss=0.65  (1892.3  examples/sec;  0.068  sec/batch)
step 14230, loss=0.81  (1845.2  examples/sec;  0.069  sec/batch)
step 14240, loss=1.02  (1954.0  examples/sec;  0.066  sec/batch)
step 14250, loss=0.92  (1782.7  examples/sec;  0.072  sec/batch)
step 14260, loss=0.89  (1766.2  examples/sec;  0.072  sec/batch)
step 14270, loss=0.78  (1908.1  examples/sec;  0.067  sec/batch)
step 14280, loss=0.74  (1874.4  examples/sec;  0.068  sec/batch)
step 14290, loss=0.76  (1959.1  examples/sec;  0.065  sec/batch)
step 14300, loss=0.87  (1917.0  examples/sec;  0.067  sec/batch)
step 14310, loss=0.79  (1946.3  examples/sec;  0.066  sec/batch)
step 14320, loss=0.70  (1967.0  examples/sec;  0.065  sec/batch)
step 14330, loss=0.85  (1914.2  examples/sec;  0.067  sec/batch)
step 14340, loss=0.67  (1862.6  examples/sec;  0.069  sec/batch)
step 14350, loss=0.83  (1868.3  examples/sec;  0.069  sec/batch)
step 14360, loss=1.03  (1876.9  examples/sec;  0.068  sec/batch)
step 14370, loss=0.64  (1842.5  examples/sec;  0.069  sec/batch)
step 14380, loss=0.70  (1882.2  examples/sec;  0.068  sec/batch)
step 14390, loss=0.71  (1776.3  examples/sec;  0.072  sec/batch)
step 14400, loss=0.73  (1967.1  examples/sec;  0.065  sec/batch)
step 14410, loss=0.85  (1893.7  examples/sec;  0.068  sec/batch)
step 14420, loss=0.79  (1931.4  examples/sec;  0.066  sec/batch)
step 14430, loss=0.66  (1905.8  examples/sec;  0.067  sec/batch)
step 14440, loss=0.96  (1910.5  examples/sec;  0.067  sec/batch)
step 14450, loss=0.86  (1907.5  examples/sec;  0.067  sec/batch)
step 14460, loss=0.72  (1934.0  examples/sec;  0.066  sec/batch)
step 14470, loss=0.83  (1843.0  examples/sec;  0.069  sec/batch)
step 14480, loss=0.78  (1894.6  examples/sec;  0.068  sec/batch)
step 14490, loss=0.79  (2005.2  examples/sec;  0.064  sec/batch)
step 14500, loss=0.81  (1885.2  examples/sec;  0.068  sec/batch)
step 14510, loss=0.70  (1915.7  examples/sec;  0.067  sec/batch)
step 14520, loss=0.66  (1837.6  examples/sec;  0.070  sec/batch)
step 14530, loss=0.78  (1920.8  examples/sec;  0.067  sec/batch)
step 14540, loss=0.82  (1821.9  examples/sec;  0.070  sec/batch)
step 14550, loss=0.83  (1684.5  examples/sec;  0.076  sec/batch)
step 14560, loss=0.85  (1816.3  examples/sec;  0.070  sec/batch)
step 14570, loss=0.87  (1890.6  examples/sec;  0.068  sec/batch)
step 14580, loss=0.86  (1851.7  examples/sec;  0.069  sec/batch)
step 14590, loss=0.79  (1812.5  examples/sec;  0.071  sec/batch)
step 14600, loss=0.73  (1869.9  examples/sec;  0.068  sec/batch)
step 14610, loss=0.75  (1957.0  examples/sec;  0.065  sec/batch)
step 14620, loss=0.86  (1974.1  examples/sec;  0.065  sec/batch)
step 14630, loss=0.75  (1933.5  examples/sec;  0.066  sec/batch)
step 14640, loss=0.73  (1892.1  examples/sec;  0.068  sec/batch)
step 14650, loss=0.85  (1986.1  examples/sec;  0.064  sec/batch)
step 14660, loss=0.74  (1930.0  examples/sec;  0.066  sec/batch)
step 14670, loss=1.02  (1882.7  examples/sec;  0.068  sec/batch)
step 14680, loss=0.85  (1855.2  examples/sec;  0.069  sec/batch)
step 14690, loss=0.92  (1814.0  examples/sec;  0.071  sec/batch)
step 14700, loss=0.79  (1758.7  examples/sec;  0.073  sec/batch)
step 14710, loss=0.80  (1888.3  examples/sec;  0.068  sec/batch)
step 14720, loss=0.72  (1876.4  examples/sec;  0.068  sec/batch)
step 14730, loss=0.82  (1863.4  examples/sec;  0.069  sec/batch)
step 14740, loss=0.94  (1902.8  examples/sec;  0.067  sec/batch)
step 14750, loss=0.81  (1912.0  examples/sec;  0.067  sec/batch)
step 14760, loss=0.89  (1842.4  examples/sec;  0.069  sec/batch)
step 14770, loss=0.86  (1871.2  examples/sec;  0.068  sec/batch)
step 14780, loss=0.81  (1938.2  examples/sec;  0.066  sec/batch)
step 14790, loss=0.76  (1807.6  examples/sec;  0.071  sec/batch)
step 14800, loss=0.72  (1892.8  examples/sec;  0.068  sec/batch)
step 14810, loss=0.89  (1835.1  examples/sec;  0.070  sec/batch)
step 14820, loss=0.87  (1834.0  examples/sec;  0.070  sec/batch)
step 14830, loss=0.76  (1934.2  examples/sec;  0.066  sec/batch)
step 14840, loss=0.77  (1673.2  examples/sec;  0.077  sec/batch)
step 14850, loss=0.71  (1826.6  examples/sec;  0.070  sec/batch)
step 14860, loss=0.75  (1878.7  examples/sec;  0.068  sec/batch)
step 14870, loss=0.80  (1840.3  examples/sec;  0.070  sec/batch)
step 14880, loss=0.71  (1904.6  examples/sec;  0.067  sec/batch)
step 14890, loss=0.71  (1864.8  examples/sec;  0.069  sec/batch)
step 14900, loss=0.72  (1816.5  examples/sec;  0.070  sec/batch)
step 14910, loss=0.72  (1811.6  examples/sec;  0.071  sec/batch)
step 14920, loss=0.76  (1841.7  examples/sec;  0.070  sec/batch)
step 14930, loss=0.71  (1938.6  examples/sec;  0.066  sec/batch)
step 14940, loss=0.68  (1940.3  examples/sec;  0.066  sec/batch)
step 14950, loss=0.77  (1811.5  examples/sec;  0.071  sec/batch)
step 14960, loss=0.85  (1805.6  examples/sec;  0.071  sec/batch)
step 14970, loss=0.70  (2013.2  examples/sec;  0.064  sec/batch)
step 14980, loss=0.72  (1899.5  examples/sec;  0.067  sec/batch)
step 14990, loss=0.83  (1825.6  examples/sec;  0.070  sec/batch)
step 15000, loss=0.77  (1877.5  examples/sec;  0.068  sec/batch)
step 15010, loss=0.72  (1945.4  examples/sec;  0.066  sec/batch)
step 15020, loss=0.68  (1849.7  examples/sec;  0.069  sec/batch)
step 15030, loss=0.74  (1858.8  examples/sec;  0.069  sec/batch)
step 15040, loss=0.78  (1866.2  examples/sec;  0.069  sec/batch)
step 15050, loss=0.78  (1930.7  examples/sec;  0.066  sec/batch)
step 15060, loss=0.70  (1835.4  examples/sec;  0.070  sec/batch)
step 15070, loss=0.86  (1834.0  examples/sec;  0.070  sec/batch)
step 15080, loss=0.84  (1816.9  examples/sec;  0.070  sec/batch)
step 15090, loss=0.73  (1831.0  examples/sec;  0.070  sec/batch)
step 15100, loss=0.88  (1895.8  examples/sec;  0.068  sec/batch)
step 15110, loss=0.63  (1825.7  examples/sec;  0.070  sec/batch)
step 15120, loss=0.79  (1862.5  examples/sec;  0.069  sec/batch)
step 15130, loss=0.84  (1615.5  examples/sec;  0.079  sec/batch)
step 15140, loss=0.72  (1914.3  examples/sec;  0.067  sec/batch)
step 15150, loss=0.78  (1739.8  examples/sec;  0.074  sec/batch)
step 15160, loss=0.88  (1934.1  examples/sec;  0.066  sec/batch)
step 15170, loss=0.83  (1956.5  examples/sec;  0.065  sec/batch)
step 15180, loss=0.71  (1874.7  examples/sec;  0.068  sec/batch)
step 15190, loss=0.79  (1830.7  examples/sec;  0.070  sec/batch)
step 15200, loss=0.99  (1895.4  examples/sec;  0.068  sec/batch)
step 15210, loss=0.68  (1961.2  examples/sec;  0.065  sec/batch)
step 15220, loss=0.64  (1962.5  examples/sec;  0.065  sec/batch)
step 15230, loss=0.69  (1927.5  examples/sec;  0.066  sec/batch)
step 15240, loss=0.77  (1852.3  examples/sec;  0.069  sec/batch)
step 15250, loss=0.87  (1881.1  examples/sec;  0.068  sec/batch)
step 15260, loss=0.67  (1869.9  examples/sec;  0.068  sec/batch)
step 15270, loss=0.78  (1901.1  examples/sec;  0.067  sec/batch)
step 15280, loss=0.75  (1865.4  examples/sec;  0.069  sec/batch)
step 15290, loss=0.61  (1931.4  examples/sec;  0.066  sec/batch)
step 15300, loss=0.73  (1857.2  examples/sec;  0.069  sec/batch)
step 15310, loss=0.61  (1776.9  examples/sec;  0.072  sec/batch)
step 15320, loss=0.92  (1926.1  examples/sec;  0.066  sec/batch)
step 15330, loss=1.01  (1849.0  examples/sec;  0.069  sec/batch)
step 15340, loss=0.70  (1850.9  examples/sec;  0.069  sec/batch)
step 15350, loss=0.79  (1815.3  examples/sec;  0.071  sec/batch)
step 15360, loss=0.81  (1833.4  examples/sec;  0.070  sec/batch)
step 15370, loss=0.68  (1916.3  examples/sec;  0.067  sec/batch)
step 15380, loss=0.88  (1857.9  examples/sec;  0.069  sec/batch)
step 15390, loss=0.90  (1863.7  examples/sec;  0.069  sec/batch)
step 15400, loss=0.73  (1832.8  examples/sec;  0.070  sec/batch)
step 15410, loss=0.74  (1914.1  examples/sec;  0.067  sec/batch)
step 15420, loss=0.72  (1854.8  examples/sec;  0.069  sec/batch)
step 15430, loss=0.69  (1847.9  examples/sec;  0.069  sec/batch)
step 15440, loss=0.88  (1842.2  examples/sec;  0.069  sec/batch)
step 15450, loss=0.82  (1857.2  examples/sec;  0.069  sec/batch)
step 15460, loss=0.81  (1866.1  examples/sec;  0.069  sec/batch)
step 15470, loss=0.80  (1741.1  examples/sec;  0.074  sec/batch)
step 15480, loss=0.91  (1974.2  examples/sec;  0.065  sec/batch)
step 15490, loss=0.73  (1823.3  examples/sec;  0.070  sec/batch)
step 15500, loss=0.82  (1906.3  examples/sec;  0.067  sec/batch)
step 15510, loss=0.66  (1841.3  examples/sec;  0.070  sec/batch)
step 15520, loss=0.84  (1831.8  examples/sec;  0.070  sec/batch)
step 15530, loss=0.73  (1878.0  examples/sec;  0.068  sec/batch)
step 15540, loss=0.85  (1849.0  examples/sec;  0.069  sec/batch)
step 15550, loss=0.86  (1958.6  examples/sec;  0.065  sec/batch)
step 15560, loss=0.64  (1756.2  examples/sec;  0.073  sec/batch)
step 15570, loss=0.75  (1867.0  examples/sec;  0.069  sec/batch)
step 15580, loss=0.88  (1851.8  examples/sec;  0.069  sec/batch)
step 15590, loss=0.72  (1872.2  examples/sec;  0.068  sec/batch)
step 15600, loss=0.67  (1891.8  examples/sec;  0.068  sec/batch)
step 15610, loss=0.84  (1866.7  examples/sec;  0.069  sec/batch)
step 15620, loss=0.71  (1865.7  examples/sec;  0.069  sec/batch)
step 15630, loss=0.85  (1930.1  examples/sec;  0.066  sec/batch)
step 15640, loss=0.84  (1950.4  examples/sec;  0.066  sec/batch)
step 15650, loss=0.54  (1893.8  examples/sec;  0.068  sec/batch)
step 15660, loss=0.72  (1887.8  examples/sec;  0.068  sec/batch)
step 15670, loss=0.79  (1909.3  examples/sec;  0.067  sec/batch)
step 15680, loss=0.88  (1919.8  examples/sec;  0.067  sec/batch)
step 15690, loss=0.79  (1834.9  examples/sec;  0.070  sec/batch)
step 15700, loss=0.87  (1935.8  examples/sec;  0.066  sec/batch)
step 15710, loss=0.84  (1853.9  examples/sec;  0.069  sec/batch)
step 15720, loss=0.75  (1820.2  examples/sec;  0.070  sec/batch)
step 15730, loss=0.80  (1852.3  examples/sec;  0.069  sec/batch)
step 15740, loss=0.67  (1874.3  examples/sec;  0.068  sec/batch)
step 15750, loss=0.65  (1778.3  examples/sec;  0.072  sec/batch)
step 15760, loss=0.82  (1879.6  examples/sec;  0.068  sec/batch)
step 15770, loss=0.65  (1844.9  examples/sec;  0.069  sec/batch)
step 15780, loss=0.69  (1906.3  examples/sec;  0.067  sec/batch)
step 15790, loss=0.73  (1851.9  examples/sec;  0.069  sec/batch)
step 15800, loss=0.68  (1869.1  examples/sec;  0.068  sec/batch)
step 15810, loss=0.70  (1849.4  examples/sec;  0.069  sec/batch)
step 15820, loss=0.80  (1612.9  examples/sec;  0.079  sec/batch)
step 15830, loss=0.71  (1838.6  examples/sec;  0.070  sec/batch)
step 15840, loss=0.82  (1962.6  examples/sec;  0.065  sec/batch)
step 15850, loss=0.74  (1898.3  examples/sec;  0.067  sec/batch)
step 15860, loss=0.73  (1986.1  examples/sec;  0.064  sec/batch)
step 15870, loss=0.84  (1929.3  examples/sec;  0.066  sec/batch)
step 15880, loss=0.99  (1764.0  examples/sec;  0.073  sec/batch)
step 15890, loss=0.61  (1903.2  examples/sec;  0.067  sec/batch)
step 15900, loss=0.59  (1920.8  examples/sec;  0.067  sec/batch)
step 15910, loss=0.86  (1661.6  examples/sec;  0.077  sec/batch)
step 15920, loss=0.76  (1967.0  examples/sec;  0.065  sec/batch)
step 15930, loss=0.86  (1825.7  examples/sec;  0.070  sec/batch)
step 15940, loss=0.69  (1829.2  examples/sec;  0.070  sec/batch)
step 15950, loss=0.77  (1792.6  examples/sec;  0.071  sec/batch)
step 15960, loss=0.69  (1811.8  examples/sec;  0.071  sec/batch)
step 15970, loss=0.78  (1849.9  examples/sec;  0.069  sec/batch)
step 15980, loss=0.72  (1894.9  examples/sec;  0.068  sec/batch)
step 15990, loss=0.62  (1798.5  examples/sec;  0.071  sec/batch)
step 16000, loss=0.83  (1856.4  examples/sec;  0.069  sec/batch)
step 16010, loss=0.69  (1940.9  examples/sec;  0.066  sec/batch)
step 16020, loss=0.81  (1840.7  examples/sec;  0.070  sec/batch)
step 16030, loss=0.86  (1509.9  examples/sec;  0.085  sec/batch)
step 16040, loss=0.75  (1928.9  examples/sec;  0.066  sec/batch)
step 16050, loss=0.85  (1859.9  examples/sec;  0.069  sec/batch)
step 16060, loss=0.84  (1959.6  examples/sec;  0.065  sec/batch)
step 16070, loss=0.85  (1870.9  examples/sec;  0.068  sec/batch)
step 16080, loss=0.85  (1932.5  examples/sec;  0.066  sec/batch)
step 16090, loss=0.69  (1885.4  examples/sec;  0.068  sec/batch)
step 16100, loss=0.84  (1818.3  examples/sec;  0.070  sec/batch)
step 16110, loss=0.69  (1951.7  examples/sec;  0.066  sec/batch)
step 16120, loss=0.70  (1930.8  examples/sec;  0.066  sec/batch)
step 16130, loss=0.65  (1710.4  examples/sec;  0.075  sec/batch)
step 16140, loss=0.67  (1884.0  examples/sec;  0.068  sec/batch)
step 16150, loss=0.63  (1878.0  examples/sec;  0.068  sec/batch)
step 16160, loss=0.92  (1969.5  examples/sec;  0.065  sec/batch)
step 16170, loss=0.85  (1936.8  examples/sec;  0.066  sec/batch)
step 16180, loss=0.79  (1888.1  examples/sec;  0.068  sec/batch)
step 16190, loss=0.89  (1920.7  examples/sec;  0.067  sec/batch)
step 16200, loss=0.85  (1776.2  examples/sec;  0.072  sec/batch)
step 16210, loss=0.88  (1966.7  examples/sec;  0.065  sec/batch)
step 16220, loss=0.86  (1990.3  examples/sec;  0.064  sec/batch)
step 16230, loss=0.77  (1914.0  examples/sec;  0.067  sec/batch)
step 16240, loss=0.60  (1883.3  examples/sec;  0.068  sec/batch)
step 16250, loss=0.73  (1829.8  examples/sec;  0.070  sec/batch)
step 16260, loss=0.85  (1879.4  examples/sec;  0.068  sec/batch)
step 16270, loss=0.64  (1908.2  examples/sec;  0.067  sec/batch)
step 16280, loss=0.85  (1823.7  examples/sec;  0.070  sec/batch)
step 16290, loss=0.85  (1911.4  examples/sec;  0.067  sec/batch)
step 16300, loss=0.94  (1839.2  examples/sec;  0.070  sec/batch)
step 16310, loss=0.73  (1895.2  examples/sec;  0.068  sec/batch)
step 16320, loss=0.76  (1849.6  examples/sec;  0.069  sec/batch)
step 16330, loss=0.82  (1814.7  examples/sec;  0.071  sec/batch)
step 16340, loss=0.73  (1814.8  examples/sec;  0.071  sec/batch)
step 16350, loss=0.89  (1999.9  examples/sec;  0.064  sec/batch)
step 16360, loss=0.64  (1840.3  examples/sec;  0.070  sec/batch)
step 16370, loss=0.93  (1895.0  examples/sec;  0.068  sec/batch)
step 16380, loss=1.04  (1808.1  examples/sec;  0.071  sec/batch)
step 16390, loss=0.83  (1779.2  examples/sec;  0.072  sec/batch)
step 16400, loss=0.65  (1837.1  examples/sec;  0.070  sec/batch)
step 16410, loss=0.67  (1893.5  examples/sec;  0.068  sec/batch)
step 16420, loss=0.85  (1849.7  examples/sec;  0.069  sec/batch)
step 16430, loss=0.59  (1891.8  examples/sec;  0.068  sec/batch)
step 16440, loss=0.79  (1880.0  examples/sec;  0.068  sec/batch)
step 16450, loss=0.78  (1840.1  examples/sec;  0.070  sec/batch)
step 16460, loss=0.69  (1937.2  examples/sec;  0.066  sec/batch)
step 16470, loss=0.95  (1831.0  examples/sec;  0.070  sec/batch)
step 16480, loss=0.75  (1876.2  examples/sec;  0.068  sec/batch)
step 16490, loss=0.62  (1883.4  examples/sec;  0.068  sec/batch)
step 16500, loss=0.77  (1884.3  examples/sec;  0.068  sec/batch)
step 16510, loss=0.72  (1796.1  examples/sec;  0.071  sec/batch)
step 16520, loss=0.72  (1835.1  examples/sec;  0.070  sec/batch)
step 16530, loss=0.82  (1887.3  examples/sec;  0.068  sec/batch)
step 16540, loss=0.95  (1838.7  examples/sec;  0.070  sec/batch)
step 16550, loss=0.64  (1871.0  examples/sec;  0.068  sec/batch)
step 16560, loss=0.85  (1796.2  examples/sec;  0.071  sec/batch)
step 16570, loss=0.77  (1858.8  examples/sec;  0.069  sec/batch)
step 16580, loss=0.94  (1830.4  examples/sec;  0.070  sec/batch)
step 16590, loss=0.80  (1849.9  examples/sec;  0.069  sec/batch)
step 16600, loss=0.68  (1959.6  examples/sec;  0.065  sec/batch)
step 16610, loss=0.92  (1779.3  examples/sec;  0.072  sec/batch)
step 16620, loss=0.78  (1901.1  examples/sec;  0.067  sec/batch)
step 16630, loss=0.75  (1567.2  examples/sec;  0.082  sec/batch)
step 16640, loss=0.68  (1884.5  examples/sec;  0.068  sec/batch)
step 16650, loss=0.87  (1942.5  examples/sec;  0.066  sec/batch)
step 16660, loss=0.77  (1925.0  examples/sec;  0.066  sec/batch)
step 16670, loss=0.86  (1927.3  examples/sec;  0.066  sec/batch)
step 16680, loss=0.59  (1885.9  examples/sec;  0.068  sec/batch)
step 16690, loss=0.64  (1900.8  examples/sec;  0.067  sec/batch)
step 16700, loss=0.66  (1702.2  examples/sec;  0.075  sec/batch)
step 16710, loss=0.85  (1928.7  examples/sec;  0.066  sec/batch)
step 16720, loss=0.76  (1958.6  examples/sec;  0.065  sec/batch)
step 16730, loss=0.77  (1867.0  examples/sec;  0.069  sec/batch)
step 16740, loss=0.79  (1852.6  examples/sec;  0.069  sec/batch)
step 16750, loss=0.72  (1773.3  examples/sec;  0.072  sec/batch)
step 16760, loss=0.82  (1913.0  examples/sec;  0.067  sec/batch)
step 16770, loss=0.95  (1866.6  examples/sec;  0.069  sec/batch)
step 16780, loss=0.83  (1873.8  examples/sec;  0.068  sec/batch)
step 16790, loss=0.62  (1934.3  examples/sec;  0.066  sec/batch)
step 16800, loss=0.77  (1938.4  examples/sec;  0.066  sec/batch)
step 16810, loss=0.78  (1849.7  examples/sec;  0.069  sec/batch)
step 16820, loss=0.89  (1836.2  examples/sec;  0.070  sec/batch)
step 16830, loss=0.85  (1958.1  examples/sec;  0.065  sec/batch)
step 16840, loss=0.88  (1977.8  examples/sec;  0.065  sec/batch)
step 16850, loss=0.86  (1753.3  examples/sec;  0.073  sec/batch)
step 16860, loss=0.62  (1922.0  examples/sec;  0.067  sec/batch)
step 16870, loss=0.68  (1860.9  examples/sec;  0.069  sec/batch)
step 16880, loss=0.80  (1666.3  examples/sec;  0.077  sec/batch)
step 16890, loss=0.80  (1889.8  examples/sec;  0.068  sec/batch)
step 16900, loss=0.79  (1961.2  examples/sec;  0.065  sec/batch)
step 16910, loss=0.84  (1956.4  examples/sec;  0.065  sec/batch)
step 16920, loss=0.74  (1898.4  examples/sec;  0.067  sec/batch)
step 16930, loss=0.83  (1911.8  examples/sec;  0.067  sec/batch)
step 16940, loss=0.70  (1839.4  examples/sec;  0.070  sec/batch)
step 16950, loss=0.67  (1893.7  examples/sec;  0.068  sec/batch)
step 16960, loss=0.61  (1876.7  examples/sec;  0.068  sec/batch)
step 16970, loss=0.79  (1892.1  examples/sec;  0.068  sec/batch)
step 16980, loss=0.67  (1884.7  examples/sec;  0.068  sec/batch)
step 16990, loss=0.68  (1943.0  examples/sec;  0.066  sec/batch)
step 17000, loss=0.77  (1831.5  examples/sec;  0.070  sec/batch)
step 17010, loss=0.79  (1849.6  examples/sec;  0.069  sec/batch)
step 17020, loss=0.74  (1795.3  examples/sec;  0.071  sec/batch)
step 17030, loss=0.65  (1886.3  examples/sec;  0.068  sec/batch)
step 17040, loss=0.73  (1861.2  examples/sec;  0.069  sec/batch)
step 17050, loss=0.98  (1805.8  examples/sec;  0.071  sec/batch)
step 17060, loss=0.88  (1875.8  examples/sec;  0.068  sec/batch)
step 17070, loss=0.70  (1838.7  examples/sec;  0.070  sec/batch)
step 17080, loss=0.78  (1877.8  examples/sec;  0.068  sec/batch)
step 17090, loss=0.77  (1950.2  examples/sec;  0.066  sec/batch)
step 17100, loss=0.59  (1838.4  examples/sec;  0.070  sec/batch)
step 17110, loss=0.71  (1902.7  examples/sec;  0.067  sec/batch)
step 17120, loss=0.97  (1859.0  examples/sec;  0.069  sec/batch)
step 17130, loss=0.91  (1901.4  examples/sec;  0.067  sec/batch)
step 17140, loss=0.82  (1916.4  examples/sec;  0.067  sec/batch)
step 17150, loss=0.68  (1867.7  examples/sec;  0.069  sec/batch)
step 17160, loss=0.53  (1864.1  examples/sec;  0.069  sec/batch)
step 17170, loss=0.77  (1928.9  examples/sec;  0.066  sec/batch)
step 17180, loss=0.60  (1904.5  examples/sec;  0.067  sec/batch)
step 17190, loss=0.98  (1870.3  examples/sec;  0.068  sec/batch)
step 17200, loss=0.83  (1874.9  examples/sec;  0.068  sec/batch)
step 17210, loss=0.80  (1842.2  examples/sec;  0.069  sec/batch)
step 17220, loss=0.76  (1921.5  examples/sec;  0.067  sec/batch)
step 17230, loss=0.71  (1943.6  examples/sec;  0.066  sec/batch)
step 17240, loss=0.76  (1860.9  examples/sec;  0.069  sec/batch)
step 17250, loss=0.65  (1826.4  examples/sec;  0.070  sec/batch)
step 17260, loss=0.71  (1942.7  examples/sec;  0.066  sec/batch)
step 17270, loss=0.70  (1990.1  examples/sec;  0.064  sec/batch)
step 17280, loss=0.75  (1916.3  examples/sec;  0.067  sec/batch)
step 17290, loss=0.63  (1829.2  examples/sec;  0.070  sec/batch)
step 17300, loss=0.51  (1781.6  examples/sec;  0.072  sec/batch)
step 17310, loss=0.94  (1821.9  examples/sec;  0.070  sec/batch)
step 17320, loss=0.87  (1798.2  examples/sec;  0.071  sec/batch)
step 17330, loss=0.78  (1848.9  examples/sec;  0.069  sec/batch)
step 17340, loss=0.73  (1851.5  examples/sec;  0.069  sec/batch)
step 17350, loss=0.81  (1872.7  examples/sec;  0.068  sec/batch)
step 17360, loss=0.92  (1807.0  examples/sec;  0.071  sec/batch)
step 17370, loss=0.92  (1887.2  examples/sec;  0.068  sec/batch)
step 17380, loss=0.75  (1898.9  examples/sec;  0.067  sec/batch)
step 17390, loss=0.87  (1823.8  examples/sec;  0.070  sec/batch)
step 17400, loss=0.86  (1892.4  examples/sec;  0.068  sec/batch)
step 17410, loss=0.78  (1798.9  examples/sec;  0.071  sec/batch)
step 17420, loss=0.90  (1765.1  examples/sec;  0.073  sec/batch)
step 17430, loss=0.78  (1945.6  examples/sec;  0.066  sec/batch)
step 17440, loss=0.79  (1877.8  examples/sec;  0.068  sec/batch)
step 17450, loss=0.76  (1819.8  examples/sec;  0.070  sec/batch)
step 17460, loss=0.76  (1836.6  examples/sec;  0.070  sec/batch)
step 17470, loss=0.85  (1833.6  examples/sec;  0.070  sec/batch)
step 17480, loss=0.74  (1805.3  examples/sec;  0.071  sec/batch)
step 17490, loss=0.86  (1851.5  examples/sec;  0.069  sec/batch)
step 17500, loss=0.75  (1858.5  examples/sec;  0.069  sec/batch)
step 17510, loss=0.81  (1843.6  examples/sec;  0.069  sec/batch)
step 17520, loss=0.74  (1846.5  examples/sec;  0.069  sec/batch)
step 17530, loss=0.73  (1853.9  examples/sec;  0.069  sec/batch)
step 17540, loss=0.57  (1938.3  examples/sec;  0.066  sec/batch)
step 17550, loss=0.69  (1836.9  examples/sec;  0.070  sec/batch)
step 17560, loss=0.86  (1972.7  examples/sec;  0.065  sec/batch)
step 17570, loss=0.82  (1942.4  examples/sec;  0.066  sec/batch)
step 17580, loss=0.67  (1868.3  examples/sec;  0.069  sec/batch)
step 17590, loss=0.71  (1955.6  examples/sec;  0.065  sec/batch)
step 17600, loss=0.77  (1848.8  examples/sec;  0.069  sec/batch)
step 17610, loss=0.69  (1892.4  examples/sec;  0.068  sec/batch)
step 17620, loss=0.95  (1830.6  examples/sec;  0.070  sec/batch)
step 17630, loss=0.59  (1982.9  examples/sec;  0.065  sec/batch)
step 17640, loss=0.86  (1855.5  examples/sec;  0.069  sec/batch)
step 17650, loss=0.99  (1750.3  examples/sec;  0.073  sec/batch)
step 17660, loss=0.73  (1862.2  examples/sec;  0.069  sec/batch)
step 17670, loss=0.82  (1894.5  examples/sec;  0.068  sec/batch)
step 17680, loss=0.79  (1785.0  examples/sec;  0.072  sec/batch)
step 17690, loss=0.75  (1652.0  examples/sec;  0.077  sec/batch)
step 17700, loss=0.77  (1831.9  examples/sec;  0.070  sec/batch)
step 17710, loss=0.60  (1856.4  examples/sec;  0.069  sec/batch)
step 17720, loss=0.88  (1928.9  examples/sec;  0.066  sec/batch)
step 17730, loss=0.80  (1832.6  examples/sec;  0.070  sec/batch)
step 17740, loss=0.73  (1878.3  examples/sec;  0.068  sec/batch)
step 17750, loss=0.73  (1976.3  examples/sec;  0.065  sec/batch)
step 17760, loss=0.70  (1930.8  examples/sec;  0.066  sec/batch)
step 17770, loss=0.74  (1938.7  examples/sec;  0.066  sec/batch)
step 17780, loss=0.69  (1817.3  examples/sec;  0.070  sec/batch)
step 17790, loss=0.70  (1880.7  examples/sec;  0.068  sec/batch)
step 17800, loss=0.65  (1931.2  examples/sec;  0.066  sec/batch)
step 17810, loss=0.81  (1883.2  examples/sec;  0.068  sec/batch)
step 17820, loss=0.83  (1912.6  examples/sec;  0.067  sec/batch)
step 17830, loss=0.82  (1938.9  examples/sec;  0.066  sec/batch)
step 17840, loss=0.77  (1829.7  examples/sec;  0.070  sec/batch)
step 17850, loss=0.83  (1894.5  examples/sec;  0.068  sec/batch)
step 17860, loss=0.64  (1723.5  examples/sec;  0.074  sec/batch)
step 17870, loss=0.78  (1875.3  examples/sec;  0.068  sec/batch)
step 17880, loss=0.70  (1956.7  examples/sec;  0.065  sec/batch)
step 17890, loss=0.73  (1916.4  examples/sec;  0.067  sec/batch)
step 17900, loss=0.70  (1889.0  examples/sec;  0.068  sec/batch)
step 17910, loss=0.71  (1884.6  examples/sec;  0.068  sec/batch)
step 17920, loss=0.83  (1766.1  examples/sec;  0.072  sec/batch)
step 17930, loss=0.92  (1887.8  examples/sec;  0.068  sec/batch)
step 17940, loss=0.67  (1911.9  examples/sec;  0.067  sec/batch)
step 17950, loss=0.80  (1947.8  examples/sec;  0.066  sec/batch)
step 17960, loss=0.85  (1681.1  examples/sec;  0.076  sec/batch)
step 17970, loss=0.66  (1899.6  examples/sec;  0.067  sec/batch)
step 17980, loss=0.72  (1799.2  examples/sec;  0.071  sec/batch)
step 17990, loss=0.72  (1859.2  examples/sec;  0.069  sec/batch)
step 18000, loss=0.71  (1896.6  examples/sec;  0.067  sec/batch)
step 18010, loss=0.82  (1794.0  examples/sec;  0.071  sec/batch)
step 18020, loss=0.70  (1806.5  examples/sec;  0.071  sec/batch)
step 18030, loss=0.78  (1958.7  examples/sec;  0.065  sec/batch)
step 18040, loss=0.76  (1955.2  examples/sec;  0.065  sec/batch)
step 18050, loss=0.67  (1935.4  examples/sec;  0.066  sec/batch)
step 18060, loss=0.70  (1856.4  examples/sec;  0.069  sec/batch)
step 18070, loss=0.80  (1892.9  examples/sec;  0.068  sec/batch)
step 18080, loss=0.77  (1715.8  examples/sec;  0.075  sec/batch)
step 18090, loss=0.80  (1938.5  examples/sec;  0.066  sec/batch)
step 18100, loss=0.57  (1947.6  examples/sec;  0.066  sec/batch)
step 18110, loss=0.81  (1883.1  examples/sec;  0.068  sec/batch)
step 18120, loss=0.83  (1807.1  examples/sec;  0.071  sec/batch)
step 18130, loss=0.71  (1898.9  examples/sec;  0.067  sec/batch)
step 18140, loss=0.70  (1878.4  examples/sec;  0.068  sec/batch)
step 18150, loss=0.74  (1855.2  examples/sec;  0.069  sec/batch)
step 18160, loss=0.63  (1859.5  examples/sec;  0.069  sec/batch)
step 18170, loss=0.83  (1815.4  examples/sec;  0.071  sec/batch)
step 18180, loss=0.74  (1784.7  examples/sec;  0.072  sec/batch)
step 18190, loss=0.69  (1828.5  examples/sec;  0.070  sec/batch)
step 18200, loss=0.68  (1875.1  examples/sec;  0.068  sec/batch)
step 18210, loss=0.74  (1902.0  examples/sec;  0.067  sec/batch)
step 18220, loss=0.77  (1939.8  examples/sec;  0.066  sec/batch)
step 18230, loss=0.82  (1783.2  examples/sec;  0.072  sec/batch)
step 18240, loss=0.79  (1919.7  examples/sec;  0.067  sec/batch)
step 18250, loss=0.92  (1928.4  examples/sec;  0.066  sec/batch)
step 18260, loss=0.75  (1899.5  examples/sec;  0.067  sec/batch)
step 18270, loss=0.78  (1725.4  examples/sec;  0.074  sec/batch)
step 18280, loss=0.74  (1850.3  examples/sec;  0.069  sec/batch)
step 18290, loss=0.75  (1841.5  examples/sec;  0.070  sec/batch)
step 18300, loss=0.68  (1853.5  examples/sec;  0.069  sec/batch)
step 18310, loss=0.78  (1909.4  examples/sec;  0.067  sec/batch)
step 18320, loss=0.74  (1869.9  examples/sec;  0.068  sec/batch)
step 18330, loss=0.75  (1965.6  examples/sec;  0.065  sec/batch)
step 18340, loss=0.72  (1903.1  examples/sec;  0.067  sec/batch)
step 18350, loss=0.79  (1971.5  examples/sec;  0.065  sec/batch)
step 18360, loss=0.67  (1765.7  examples/sec;  0.072  sec/batch)
step 18370, loss=0.68  (1877.4  examples/sec;  0.068  sec/batch)
step 18380, loss=0.83  (1911.0  examples/sec;  0.067  sec/batch)
step 18390, loss=0.71  (1841.3  examples/sec;  0.070  sec/batch)
step 18400, loss=0.77  (1807.9  examples/sec;  0.071  sec/batch)
step 18410, loss=0.70  (1915.1  examples/sec;  0.067  sec/batch)
step 18420, loss=0.89  (1810.0  examples/sec;  0.071  sec/batch)
step 18430, loss=0.67  (1700.7  examples/sec;  0.075  sec/batch)
step 18440, loss=0.75  (1849.4  examples/sec;  0.069  sec/batch)
step 18450, loss=0.67  (1920.9  examples/sec;  0.067  sec/batch)
step 18460, loss=0.87  (1802.3  examples/sec;  0.071  sec/batch)
step 18470, loss=0.70  (1847.5  examples/sec;  0.069  sec/batch)
step 18480, loss=0.93  (1829.4  examples/sec;  0.070  sec/batch)
step 18490, loss=0.60  (1803.7  examples/sec;  0.071  sec/batch)
step 18500, loss=0.73  (1940.4  examples/sec;  0.066  sec/batch)
step 18510, loss=0.72  (1603.4  examples/sec;  0.080  sec/batch)
step 18520, loss=0.94  (1881.2  examples/sec;  0.068  sec/batch)
step 18530, loss=0.74  (1880.2  examples/sec;  0.068  sec/batch)
step 18540, loss=0.93  (1890.4  examples/sec;  0.068  sec/batch)
step 18550, loss=0.93  (1886.7  examples/sec;  0.068  sec/batch)
step 18560, loss=0.89  (1893.2  examples/sec;  0.068  sec/batch)
step 18570, loss=0.67  (1825.0  examples/sec;  0.070  sec/batch)
step 18580, loss=0.68  (1855.2  examples/sec;  0.069  sec/batch)
step 18590, loss=0.76  (2019.0  examples/sec;  0.063  sec/batch)
step 18600, loss=0.72  (1813.4  examples/sec;  0.071  sec/batch)
step 18610, loss=0.66  (1869.0  examples/sec;  0.068  sec/batch)
step 18620, loss=0.69  (1984.4  examples/sec;  0.065  sec/batch)
step 18630, loss=0.75  (1844.6  examples/sec;  0.069  sec/batch)
step 18640, loss=0.77  (1895.3  examples/sec;  0.068  sec/batch)
step 18650, loss=0.86  (1966.7  examples/sec;  0.065  sec/batch)
step 18660, loss=0.65  (1878.2  examples/sec;  0.068  sec/batch)
step 18670, loss=0.73  (1951.2  examples/sec;  0.066  sec/batch)
step 18680, loss=0.89  (1892.8  examples/sec;  0.068  sec/batch)
step 18690, loss=0.55  (1971.1  examples/sec;  0.065  sec/batch)
step 18700, loss=0.67  (1760.6  examples/sec;  0.073  sec/batch)
step 18710, loss=0.88  (1771.2  examples/sec;  0.072  sec/batch)
step 18720, loss=0.81  (1865.1  examples/sec;  0.069  sec/batch)
step 18730, loss=0.72  (1945.9  examples/sec;  0.066  sec/batch)
step 18740, loss=0.83  (1924.8  examples/sec;  0.066  sec/batch)
step 18750, loss=0.82  (1919.3  examples/sec;  0.067  sec/batch)
step 18760, loss=0.60  (1840.2  examples/sec;  0.070  sec/batch)
step 18770, loss=0.71  (2001.4  examples/sec;  0.064  sec/batch)
step 18780, loss=0.83  (1903.4  examples/sec;  0.067  sec/batch)
step 18790, loss=0.59  (1889.9  examples/sec;  0.068  sec/batch)
step 18800, loss=0.76  (1929.9  examples/sec;  0.066  sec/batch)
step 18810, loss=0.68  (1901.5  examples/sec;  0.067  sec/batch)
step 18820, loss=0.77  (1922.3  examples/sec;  0.067  sec/batch)
step 18830, loss=0.78  (1982.7  examples/sec;  0.065  sec/batch)
step 18840, loss=0.73  (1899.2  examples/sec;  0.067  sec/batch)
step 18850, loss=0.86  (1881.4  examples/sec;  0.068  sec/batch)
step 18860, loss=0.67  (1880.9  examples/sec;  0.068  sec/batch)
step 18870, loss=0.65  (1805.1  examples/sec;  0.071  sec/batch)
step 18880, loss=0.84  (1868.6  examples/sec;  0.068  sec/batch)
step 18890, loss=0.65  (1803.9  examples/sec;  0.071  sec/batch)
step 18900, loss=0.73  (1936.0  examples/sec;  0.066  sec/batch)
step 18910, loss=0.76  (1826.5  examples/sec;  0.070  sec/batch)
step 18920, loss=0.76  (1918.9  examples/sec;  0.067  sec/batch)
step 18930, loss=0.80  (1802.5  examples/sec;  0.071  sec/batch)
step 18940, loss=0.63  (1874.5  examples/sec;  0.068  sec/batch)
step 18950, loss=1.00  (1889.3  examples/sec;  0.068  sec/batch)
step 18960, loss=0.71  (1842.9  examples/sec;  0.069  sec/batch)
step 18970, loss=0.74  (1817.6  examples/sec;  0.070  sec/batch)
step 18980, loss=0.73  (1902.7  examples/sec;  0.067  sec/batch)
step 18990, loss=0.58  (1896.0  examples/sec;  0.068  sec/batch)
step 19000, loss=0.66  (1967.4  examples/sec;  0.065  sec/batch)
step 19010, loss=0.71  (1921.6  examples/sec;  0.067  sec/batch)
step 19020, loss=0.71  (1972.6  examples/sec;  0.065  sec/batch)
step 19030, loss=0.89  (1879.1  examples/sec;  0.068  sec/batch)
step 19040, loss=0.63  (1981.8  examples/sec;  0.065  sec/batch)
step 19050, loss=0.85  (1954.6  examples/sec;  0.065  sec/batch)
step 19060, loss=0.70  (1722.7  examples/sec;  0.074  sec/batch)
step 19070, loss=0.64  (1917.1  examples/sec;  0.067  sec/batch)
step 19080, loss=0.80  (1809.4  examples/sec;  0.071  sec/batch)
step 19090, loss=0.87  (1968.4  examples/sec;  0.065  sec/batch)
step 19100, loss=0.83  (1900.1  examples/sec;  0.067  sec/batch)
step 19110, loss=0.65  (1830.4  examples/sec;  0.070  sec/batch)
step 19120, loss=0.47  (1851.8  examples/sec;  0.069  sec/batch)
step 19130, loss=0.79  (1799.3  examples/sec;  0.071  sec/batch)
step 19140, loss=0.77  (1844.2  examples/sec;  0.069  sec/batch)
step 19150, loss=0.75  (1906.5  examples/sec;  0.067  sec/batch)
step 19160, loss=0.74  (1941.4  examples/sec;  0.066  sec/batch)
step 19170, loss=0.86  (1868.3  examples/sec;  0.069  sec/batch)
step 19180, loss=0.79  (1862.8  examples/sec;  0.069  sec/batch)
step 19190, loss=0.83  (1881.9  examples/sec;  0.068  sec/batch)
step 19200, loss=0.72  (1911.5  examples/sec;  0.067  sec/batch)
step 19210, loss=0.76  (2003.0  examples/sec;  0.064  sec/batch)
step 19220, loss=0.64  (1930.9  examples/sec;  0.066  sec/batch)
step 19230, loss=0.69  (1736.2  examples/sec;  0.074  sec/batch)
step 19240, loss=0.86  (1916.9  examples/sec;  0.067  sec/batch)
step 19250, loss=0.81  (1813.8  examples/sec;  0.071  sec/batch)
step 19260, loss=0.62  (1860.5  examples/sec;  0.069  sec/batch)
step 19270, loss=0.84  (1873.5  examples/sec;  0.068  sec/batch)
step 19280, loss=0.88  (1837.4  examples/sec;  0.070  sec/batch)
step 19290, loss=0.73  (1888.3  examples/sec;  0.068  sec/batch)
step 19300, loss=0.65  (1869.6  examples/sec;  0.068  sec/batch)
step 19310, loss=0.88  (1951.5  examples/sec;  0.066  sec/batch)
step 19320, loss=0.83  (2013.9  examples/sec;  0.064  sec/batch)
step 19330, loss=0.74  (1848.9  examples/sec;  0.069  sec/batch)
step 19340, loss=0.77  (1914.7  examples/sec;  0.067  sec/batch)
step 19350, loss=0.80  (1935.0  examples/sec;  0.066  sec/batch)
step 19360, loss=0.74  (1849.9  examples/sec;  0.069  sec/batch)
step 19370, loss=0.67  (1920.6  examples/sec;  0.067  sec/batch)
step 19380, loss=0.78  (1746.2  examples/sec;  0.073  sec/batch)
step 19390, loss=0.81  (1969.9  examples/sec;  0.065  sec/batch)
step 19400, loss=0.82  (1903.1  examples/sec;  0.067  sec/batch)
step 19410, loss=0.72  (1863.7  examples/sec;  0.069  sec/batch)
step 19420, loss=0.76  (1896.1  examples/sec;  0.068  sec/batch)
step 19430, loss=0.82  (1881.8  examples/sec;  0.068  sec/batch)
step 19440, loss=0.92  (1860.9  examples/sec;  0.069  sec/batch)
step 19450, loss=0.72  (1942.0  examples/sec;  0.066  sec/batch)
step 19460, loss=0.95  (1991.3  examples/sec;  0.064  sec/batch)
step 19470, loss=0.71  (1917.7  examples/sec;  0.067  sec/batch)
step 19480, loss=0.60  (1922.6  examples/sec;  0.067  sec/batch)
step 19490, loss=0.84  (1574.2  examples/sec;  0.081  sec/batch)
step 19500, loss=0.67  (1915.3  examples/sec;  0.067  sec/batch)
step 19510, loss=0.68  (1959.0  examples/sec;  0.065  sec/batch)
step 19520, loss=0.74  (1883.8  examples/sec;  0.068  sec/batch)
step 19530, loss=0.66  (1908.1  examples/sec;  0.067  sec/batch)
step 19540, loss=0.71  (1932.1  examples/sec;  0.066  sec/batch)
step 19550, loss=0.63  (1887.9  examples/sec;  0.068  sec/batch)
step 19560, loss=0.65  (1826.2  examples/sec;  0.070  sec/batch)
step 19570, loss=0.74  (1900.7  examples/sec;  0.067  sec/batch)
step 19580, loss=0.72  (1980.4  examples/sec;  0.065  sec/batch)
step 19590, loss=0.78  (1871.0  examples/sec;  0.068  sec/batch)
step 19600, loss=0.75  (1870.0  examples/sec;  0.068  sec/batch)
step 19610, loss=0.81  (1775.8  examples/sec;  0.072  sec/batch)
step 19620, loss=0.63  (1876.4  examples/sec;  0.068  sec/batch)
step 19630, loss=0.90  (1942.2  examples/sec;  0.066  sec/batch)
step 19640, loss=0.81  (1936.8  examples/sec;  0.066  sec/batch)
step 19650, loss=0.72  (1669.3  examples/sec;  0.077  sec/batch)
step 19660, loss=0.85  (1880.7  examples/sec;  0.068  sec/batch)
step 19670, loss=0.70  (1843.4  examples/sec;  0.069  sec/batch)
step 19680, loss=0.86  (1954.3  examples/sec;  0.065  sec/batch)
step 19690, loss=0.60  (1867.9  examples/sec;  0.069  sec/batch)
step 19700, loss=0.72  (1870.0  examples/sec;  0.068  sec/batch)
step 19710, loss=0.95  (1805.3  examples/sec;  0.071  sec/batch)
step 19720, loss=0.68  (1863.0  examples/sec;  0.069  sec/batch)
step 19730, loss=0.60  (1738.7  examples/sec;  0.074  sec/batch)
step 19740, loss=0.75  (1944.2  examples/sec;  0.066  sec/batch)
step 19750, loss=0.77  (1877.4  examples/sec;  0.068  sec/batch)
step 19760, loss=0.77  (1935.2  examples/sec;  0.066  sec/batch)
step 19770, loss=0.81  (1957.7  examples/sec;  0.065  sec/batch)
step 19780, loss=0.78  (1862.3  examples/sec;  0.069  sec/batch)
step 19790, loss=0.85  (1894.2  examples/sec;  0.068  sec/batch)
step 19800, loss=0.67  (1857.2  examples/sec;  0.069  sec/batch)
step 19810, loss=0.73  (1987.0  examples/sec;  0.064  sec/batch)
step 19820, loss=0.69  (1719.6  examples/sec;  0.074  sec/batch)
step 19830, loss=0.62  (1932.7  examples/sec;  0.066  sec/batch)
step 19840, loss=0.92  (1854.2  examples/sec;  0.069  sec/batch)
step 19850, loss=0.83  (1860.1  examples/sec;  0.069  sec/batch)
step 19860, loss=0.79  (1880.4  examples/sec;  0.068  sec/batch)
step 19870, loss=0.80  (1851.5  examples/sec;  0.069  sec/batch)
step 19880, loss=0.79  (1890.3  examples/sec;  0.068  sec/batch)
step 19890, loss=0.71  (1916.9  examples/sec;  0.067  sec/batch)
step 19900, loss=0.84  (1937.1  examples/sec;  0.066  sec/batch)
step 19910, loss=0.75  (1962.1  examples/sec;  0.065  sec/batch)
step 19920, loss=0.79  (1872.6  examples/sec;  0.068  sec/batch)
step 19930, loss=0.99  (1875.9  examples/sec;  0.068  sec/batch)
step 19940, loss=0.66  (1714.2  examples/sec;  0.075  sec/batch)
step 19950, loss=0.81  (1896.3  examples/sec;  0.068  sec/batch)
step 19960, loss=0.73  (1754.8  examples/sec;  0.073  sec/batch)
step 19970, loss=0.65  (1817.8  examples/sec;  0.070  sec/batch)
step 19980, loss=0.78  (1905.0  examples/sec;  0.067  sec/batch)
step 19990, loss=0.86  (1887.0  examples/sec;  0.068  sec/batch)
step 20000, loss=0.61  (1883.5  examples/sec;  0.068  sec/batch)
step 20010, loss=0.67  (1911.7  examples/sec;  0.067  sec/batch)
step 20020, loss=0.71  (1935.9  examples/sec;  0.066  sec/batch)
step 20030, loss=0.73  (1925.4  examples/sec;  0.066  sec/batch)
step 20040, loss=0.98  (1935.4  examples/sec;  0.066  sec/batch)
step 20050, loss=0.71  (1892.8  examples/sec;  0.068  sec/batch)
step 20060, loss=0.81  (1920.4  examples/sec;  0.067  sec/batch)
step 20070, loss=0.68  (1839.6  examples/sec;  0.070  sec/batch)
step 20080, loss=0.70  (1892.7  examples/sec;  0.068  sec/batch)
step 20090, loss=0.77  (1806.5  examples/sec;  0.071  sec/batch)
step 20100, loss=0.86  (2038.4  examples/sec;  0.063  sec/batch)
step 20110, loss=0.83  (1916.6  examples/sec;  0.067  sec/batch)
step 20120, loss=0.84  (1788.5  examples/sec;  0.072  sec/batch)
step 20130, loss=0.91  (1885.4  examples/sec;  0.068  sec/batch)
step 20140, loss=0.72  (1866.4  examples/sec;  0.069  sec/batch)
step 20150, loss=0.68  (1854.7  examples/sec;  0.069  sec/batch)
step 20160, loss=0.88  (1855.3  examples/sec;  0.069  sec/batch)
step 20170, loss=0.73  (1890.5  examples/sec;  0.068  sec/batch)
step 20180, loss=0.62  (1867.8  examples/sec;  0.069  sec/batch)
step 20190, loss=0.68  (1863.6  examples/sec;  0.069  sec/batch)
step 20200, loss=0.74  (1875.1  examples/sec;  0.068  sec/batch)
step 20210, loss=0.83  (1865.6  examples/sec;  0.069  sec/batch)
step 20220, loss=0.65  (1898.9  examples/sec;  0.067  sec/batch)
step 20230, loss=0.76  (1903.9  examples/sec;  0.067  sec/batch)
step 20240, loss=0.92  (1902.0  examples/sec;  0.067  sec/batch)
step 20250, loss=0.55  (1835.2  examples/sec;  0.070  sec/batch)
step 20260, loss=0.64  (1943.0  examples/sec;  0.066  sec/batch)
step 20270, loss=0.76  (1808.9  examples/sec;  0.071  sec/batch)
step 20280, loss=0.70  (1889.3  examples/sec;  0.068  sec/batch)
step 20290, loss=0.75  (1686.1  examples/sec;  0.076  sec/batch)
step 20300, loss=0.60  (1895.3  examples/sec;  0.068  sec/batch)
step 20310, loss=0.79  (1926.9  examples/sec;  0.066  sec/batch)
step 20320, loss=0.69  (1916.4  examples/sec;  0.067  sec/batch)
step 20330, loss=0.64  (1849.3  examples/sec;  0.069  sec/batch)
step 20340, loss=0.75  (1935.6  examples/sec;  0.066  sec/batch)
step 20350, loss=0.78  (1868.8  examples/sec;  0.068  sec/batch)
step 20360, loss=0.81  (1913.0  examples/sec;  0.067  sec/batch)
step 20370, loss=0.79  (1987.4  examples/sec;  0.064  sec/batch)
step 20380, loss=0.61  (1876.3  examples/sec;  0.068  sec/batch)
step 20390, loss=0.67  (1957.6  examples/sec;  0.065  sec/batch)
step 20400, loss=0.71  (1914.2  examples/sec;  0.067  sec/batch)
step 20410, loss=0.74  (1939.3  examples/sec;  0.066  sec/batch)
step 20420, loss=0.75  (1946.4  examples/sec;  0.066  sec/batch)
step 20430, loss=0.67  (1822.2  examples/sec;  0.070  sec/batch)
step 20440, loss=0.64  (1914.2  examples/sec;  0.067  sec/batch)
step 20450, loss=0.67  (1852.0  examples/sec;  0.069  sec/batch)
step 20460, loss=0.62  (1979.6  examples/sec;  0.065  sec/batch)
step 20470, loss=0.76  (1872.1  examples/sec;  0.068  sec/batch)
step 20480, loss=0.68  (1949.0  examples/sec;  0.066  sec/batch)
step 20490, loss=0.82  (1938.5  examples/sec;  0.066  sec/batch)
step 20500, loss=0.70  (1944.8  examples/sec;  0.066  sec/batch)
step 20510, loss=0.83  (1864.5  examples/sec;  0.069  sec/batch)
step 20520, loss=0.62  (1932.6  examples/sec;  0.066  sec/batch)
step 20530, loss=0.61  (1882.6  examples/sec;  0.068  sec/batch)
step 20540, loss=0.67  (1909.2  examples/sec;  0.067  sec/batch)
step 20550, loss=0.76  (1916.2  examples/sec;  0.067  sec/batch)
step 20560, loss=0.62  (1888.2  examples/sec;  0.068  sec/batch)
step 20570, loss=0.83  (1892.3  examples/sec;  0.068  sec/batch)
step 20580, loss=0.67  (1849.0  examples/sec;  0.069  sec/batch)
step 20590, loss=0.82  (1895.1  examples/sec;  0.068  sec/batch)
step 20600, loss=0.63  (1818.4  examples/sec;  0.070  sec/batch)
step 20610, loss=0.77  (1830.6  examples/sec;  0.070  sec/batch)
step 20620, loss=0.79  (1880.0  examples/sec;  0.068  sec/batch)
step 20630, loss=0.74  (1915.0  examples/sec;  0.067  sec/batch)
step 20640, loss=0.74  (1801.5  examples/sec;  0.071  sec/batch)
step 20650, loss=0.87  (1889.9  examples/sec;  0.068  sec/batch)
step 20660, loss=0.80  (1845.9  examples/sec;  0.069  sec/batch)
step 20670, loss=0.67  (1921.6  examples/sec;  0.067  sec/batch)
step 20680, loss=0.78  (1880.9  examples/sec;  0.068  sec/batch)
step 20690, loss=0.77  (1953.7  examples/sec;  0.066  sec/batch)
step 20700, loss=0.82  (1900.9  examples/sec;  0.067  sec/batch)
step 20710, loss=0.82  (1896.2  examples/sec;  0.068  sec/batch)
step 20720, loss=0.89  (1922.4  examples/sec;  0.067  sec/batch)
step 20730, loss=0.80  (1871.2  examples/sec;  0.068  sec/batch)
step 20740, loss=0.77  (1850.6  examples/sec;  0.069  sec/batch)
step 20750, loss=0.82  (1990.9  examples/sec;  0.064  sec/batch)
step 20760, loss=0.66  (1956.1  examples/sec;  0.065  sec/batch)
step 20770, loss=0.73  (1896.5  examples/sec;  0.067  sec/batch)
step 20780, loss=0.82  (1977.5  examples/sec;  0.065  sec/batch)
step 20790, loss=0.82  (1917.2  examples/sec;  0.067  sec/batch)
step 20800, loss=0.69  (1985.5  examples/sec;  0.064  sec/batch)
step 20810, loss=0.84  (1911.6  examples/sec;  0.067  sec/batch)
step 20820, loss=0.69  (1888.8  examples/sec;  0.068  sec/batch)
step 20830, loss=0.75  (1667.2  examples/sec;  0.077  sec/batch)
step 20840, loss=0.69  (1855.3  examples/sec;  0.069  sec/batch)
step 20850, loss=0.81  (1961.8  examples/sec;  0.065  sec/batch)
step 20860, loss=0.80  (1872.7  examples/sec;  0.068  sec/batch)
step 20870, loss=0.79  (1906.8  examples/sec;  0.067  sec/batch)
step 20880, loss=0.65  (1826.1  examples/sec;  0.070  sec/batch)
step 20890, loss=0.80  (1857.1  examples/sec;  0.069  sec/batch)
step 20900, loss=0.61  (1863.6  examples/sec;  0.069  sec/batch)
step 20910, loss=0.94  (2061.1  examples/sec;  0.062  sec/batch)
step 20920, loss=0.68  (1891.7  examples/sec;  0.068  sec/batch)
step 20930, loss=0.67  (1637.6  examples/sec;  0.078  sec/batch)
step 20940, loss=0.71  (1962.9  examples/sec;  0.065  sec/batch)
step 20950, loss=0.62  (1843.8  examples/sec;  0.069  sec/batch)
step 20960, loss=0.70  (1885.5  examples/sec;  0.068  sec/batch)
step 20970, loss=0.89  (1941.1  examples/sec;  0.066  sec/batch)
step 20980, loss=0.73  (1863.3  examples/sec;  0.069  sec/batch)
step 20990, loss=0.90  (1893.0  examples/sec;  0.068  sec/batch)
step 21000, loss=0.61  (1739.0  examples/sec;  0.074  sec/batch)
step 21010, loss=0.76  (1895.3  examples/sec;  0.068  sec/batch)
step 21020, loss=0.69  (1898.1  examples/sec;  0.067  sec/batch)
step 21030, loss=0.72  (1689.5  examples/sec;  0.076  sec/batch)
step 21040, loss=0.75  (1870.6  examples/sec;  0.068  sec/batch)
step 21050, loss=0.81  (1872.7  examples/sec;  0.068  sec/batch)
step 21060, loss=0.75  (1894.8  examples/sec;  0.068  sec/batch)
step 21070, loss=0.72  (1647.8  examples/sec;  0.078  sec/batch)
step 21080, loss=0.66  (1873.7  examples/sec;  0.068  sec/batch)
step 21090, loss=0.60  (1943.7  examples/sec;  0.066  sec/batch)
step 21100, loss=0.82  (1769.4  examples/sec;  0.072  sec/batch)
step 21110, loss=0.67  (1932.1  examples/sec;  0.066  sec/batch)
step 21120, loss=0.79  (1855.0  examples/sec;  0.069  sec/batch)
step 21130, loss=0.91  (1912.7  examples/sec;  0.067  sec/batch)
step 21140, loss=0.70  (1938.6  examples/sec;  0.066  sec/batch)
step 21150, loss=0.66  (1827.3  examples/sec;  0.070  sec/batch)
step 21160, loss=0.88  (1897.1  examples/sec;  0.067  sec/batch)
step 21170, loss=0.74  (1894.5  examples/sec;  0.068  sec/batch)
step 21180, loss=0.77  (1747.2  examples/sec;  0.073  sec/batch)
step 21190, loss=0.68  (1783.0  examples/sec;  0.072  sec/batch)
step 21200, loss=0.82  (1887.4  examples/sec;  0.068  sec/batch)
step 21210, loss=0.89  (1847.1  examples/sec;  0.069  sec/batch)
step 21220, loss=0.73  (1866.3  examples/sec;  0.069  sec/batch)
step 21230, loss=0.73  (1848.6  examples/sec;  0.069  sec/batch)
step 21240, loss=0.70  (1943.3  examples/sec;  0.066  sec/batch)
step 21250, loss=0.72  (1974.5  examples/sec;  0.065  sec/batch)
step 21260, loss=0.71  (1959.9  examples/sec;  0.065  sec/batch)
step 21270, loss=0.59  (1920.6  examples/sec;  0.067  sec/batch)
step 21280, loss=0.70  (1911.5  examples/sec;  0.067  sec/batch)
step 21290, loss=0.84  (1931.4  examples/sec;  0.066  sec/batch)
step 21300, loss=0.75  (1877.9  examples/sec;  0.068  sec/batch)
step 21310, loss=0.78  (1855.7  examples/sec;  0.069  sec/batch)
step 21320, loss=0.72  (1891.5  examples/sec;  0.068  sec/batch)
step 21330, loss=0.73  (1833.8  examples/sec;  0.070  sec/batch)
step 21340, loss=0.74  (1947.3  examples/sec;  0.066  sec/batch)
step 21350, loss=0.71  (1905.3  examples/sec;  0.067  sec/batch)
step 21360, loss=0.67  (1765.0  examples/sec;  0.073  sec/batch)
step 21370, loss=0.55  (1901.2  examples/sec;  0.067  sec/batch)
step 21380, loss=0.90  (1912.6  examples/sec;  0.067  sec/batch)
step 21390, loss=0.63  (1870.3  examples/sec;  0.068  sec/batch)
step 21400, loss=0.78  (1909.6  examples/sec;  0.067  sec/batch)
step 21410, loss=0.73  (1908.8  examples/sec;  0.067  sec/batch)
step 21420, loss=0.88  (1884.8  examples/sec;  0.068  sec/batch)
step 21430, loss=0.69  (1943.0  examples/sec;  0.066  sec/batch)
step 21440, loss=0.75  (1875.6  examples/sec;  0.068  sec/batch)
step 21450, loss=0.90  (1936.0  examples/sec;  0.066  sec/batch)
step 21460, loss=0.70  (1901.8  examples/sec;  0.067  sec/batch)
step 21470, loss=0.62  (1848.5  examples/sec;  0.069  sec/batch)
step 21480, loss=0.72  (1855.2  examples/sec;  0.069  sec/batch)
step 21490, loss=0.83  (1900.0  examples/sec;  0.067  sec/batch)
step 21500, loss=0.59  (1907.7  examples/sec;  0.067  sec/batch)
step 21510, loss=0.67  (1871.7  examples/sec;  0.068  sec/batch)
step 21520, loss=0.78  (1934.0  examples/sec;  0.066  sec/batch)
step 21530, loss=0.80  (1912.1  examples/sec;  0.067  sec/batch)
step 21540, loss=0.77  (1978.0  examples/sec;  0.065  sec/batch)
step 21550, loss=0.72  (1912.0  examples/sec;  0.067  sec/batch)
step 21560, loss=0.72  (1747.9  examples/sec;  0.073  sec/batch)
step 21570, loss=0.59  (1897.7  examples/sec;  0.067  sec/batch)
step 21580, loss=0.65  (1937.4  examples/sec;  0.066  sec/batch)
step 21590, loss=0.81  (1888.9  examples/sec;  0.068  sec/batch)
step 21600, loss=0.66  (1867.1  examples/sec;  0.069  sec/batch)
step 21610, loss=0.77  (1892.5  examples/sec;  0.068  sec/batch)
step 21620, loss=0.84  (1831.1  examples/sec;  0.070  sec/batch)
step 21630, loss=0.74  (1932.9  examples/sec;  0.066  sec/batch)
step 21640, loss=0.71  (1781.1  examples/sec;  0.072  sec/batch)
step 21650, loss=0.71  (1884.2  examples/sec;  0.068  sec/batch)
step 21660, loss=0.84  (1864.3  examples/sec;  0.069  sec/batch)
step 21670, loss=0.76  (1907.8  examples/sec;  0.067  sec/batch)
step 21680, loss=0.81  (1905.1  examples/sec;  0.067  sec/batch)
step 21690, loss=0.69  (1939.6  examples/sec;  0.066  sec/batch)
step 21700, loss=0.69  (1898.2  examples/sec;  0.067  sec/batch)
step 21710, loss=0.66  (1927.8  examples/sec;  0.066  sec/batch)
step 21720, loss=0.60  (1898.3  examples/sec;  0.067  sec/batch)
step 21730, loss=0.77  (1884.9  examples/sec;  0.068  sec/batch)
step 21740, loss=0.63  (1902.6  examples/sec;  0.067  sec/batch)
step 21750, loss=0.61  (1999.9  examples/sec;  0.064  sec/batch)
step 21760, loss=0.55  (1922.4  examples/sec;  0.067  sec/batch)
step 21770, loss=0.65  (1928.7  examples/sec;  0.066  sec/batch)
step 21780, loss=0.93  (1901.1  examples/sec;  0.067  sec/batch)
step 21790, loss=0.86  (1862.7  examples/sec;  0.069  sec/batch)
step 21800, loss=0.67  (1928.9  examples/sec;  0.066  sec/batch)
step 21810, loss=0.73  (1892.0  examples/sec;  0.068  sec/batch)
step 21820, loss=0.75  (1904.0  examples/sec;  0.067  sec/batch)
step 21830, loss=0.77  (1937.1  examples/sec;  0.066  sec/batch)
step 21840, loss=0.72  (1833.0  examples/sec;  0.070  sec/batch)
step 21850, loss=0.76  (1839.6  examples/sec;  0.070  sec/batch)
step 21860, loss=0.78  (2019.7  examples/sec;  0.063  sec/batch)
step 21870, loss=0.63  (1845.6  examples/sec;  0.069  sec/batch)
step 21880, loss=0.72  (1914.9  examples/sec;  0.067  sec/batch)
step 21890, loss=0.87  (1965.9  examples/sec;  0.065  sec/batch)
step 21900, loss=0.62  (1852.6  examples/sec;  0.069  sec/batch)
step 21910, loss=0.71  (1832.6  examples/sec;  0.070  sec/batch)
step 21920, loss=0.94  (1961.5  examples/sec;  0.065  sec/batch)
step 21930, loss=0.75  (1970.2  examples/sec;  0.065  sec/batch)
step 21940, loss=0.71  (1920.5  examples/sec;  0.067  sec/batch)
step 21950, loss=0.77  (1913.6  examples/sec;  0.067  sec/batch)
step 21960, loss=0.62  (1835.8  examples/sec;  0.070  sec/batch)
step 21970, loss=0.60  (1888.5  examples/sec;  0.068  sec/batch)
step 21980, loss=0.75  (1855.5  examples/sec;  0.069  sec/batch)
step 21990, loss=0.72  (1984.0  examples/sec;  0.065  sec/batch)
step 22000, loss=0.71  (1834.7  examples/sec;  0.070  sec/batch)
step 22010, loss=0.64  (1943.2  examples/sec;  0.066  sec/batch)
step 22020, loss=0.75  (1650.7  examples/sec;  0.078  sec/batch)
step 22030, loss=0.71  (1899.2  examples/sec;  0.067  sec/batch)
step 22040, loss=0.89  (1786.7  examples/sec;  0.072  sec/batch)
step 22050, loss=0.79  (1884.5  examples/sec;  0.068  sec/batch)
step 22060, loss=0.70  (1823.8  examples/sec;  0.070  sec/batch)
step 22070, loss=0.73  (1951.8  examples/sec;  0.066  sec/batch)
step 22080, loss=0.73  (1949.4  examples/sec;  0.066  sec/batch)
step 22090, loss=0.64  (1940.5  examples/sec;  0.066  sec/batch)
step 22100, loss=0.72  (1883.9  examples/sec;  0.068  sec/batch)
step 22110, loss=0.71  (1875.0  examples/sec;  0.068  sec/batch)
step 22120, loss=0.76  (1871.2  examples/sec;  0.068  sec/batch)
step 22130, loss=0.77  (1887.5  examples/sec;  0.068  sec/batch)
step 22140, loss=0.86  (1843.2  examples/sec;  0.069  sec/batch)
step 22150, loss=0.98  (1910.8  examples/sec;  0.067  sec/batch)
step 22160, loss=0.72  (1772.4  examples/sec;  0.072  sec/batch)
step 22170, loss=0.73  (1857.5  examples/sec;  0.069  sec/batch)
step 22180, loss=0.79  (1921.3  examples/sec;  0.067  sec/batch)
step 22190, loss=0.64  (1829.3  examples/sec;  0.070  sec/batch)
step 22200, loss=0.85  (1925.5  examples/sec;  0.066  sec/batch)
step 22210, loss=0.89  (1908.1  examples/sec;  0.067  sec/batch)
step 22220, loss=0.69  (1900.9  examples/sec;  0.067  sec/batch)
step 22230, loss=0.68  (1618.3  examples/sec;  0.079  sec/batch)
step 22240, loss=0.86  (1912.0  examples/sec;  0.067  sec/batch)
step 22250, loss=0.79  (1880.9  examples/sec;  0.068  sec/batch)
step 22260, loss=0.76  (1892.8  examples/sec;  0.068  sec/batch)
step 22270, loss=0.89  (1889.1  examples/sec;  0.068  sec/batch)
step 22280, loss=0.73  (1882.3  examples/sec;  0.068  sec/batch)
step 22290, loss=0.53  (1963.1  examples/sec;  0.065  sec/batch)
step 22300, loss=0.71  (1895.6  examples/sec;  0.068  sec/batch)
step 22310, loss=0.78  (1824.9  examples/sec;  0.070  sec/batch)
step 22320, loss=0.71  (1630.6  examples/sec;  0.079  sec/batch)
step 22330, loss=0.67  (1926.0  examples/sec;  0.066  sec/batch)
step 22340, loss=0.78  (1842.8  examples/sec;  0.069  sec/batch)
step 22350, loss=0.84  (1888.0  examples/sec;  0.068  sec/batch)
step 22360, loss=0.86  (1799.9  examples/sec;  0.071  sec/batch)
step 22370, loss=0.80  (1841.6  examples/sec;  0.070  sec/batch)
step 22380, loss=0.71  (1816.5  examples/sec;  0.070  sec/batch)
step 22390, loss=0.65  (1919.6  examples/sec;  0.067  sec/batch)
step 22400, loss=0.73  (1880.6  examples/sec;  0.068  sec/batch)
step 22410, loss=0.63  (1823.2  examples/sec;  0.070  sec/batch)
step 22420, loss=0.80  (1925.0  examples/sec;  0.066  sec/batch)
step 22430, loss=0.94  (1970.1  examples/sec;  0.065  sec/batch)
step 22440, loss=0.69  (1897.0  examples/sec;  0.067  sec/batch)
step 22450, loss=0.67  (1964.4  examples/sec;  0.065  sec/batch)
step 22460, loss=0.68  (1893.5  examples/sec;  0.068  sec/batch)
step 22470, loss=0.79  (1959.6  examples/sec;  0.065  sec/batch)
step 22480, loss=0.56  (1953.1  examples/sec;  0.066  sec/batch)
step 22490, loss=0.76  (1907.4  examples/sec;  0.067  sec/batch)
step 22500, loss=0.73  (1899.1  examples/sec;  0.067  sec/batch)
step 22510, loss=0.89  (1820.3  examples/sec;  0.070  sec/batch)
step 22520, loss=0.81  (1957.7  examples/sec;  0.065  sec/batch)
step 22530, loss=0.67  (1830.1  examples/sec;  0.070  sec/batch)
step 22540, loss=0.83  (1895.1  examples/sec;  0.068  sec/batch)
step 22550, loss=0.72  (1889.5  examples/sec;  0.068  sec/batch)
step 22560, loss=0.77  (1906.0  examples/sec;  0.067  sec/batch)
step 22570, loss=0.80  (1908.7  examples/sec;  0.067  sec/batch)
step 22580, loss=0.84  (1868.6  examples/sec;  0.069  sec/batch)
step 22590, loss=0.63  (1886.7  examples/sec;  0.068  sec/batch)
step 22600, loss=0.92  (1881.8  examples/sec;  0.068  sec/batch)
step 22610, loss=0.72  (1658.3  examples/sec;  0.077  sec/batch)
step 22620, loss=0.57  (1935.4  examples/sec;  0.066  sec/batch)
step 22630, loss=0.70  (1841.4  examples/sec;  0.070  sec/batch)
step 22640, loss=0.68  (1845.9  examples/sec;  0.069  sec/batch)
step 22650, loss=0.61  (1942.3  examples/sec;  0.066  sec/batch)
step 22660, loss=0.80  (1961.7  examples/sec;  0.065  sec/batch)
step 22670, loss=0.57  (1897.7  examples/sec;  0.067  sec/batch)
step 22680, loss=0.59  (1910.0  examples/sec;  0.067  sec/batch)
step 22690, loss=0.77  (1913.3  examples/sec;  0.067  sec/batch)
step 22700, loss=0.87  (1921.3  examples/sec;  0.067  sec/batch)
step 22710, loss=0.87  (1856.6  examples/sec;  0.069  sec/batch)
step 22720, loss=0.76  (1938.3  examples/sec;  0.066  sec/batch)
step 22730, loss=0.78  (1846.6  examples/sec;  0.069  sec/batch)
step 22740, loss=0.74  (1980.0  examples/sec;  0.065  sec/batch)
step 22750, loss=0.77  (1908.1  examples/sec;  0.067  sec/batch)
step 22760, loss=0.73  (1976.7  examples/sec;  0.065  sec/batch)
step 22770, loss=0.78  (1873.2  examples/sec;  0.068  sec/batch)
step 22780, loss=0.74  (1839.2  examples/sec;  0.070  sec/batch)
step 22790, loss=0.72  (1881.7  examples/sec;  0.068  sec/batch)
step 22800, loss=0.83  (1942.2  examples/sec;  0.066  sec/batch)
step 22810, loss=0.74  (1899.8  examples/sec;  0.067  sec/batch)
step 22820, loss=0.67  (1716.1  examples/sec;  0.075  sec/batch)
step 22830, loss=0.76  (1886.4  examples/sec;  0.068  sec/batch)
step 22840, loss=0.77  (1927.7  examples/sec;  0.066  sec/batch)
step 22850, loss=0.73  (1964.3  examples/sec;  0.065  sec/batch)
step 22860, loss=0.67  (1919.4  examples/sec;  0.067  sec/batch)
step 22870, loss=0.73  (2003.0  examples/sec;  0.064  sec/batch)
step 22880, loss=0.72  (1911.3  examples/sec;  0.067  sec/batch)
step 22890, loss=0.67  (1884.1  examples/sec;  0.068  sec/batch)
step 22900, loss=0.80  (1855.9  examples/sec;  0.069  sec/batch)
step 22910, loss=0.74  (1966.3  examples/sec;  0.065  sec/batch)
step 22920, loss=0.71  (1964.6  examples/sec;  0.065  sec/batch)
step 22930, loss=0.74  (1970.8  examples/sec;  0.065  sec/batch)
step 22940, loss=0.73  (1892.0  examples/sec;  0.068  sec/batch)
step 22950, loss=0.54  (1989.1  examples/sec;  0.064  sec/batch)
step 22960, loss=0.66  (1910.4  examples/sec;  0.067  sec/batch)
step 22970, loss=0.80  (1870.7  examples/sec;  0.068  sec/batch)
step 22980, loss=0.67  (1927.1  examples/sec;  0.066  sec/batch)
step 22990, loss=0.64  (1907.9  examples/sec;  0.067  sec/batch)
step 23000, loss=1.06  (1820.9  examples/sec;  0.070  sec/batch)
step 23010, loss=0.59  (1817.6  examples/sec;  0.070  sec/batch)
step 23020, loss=0.85  (1875.1  examples/sec;  0.068  sec/batch)
step 23030, loss=0.82  (1894.3  examples/sec;  0.068  sec/batch)
step 23040, loss=0.81  (1882.9  examples/sec;  0.068  sec/batch)
step 23050, loss=0.68  (1890.1  examples/sec;  0.068  sec/batch)
step 23060, loss=0.65  (1943.3  examples/sec;  0.066  sec/batch)
step 23070, loss=0.67  (1703.1  examples/sec;  0.075  sec/batch)
step 23080, loss=0.76  (1953.2  examples/sec;  0.066  sec/batch)
step 23090, loss=0.72  (1909.9  examples/sec;  0.067  sec/batch)
step 23100, loss=0.68  (1865.7  examples/sec;  0.069  sec/batch)
step 23110, loss=0.77  (1932.8  examples/sec;  0.066  sec/batch)
step 23120, loss=0.82  (1921.0  examples/sec;  0.067  sec/batch)
step 23130, loss=0.65  (1933.5  examples/sec;  0.066  sec/batch)
step 23140, loss=0.75  (1868.8  examples/sec;  0.068  sec/batch)
step 23150, loss=0.76  (1946.2  examples/sec;  0.066  sec/batch)
step 23160, loss=0.65  (1857.2  examples/sec;  0.069  sec/batch)
step 23170, loss=0.73  (1849.0  examples/sec;  0.069  sec/batch)
step 23180, loss=0.87  (1963.5  examples/sec;  0.065  sec/batch)
step 23190, loss=0.65  (1844.9  examples/sec;  0.069  sec/batch)
step 23200, loss=0.82  (1834.8  examples/sec;  0.070  sec/batch)
step 23210, loss=0.78  (2018.3  examples/sec;  0.063  sec/batch)
step 23220, loss=0.61  (1873.4  examples/sec;  0.068  sec/batch)
step 23230, loss=0.73  (1924.3  examples/sec;  0.067  sec/batch)
step 23240, loss=0.90  (1868.1  examples/sec;  0.069  sec/batch)
step 23250, loss=0.69  (1942.3  examples/sec;  0.066  sec/batch)
step 23260, loss=0.78  (1839.7  examples/sec;  0.070  sec/batch)
step 23270, loss=0.71  (1961.7  examples/sec;  0.065  sec/batch)
step 23280, loss=0.76  (1929.2  examples/sec;  0.066  sec/batch)
step 23290, loss=0.75  (1942.2  examples/sec;  0.066  sec/batch)
step 23300, loss=0.67  (1884.0  examples/sec;  0.068  sec/batch)
step 23310, loss=0.73  (1925.6  examples/sec;  0.066  sec/batch)
step 23320, loss=0.77  (1958.6  examples/sec;  0.065  sec/batch)
step 23330, loss=0.70  (1807.1  examples/sec;  0.071  sec/batch)
step 23340, loss=0.59  (1939.1  examples/sec;  0.066  sec/batch)
step 23350, loss=0.70  (1880.7  examples/sec;  0.068  sec/batch)
step 23360, loss=0.78  (1870.3  examples/sec;  0.068  sec/batch)
step 23370, loss=0.62  (1864.3  examples/sec;  0.069  sec/batch)
step 23380, loss=0.77  (1823.9  examples/sec;  0.070  sec/batch)
step 23390, loss=0.69  (1885.2  examples/sec;  0.068  sec/batch)
step 23400, loss=0.69  (1875.7  examples/sec;  0.068  sec/batch)
step 23410, loss=0.73  (1974.9  examples/sec;  0.065  sec/batch)
step 23420, loss=0.58  (1880.4  examples/sec;  0.068  sec/batch)
step 23430, loss=0.72  (1953.3  examples/sec;  0.066  sec/batch)
step 23440, loss=0.74  (1917.1  examples/sec;  0.067  sec/batch)
step 23450, loss=0.59  (1859.9  examples/sec;  0.069  sec/batch)
step 23460, loss=0.88  (1912.5  examples/sec;  0.067  sec/batch)
step 23470, loss=0.69  (1976.0  examples/sec;  0.065  sec/batch)
step 23480, loss=0.66  (1951.3  examples/sec;  0.066  sec/batch)
step 23490, loss=1.00  (1919.7  examples/sec;  0.067  sec/batch)
step 23500, loss=0.78  (1864.0  examples/sec;  0.069  sec/batch)
step 23510, loss=0.73  (1896.4  examples/sec;  0.067  sec/batch)
step 23520, loss=0.82  (1796.3  examples/sec;  0.071  sec/batch)
step 23530, loss=0.68  (1849.5  examples/sec;  0.069  sec/batch)
step 23540, loss=0.76  (1921.1  examples/sec;  0.067  sec/batch)
step 23550, loss=0.67  (1817.4  examples/sec;  0.070  sec/batch)
step 23560, loss=0.66  (1920.9  examples/sec;  0.067  sec/batch)
step 23570, loss=0.74  (1893.5  examples/sec;  0.068  sec/batch)
step 23580, loss=0.65  (1919.6  examples/sec;  0.067  sec/batch)
step 23590, loss=0.65  (1899.7  examples/sec;  0.067  sec/batch)
step 23600, loss=0.65  (1950.6  examples/sec;  0.066  sec/batch)
step 23610, loss=0.81  (1828.1  examples/sec;  0.070  sec/batch)
step 23620, loss=0.78  (1582.9  examples/sec;  0.081  sec/batch)
step 23630, loss=0.51  (1928.5  examples/sec;  0.066  sec/batch)
step 23640, loss=0.62  (1901.6  examples/sec;  0.067  sec/batch)
step 23650, loss=0.71  (1812.7  examples/sec;  0.071  sec/batch)
step 23660, loss=0.82  (1937.0  examples/sec;  0.066  sec/batch)
step 23670, loss=0.67  (1853.0  examples/sec;  0.069  sec/batch)
step 23680, loss=1.01  (1890.0  examples/sec;  0.068  sec/batch)
step 23690, loss=0.81  (2021.2  examples/sec;  0.063  sec/batch)
step 23700, loss=0.72  (1721.8  examples/sec;  0.074  sec/batch)
step 23710, loss=0.73  (1787.4  examples/sec;  0.072  sec/batch)
step 23720, loss=0.62  (1918.2  examples/sec;  0.067  sec/batch)
step 23730, loss=0.81  (1872.0  examples/sec;  0.068  sec/batch)
step 23740, loss=0.89  (1894.6  examples/sec;  0.068  sec/batch)
step 23750, loss=0.63  (1873.8  examples/sec;  0.068  sec/batch)
step 23760, loss=0.74  (1926.4  examples/sec;  0.066  sec/batch)
step 23770, loss=0.97  (2031.8  examples/sec;  0.063  sec/batch)
step 23780, loss=0.74  (1895.6  examples/sec;  0.068  sec/batch)
step 23790, loss=0.63  (1970.6  examples/sec;  0.065  sec/batch)
step 23800, loss=0.74  (1845.3  examples/sec;  0.069  sec/batch)
step 23810, loss=0.78  (1896.7  examples/sec;  0.067  sec/batch)
step 23820, loss=0.70  (1945.6  examples/sec;  0.066  sec/batch)
step 23830, loss=0.74  (1928.7  examples/sec;  0.066  sec/batch)
step 23840, loss=0.69  (1908.4  examples/sec;  0.067  sec/batch)
step 23850, loss=0.75  (1829.8  examples/sec;  0.070  sec/batch)
step 23860, loss=0.78  (1975.1  examples/sec;  0.065  sec/batch)
step 23870, loss=0.74  (1972.4  examples/sec;  0.065  sec/batch)
step 23880, loss=0.54  (1859.5  examples/sec;  0.069  sec/batch)
step 23890, loss=0.86  (1975.5  examples/sec;  0.065  sec/batch)
step 23900, loss=0.83  (1898.2  examples/sec;  0.067  sec/batch)
step 23910, loss=0.88  (1886.6  examples/sec;  0.068  sec/batch)
step 23920, loss=0.82  (1656.2  examples/sec;  0.077  sec/batch)
step 23930, loss=0.77  (1660.8  examples/sec;  0.077  sec/batch)
step 23940, loss=0.74  (1941.6  examples/sec;  0.066  sec/batch)
step 23950, loss=0.97  (1956.3  examples/sec;  0.065  sec/batch)
step 23960, loss=0.81  (1889.0  examples/sec;  0.068  sec/batch)
step 23970, loss=0.69  (1900.7  examples/sec;  0.067  sec/batch)
step 23980, loss=0.90  (1836.3  examples/sec;  0.070  sec/batch)
step 23990, loss=0.64  (1918.5  examples/sec;  0.067  sec/batch)
step 24000, loss=0.71  (1880.9  examples/sec;  0.068  sec/batch)
step 24010, loss=0.82  (1826.7  examples/sec;  0.070  sec/batch)
step 24020, loss=0.64  (1916.2  examples/sec;  0.067  sec/batch)
step 24030, loss=0.79  (2003.6  examples/sec;  0.064  sec/batch)
step 24040, loss=0.70  (1850.1  examples/sec;  0.069  sec/batch)
step 24050, loss=0.77  (1821.0  examples/sec;  0.070  sec/batch)
step 24060, loss=0.75  (1968.9  examples/sec;  0.065  sec/batch)
step 24070, loss=0.66  (1874.3  examples/sec;  0.068  sec/batch)
step 24080, loss=0.64  (1872.3  examples/sec;  0.068  sec/batch)
step 24090, loss=0.75  (1683.1  examples/sec;  0.076  sec/batch)
step 24100, loss=0.74  (1868.7  examples/sec;  0.068  sec/batch)
step 24110, loss=0.90  (1915.9  examples/sec;  0.067  sec/batch)
step 24120, loss=0.73  (2021.1  examples/sec;  0.063  sec/batch)
step 24130, loss=0.76  (1792.8  examples/sec;  0.071  sec/batch)
step 24140, loss=0.79  (1923.9  examples/sec;  0.067  sec/batch)
step 24150, loss=0.91  (1949.8  examples/sec;  0.066  sec/batch)
step 24160, loss=0.75  (1910.1  examples/sec;  0.067  sec/batch)
step 24170, loss=0.69  (1891.8  examples/sec;  0.068  sec/batch)
step 24180, loss=0.79  (1852.3  examples/sec;  0.069  sec/batch)
step 24190, loss=0.60  (1852.6  examples/sec;  0.069  sec/batch)
step 24200, loss=0.80  (1991.8  examples/sec;  0.064  sec/batch)
step 24210, loss=0.69  (1962.0  examples/sec;  0.065  sec/batch)
step 24220, loss=0.67  (1948.9  examples/sec;  0.066  sec/batch)
step 24230, loss=0.76  (1784.4  examples/sec;  0.072  sec/batch)
step 24240, loss=0.51  (1887.0  examples/sec;  0.068  sec/batch)
step 24250, loss=0.68  (1855.6  examples/sec;  0.069  sec/batch)
step 24260, loss=0.91  (1823.8  examples/sec;  0.070  sec/batch)
step 24270, loss=0.56  (1847.2  examples/sec;  0.069  sec/batch)
step 24280, loss=0.73  (1885.5  examples/sec;  0.068  sec/batch)
step 24290, loss=0.78  (1951.6  examples/sec;  0.066  sec/batch)
step 24300, loss=0.67  (1917.6  examples/sec;  0.067  sec/batch)
step 24310, loss=0.66  (1906.8  examples/sec;  0.067  sec/batch)
step 24320, loss=0.78  (1904.7  examples/sec;  0.067  sec/batch)
step 24330, loss=0.66  (1829.4  examples/sec;  0.070  sec/batch)
step 24340, loss=0.65  (1985.9  examples/sec;  0.064  sec/batch)
step 24350, loss=0.67  (1704.8  examples/sec;  0.075  sec/batch)
step 24360, loss=0.62  (1831.6  examples/sec;  0.070  sec/batch)
step 24370, loss=0.63  (1881.2  examples/sec;  0.068  sec/batch)
step 24380, loss=0.59  (1925.8  examples/sec;  0.066  sec/batch)
step 24390, loss=0.81  (1909.5  examples/sec;  0.067  sec/batch)
step 24400, loss=0.71  (1872.7  examples/sec;  0.068  sec/batch)
step 24410, loss=0.71  (1882.8  examples/sec;  0.068  sec/batch)
step 24420, loss=0.88  (1887.1  examples/sec;  0.068  sec/batch)
step 24430, loss=0.92  (2021.0  examples/sec;  0.063  sec/batch)
step 24440, loss=0.79  (1916.6  examples/sec;  0.067  sec/batch)
step 24450, loss=0.80  (1845.2  examples/sec;  0.069  sec/batch)
step 24460, loss=0.70  (1850.9  examples/sec;  0.069  sec/batch)
step 24470, loss=0.85  (2050.2  examples/sec;  0.062  sec/batch)
step 24480, loss=0.80  (1812.5  examples/sec;  0.071  sec/batch)
step 24490, loss=0.68  (1909.6  examples/sec;  0.067  sec/batch)
step 24500, loss=0.72  (1870.1  examples/sec;  0.068  sec/batch)
step 24510, loss=0.69  (1566.7  examples/sec;  0.082  sec/batch)
step 24520, loss=0.62  (1820.3  examples/sec;  0.070  sec/batch)
step 24530, loss=0.73  (1962.8  examples/sec;  0.065  sec/batch)
step 24540, loss=0.73  (1873.3  examples/sec;  0.068  sec/batch)
step 24550, loss=0.66  (1915.1  examples/sec;  0.067  sec/batch)
step 24560, loss=0.78  (1904.3  examples/sec;  0.067  sec/batch)
step 24570, loss=0.84  (1892.0  examples/sec;  0.068  sec/batch)
step 24580, loss=0.72  (1956.6  examples/sec;  0.065  sec/batch)
step 24590, loss=0.73  (1907.7  examples/sec;  0.067  sec/batch)
step 24600, loss=0.69  (1852.0  examples/sec;  0.069  sec/batch)
step 24610, loss=0.61  (1967.7  examples/sec;  0.065  sec/batch)
step 24620, loss=0.70  (1989.7  examples/sec;  0.064  sec/batch)
step 24630, loss=0.64  (1879.7  examples/sec;  0.068  sec/batch)
step 24640, loss=0.77  (1976.7  examples/sec;  0.065  sec/batch)
step 24650, loss=0.70  (1868.3  examples/sec;  0.069  sec/batch)
step 24660, loss=0.85  (1912.5  examples/sec;  0.067  sec/batch)
step 24670, loss=0.70  (1884.6  examples/sec;  0.068  sec/batch)
step 24680, loss=0.53  (1863.4  examples/sec;  0.069  sec/batch)
step 24690, loss=0.76  (1844.5  examples/sec;  0.069  sec/batch)
step 24700, loss=0.74  (1886.3  examples/sec;  0.068  sec/batch)
step 24710, loss=0.79  (1874.5  examples/sec;  0.068  sec/batch)
step 24720, loss=0.85  (1829.3  examples/sec;  0.070  sec/batch)
step 24730, loss=0.78  (1927.1  examples/sec;  0.066  sec/batch)
step 24740, loss=0.60  (1904.0  examples/sec;  0.067  sec/batch)
step 24750, loss=0.56  (1883.4  examples/sec;  0.068  sec/batch)
step 24760, loss=0.84  (1857.6  examples/sec;  0.069  sec/batch)
step 24770, loss=0.52  (1999.8  examples/sec;  0.064  sec/batch)
step 24780, loss=0.71  (2006.6  examples/sec;  0.064  sec/batch)
step 24790, loss=0.67  (1919.4  examples/sec;  0.067  sec/batch)
step 24800, loss=0.83  (1854.6  examples/sec;  0.069  sec/batch)
step 24810, loss=0.69  (1826.5  examples/sec;  0.070  sec/batch)
step 24820, loss=0.79  (1920.0  examples/sec;  0.067  sec/batch)
step 24830, loss=0.77  (1914.4  examples/sec;  0.067  sec/batch)
step 24840, loss=0.85  (1863.6  examples/sec;  0.069  sec/batch)
step 24850, loss=0.75  (1896.9  examples/sec;  0.067  sec/batch)
step 24860, loss=0.68  (1882.8  examples/sec;  0.068  sec/batch)
step 24870, loss=0.70  (1833.0  examples/sec;  0.070  sec/batch)
step 24880, loss=0.76  (1873.3  examples/sec;  0.068  sec/batch)
step 24890, loss=0.70  (1879.1  examples/sec;  0.068  sec/batch)
step 24900, loss=0.63  (1860.7  examples/sec;  0.069  sec/batch)
step 24910, loss=0.57  (1902.4  examples/sec;  0.067  sec/batch)
step 24920, loss=0.73  (1885.3  examples/sec;  0.068  sec/batch)
step 24930, loss=0.70  (1944.5  examples/sec;  0.066  sec/batch)
step 24940, loss=0.93  (1953.9  examples/sec;  0.066  sec/batch)
step 24950, loss=0.71  (1723.5  examples/sec;  0.074  sec/batch)
step 24960, loss=0.75  (1957.9  examples/sec;  0.065  sec/batch)
step 24970, loss=0.68  (1948.6  examples/sec;  0.066  sec/batch)
step 24980, loss=0.91  (1836.0  examples/sec;  0.070  sec/batch)
step 24990, loss=0.71  (1913.8  examples/sec;  0.067  sec/batch)
step 25000, loss=0.71  (1870.4  examples/sec;  0.068  sec/batch)
step 25010, loss=0.91  (1923.2  examples/sec;  0.067  sec/batch)
step 25020, loss=0.60  (1925.9  examples/sec;  0.066  sec/batch)
step 25030, loss=0.92  (1888.8  examples/sec;  0.068  sec/batch)
step 25040, loss=0.78  (1903.2  examples/sec;  0.067  sec/batch)
step 25050, loss=0.64  (1771.5  examples/sec;  0.072  sec/batch)
step 25060, loss=0.74  (1853.3  examples/sec;  0.069  sec/batch)
step 25070, loss=0.70  (1920.7  examples/sec;  0.067  sec/batch)
step 25080, loss=0.62  (1983.0  examples/sec;  0.065  sec/batch)
step 25090, loss=0.89  (1857.1  examples/sec;  0.069  sec/batch)
step 25100, loss=0.70  (1821.1  examples/sec;  0.070  sec/batch)
step 25110, loss=0.74  (1769.1  examples/sec;  0.072  sec/batch)
step 25120, loss=0.82  (1978.1  examples/sec;  0.065  sec/batch)
step 25130, loss=0.87  (1989.5  examples/sec;  0.064  sec/batch)
step 25140, loss=0.87  (1877.8  examples/sec;  0.068  sec/batch)
step 25150, loss=0.64  (1942.9  examples/sec;  0.066  sec/batch)
step 25160, loss=0.64  (1846.7  examples/sec;  0.069  sec/batch)
step 25170, loss=0.75  (1910.7  examples/sec;  0.067  sec/batch)
step 25180, loss=0.68  (1888.2  examples/sec;  0.068  sec/batch)
step 25190, loss=0.64  (1702.3  examples/sec;  0.075  sec/batch)
step 25200, loss=0.61  (1907.7  examples/sec;  0.067  sec/batch)
step 25210, loss=0.83  (1942.6  examples/sec;  0.066  sec/batch)
step 25220, loss=0.69  (1988.0  examples/sec;  0.064  sec/batch)
step 25230, loss=0.89  (1917.3  examples/sec;  0.067  sec/batch)
step 25240, loss=0.87  (1938.0  examples/sec;  0.066  sec/batch)
step 25250, loss=0.88  (1830.7  examples/sec;  0.070  sec/batch)
step 25260, loss=0.53  (1896.9  examples/sec;  0.067  sec/batch)
step 25270, loss=0.77  (1935.8  examples/sec;  0.066  sec/batch)
step 25280, loss=0.82  (1865.9  examples/sec;  0.069  sec/batch)
step 25290, loss=0.76  (1699.7  examples/sec;  0.075  sec/batch)
step 25300, loss=0.66  (1834.4  examples/sec;  0.070  sec/batch)
step 25310, loss=0.84  (1900.4  examples/sec;  0.067  sec/batch)
step 25320, loss=0.75  (1937.7  examples/sec;  0.066  sec/batch)
step 25330, loss=0.79  (1918.2  examples/sec;  0.067  sec/batch)
step 25340, loss=0.72  (1875.3  examples/sec;  0.068  sec/batch)
step 25350, loss=0.63  (1704.7  examples/sec;  0.075  sec/batch)
step 25360, loss=0.78  (1877.5  examples/sec;  0.068  sec/batch)
step 25370, loss=0.57  (1966.8  examples/sec;  0.065  sec/batch)
step 25380, loss=0.74  (1649.6  examples/sec;  0.078  sec/batch)
step 25390, loss=0.82  (1725.0  examples/sec;  0.074  sec/batch)
step 25400, loss=0.72  (1957.8  examples/sec;  0.065  sec/batch)
step 25410, loss=0.76  (1974.1  examples/sec;  0.065  sec/batch)
step 25420, loss=0.75  (1910.4  examples/sec;  0.067  sec/batch)
step 25430, loss=0.65  (1862.8  examples/sec;  0.069  sec/batch)
step 25440, loss=0.72  (1860.0  examples/sec;  0.069  sec/batch)
step 25450, loss=0.69  (1979.0  examples/sec;  0.065  sec/batch)
step 25460, loss=0.62  (1972.4  examples/sec;  0.065  sec/batch)
step 25470, loss=0.72  (1822.3  examples/sec;  0.070  sec/batch)
step 25480, loss=0.68  (1893.1  examples/sec;  0.068  sec/batch)
step 25490, loss=0.88  (1947.0  examples/sec;  0.066  sec/batch)
step 25500, loss=0.69  (1943.9  examples/sec;  0.066  sec/batch)
step 25510, loss=0.75  (1853.4  examples/sec;  0.069  sec/batch)
step 25520, loss=0.64  (1869.7  examples/sec;  0.068  sec/batch)
step 25530, loss=0.67  (1855.7  examples/sec;  0.069  sec/batch)
step 25540, loss=0.73  (1905.7  examples/sec;  0.067  sec/batch)
step 25550, loss=0.64  (1888.5  examples/sec;  0.068  sec/batch)
step 25560, loss=0.61  (1927.8  examples/sec;  0.066  sec/batch)
step 25570, loss=0.61  (1900.1  examples/sec;  0.067  sec/batch)
step 25580, loss=0.66  (1959.6  examples/sec;  0.065  sec/batch)
step 25590, loss=0.70  (1934.1  examples/sec;  0.066  sec/batch)
step 25600, loss=0.73  (1894.3  examples/sec;  0.068  sec/batch)
step 25610, loss=0.61  (1862.6  examples/sec;  0.069  sec/batch)
step 25620, loss=0.60  (1962.6  examples/sec;  0.065  sec/batch)
step 25630, loss=0.76  (1699.6  examples/sec;  0.075  sec/batch)
step 25640, loss=0.66  (1833.8  examples/sec;  0.070  sec/batch)
step 25650, loss=0.58  (1932.3  examples/sec;  0.066  sec/batch)
step 25660, loss=0.91  (1892.5  examples/sec;  0.068  sec/batch)
step 25670, loss=0.69  (1962.0  examples/sec;  0.065  sec/batch)
step 25680, loss=0.83  (1861.3  examples/sec;  0.069  sec/batch)
step 25690, loss=0.74  (1885.2  examples/sec;  0.068  sec/batch)
step 25700, loss=0.75  (1838.6  examples/sec;  0.070  sec/batch)
step 25710, loss=0.85  (1950.6  examples/sec;  0.066  sec/batch)
step 25720, loss=0.70  (1891.3  examples/sec;  0.068  sec/batch)
step 25730, loss=0.71  (1951.5  examples/sec;  0.066  sec/batch)
step 25740, loss=0.75  (1955.3  examples/sec;  0.065  sec/batch)
step 25750, loss=0.68  (1830.2  examples/sec;  0.070  sec/batch)
step 25760, loss=0.77  (1951.5  examples/sec;  0.066  sec/batch)
step 25770, loss=0.72  (1776.7  examples/sec;  0.072  sec/batch)
step 25780, loss=0.76  (1916.4  examples/sec;  0.067  sec/batch)
step 25790, loss=0.74  (1880.7  examples/sec;  0.068  sec/batch)
step 25800, loss=0.73  (1854.5  examples/sec;  0.069  sec/batch)
step 25810, loss=0.72  (1857.0  examples/sec;  0.069  sec/batch)
step 25820, loss=0.78  (1870.6  examples/sec;  0.068  sec/batch)
step 25830, loss=0.78  (1960.4  examples/sec;  0.065  sec/batch)
step 25840, loss=0.67  (1904.4  examples/sec;  0.067  sec/batch)
step 25850, loss=0.75  (1929.9  examples/sec;  0.066  sec/batch)
step 25860, loss=0.61  (1945.5  examples/sec;  0.066  sec/batch)
step 25870, loss=0.81  (1780.8  examples/sec;  0.072  sec/batch)
step 25880, loss=0.70  (1929.3  examples/sec;  0.066  sec/batch)
step 25890, loss=0.65  (1820.8  examples/sec;  0.070  sec/batch)
step 25900, loss=0.71  (1884.6  examples/sec;  0.068  sec/batch)
step 25910, loss=0.56  (1977.6  examples/sec;  0.065  sec/batch)
step 25920, loss=0.71  (1906.0  examples/sec;  0.067  sec/batch)
step 25930, loss=0.69  (1865.3  examples/sec;  0.069  sec/batch)
step 25940, loss=0.63  (1596.0  examples/sec;  0.080  sec/batch)
step 25950, loss=0.69  (1892.2  examples/sec;  0.068  sec/batch)
step 25960, loss=0.67  (1862.8  examples/sec;  0.069  sec/batch)
step 25970, loss=0.71  (1811.2  examples/sec;  0.071  sec/batch)
step 25980, loss=0.77  (1935.7  examples/sec;  0.066  sec/batch)
step 25990, loss=0.88  (1889.6  examples/sec;  0.068  sec/batch)
step 26000, loss=0.90  (1849.8  examples/sec;  0.069  sec/batch)
step 26010, loss=0.58  (1841.3  examples/sec;  0.070  sec/batch)
step 26020, loss=0.85  (1898.0  examples/sec;  0.067  sec/batch)
step 26030, loss=0.66  (1904.6  examples/sec;  0.067  sec/batch)
step 26040, loss=0.64  (1913.8  examples/sec;  0.067  sec/batch)
step 26050, loss=0.83  (1740.8  examples/sec;  0.074  sec/batch)
step 26060, loss=0.79  (1834.3  examples/sec;  0.070  sec/batch)
step 26070, loss=0.78  (1956.9  examples/sec;  0.065  sec/batch)
step 26080, loss=0.65  (1905.1  examples/sec;  0.067  sec/batch)
step 26090, loss=0.67  (1968.9  examples/sec;  0.065  sec/batch)
step 26100, loss=0.54  (1901.8  examples/sec;  0.067  sec/batch)
step 26110, loss=0.66  (1867.9  examples/sec;  0.069  sec/batch)
step 26120, loss=0.73  (1861.8  examples/sec;  0.069  sec/batch)
step 26130, loss=0.80  (1846.9  examples/sec;  0.069  sec/batch)
step 26140, loss=0.79  (1854.7  examples/sec;  0.069  sec/batch)
step 26150, loss=0.88  (1909.5  examples/sec;  0.067  sec/batch)
step 26160, loss=0.79  (1895.3  examples/sec;  0.068  sec/batch)
step 26170, loss=0.61  (1869.5  examples/sec;  0.068  sec/batch)
step 26180, loss=0.56  (1944.6  examples/sec;  0.066  sec/batch)
step 26190, loss=0.88  (1924.6  examples/sec;  0.067  sec/batch)
step 26200, loss=0.79  (1831.6  examples/sec;  0.070  sec/batch)
step 26210, loss=0.72  (1887.3  examples/sec;  0.068  sec/batch)
step 26220, loss=0.76  (1888.5  examples/sec;  0.068  sec/batch)
step 26230, loss=0.60  (1906.8  examples/sec;  0.067  sec/batch)
step 26240, loss=0.62  (1895.4  examples/sec;  0.068  sec/batch)
step 26250, loss=0.70  (1897.6  examples/sec;  0.067  sec/batch)
step 26260, loss=0.83  (2001.6  examples/sec;  0.064  sec/batch)
step 26270, loss=0.56  (1897.0  examples/sec;  0.067  sec/batch)
step 26280, loss=0.80  (1893.2  examples/sec;  0.068  sec/batch)
step 26290, loss=0.83  (1949.0  examples/sec;  0.066  sec/batch)
step 26300, loss=0.77  (1552.8  examples/sec;  0.082  sec/batch)
step 26310, loss=0.69  (1928.9  examples/sec;  0.066  sec/batch)
step 26320, loss=0.54  (1860.0  examples/sec;  0.069  sec/batch)
step 26330, loss=0.67  (1735.7  examples/sec;  0.074  sec/batch)
step 26340, loss=0.78  (1929.7  examples/sec;  0.066  sec/batch)
step 26350, loss=0.66  (1900.4  examples/sec;  0.067  sec/batch)
step 26360, loss=0.68  (1901.7  examples/sec;  0.067  sec/batch)
step 26370, loss=0.69  (1898.0  examples/sec;  0.067  sec/batch)
step 26380, loss=0.60  (1913.6  examples/sec;  0.067  sec/batch)
step 26390, loss=0.74  (1984.2  examples/sec;  0.065  sec/batch)
step 26400, loss=0.86  (1908.6  examples/sec;  0.067  sec/batch)
step 26410, loss=0.78  (1921.6  examples/sec;  0.067  sec/batch)
step 26420, loss=0.80  (1920.6  examples/sec;  0.067  sec/batch)
step 26430, loss=0.82  (1877.3  examples/sec;  0.068  sec/batch)
step 26440, loss=0.77  (1949.5  examples/sec;  0.066  sec/batch)
step 26450, loss=0.70  (1907.7  examples/sec;  0.067  sec/batch)
step 26460, loss=0.54  (1947.2  examples/sec;  0.066  sec/batch)
step 26470, loss=0.60  (1934.7  examples/sec;  0.066  sec/batch)
step 26480, loss=0.71  (1875.9  examples/sec;  0.068  sec/batch)
step 26490, loss=0.65  (1849.1  examples/sec;  0.069  sec/batch)
step 26500, loss=0.63  (1901.0  examples/sec;  0.067  sec/batch)
step 26510, loss=0.63  (1930.3  examples/sec;  0.066  sec/batch)
step 26520, loss=0.85  (2000.1  examples/sec;  0.064  sec/batch)
step 26530, loss=0.62  (2035.7  examples/sec;  0.063  sec/batch)
step 26540, loss=0.71  (1599.8  examples/sec;  0.080  sec/batch)
step 26550, loss=0.68  (1884.5  examples/sec;  0.068  sec/batch)
step 26560, loss=0.84  (1890.7  examples/sec;  0.068  sec/batch)
step 26570, loss=0.69  (1840.4  examples/sec;  0.070  sec/batch)
step 26580, loss=0.83  (1836.4  examples/sec;  0.070  sec/batch)
step 26590, loss=0.80  (1911.5  examples/sec;  0.067  sec/batch)
step 26600, loss=0.69  (1914.9  examples/sec;  0.067  sec/batch)
step 26610, loss=0.81  (1896.9  examples/sec;  0.067  sec/batch)
step 26620, loss=0.66  (1927.2  examples/sec;  0.066  sec/batch)
step 26630, loss=0.84  (1714.9  examples/sec;  0.075  sec/batch)
step 26640, loss=0.71  (1858.3  examples/sec;  0.069  sec/batch)
step 26650, loss=0.85  (1943.6  examples/sec;  0.066  sec/batch)
step 26660, loss=0.57  (1774.6  examples/sec;  0.072  sec/batch)
step 26670, loss=0.62  (1963.3  examples/sec;  0.065  sec/batch)
step 26680, loss=0.64  (1912.1  examples/sec;  0.067  sec/batch)
step 26690, loss=0.73  (1940.8  examples/sec;  0.066  sec/batch)
step 26700, loss=0.71  (1944.2  examples/sec;  0.066  sec/batch)
step 26710, loss=0.70  (1892.8  examples/sec;  0.068  sec/batch)
step 26720, loss=0.80  (1915.0  examples/sec;  0.067  sec/batch)
step 26730, loss=0.74  (1804.2  examples/sec;  0.071  sec/batch)
step 26740, loss=0.56  (1889.2  examples/sec;  0.068  sec/batch)
step 26750, loss=0.72  (1823.5  examples/sec;  0.070  sec/batch)
step 26760, loss=0.63  (1967.5  examples/sec;  0.065  sec/batch)
step 26770, loss=0.71  (1894.1  examples/sec;  0.068  sec/batch)
step 26780, loss=0.68  (1877.3  examples/sec;  0.068  sec/batch)
step 26790, loss=0.73  (1890.3  examples/sec;  0.068  sec/batch)
step 26800, loss=0.71  (1873.9  examples/sec;  0.068  sec/batch)
step 26810, loss=0.76  (1867.2  examples/sec;  0.069  sec/batch)
step 26820, loss=0.76  (1905.0  examples/sec;  0.067  sec/batch)
step 26830, loss=0.62  (1957.2  examples/sec;  0.065  sec/batch)
step 26840, loss=0.62  (1863.6  examples/sec;  0.069  sec/batch)
step 26850, loss=0.69  (1898.0  examples/sec;  0.067  sec/batch)
step 26860, loss=0.83  (1955.7  examples/sec;  0.065  sec/batch)
step 26870, loss=0.75  (1862.1  examples/sec;  0.069  sec/batch)
step 26880, loss=0.71  (1879.8  examples/sec;  0.068  sec/batch)
step 26890, loss=0.74  (1940.7  examples/sec;  0.066  sec/batch)
step 26900, loss=0.83  (1869.4  examples/sec;  0.068  sec/batch)
step 26910, loss=0.77  (1873.4  examples/sec;  0.068  sec/batch)
step 26920, loss=0.82  (1905.8  examples/sec;  0.067  sec/batch)
step 26930, loss=0.63  (1866.2  examples/sec;  0.069  sec/batch)
step 26940, loss=0.76  (1966.7  examples/sec;  0.065  sec/batch)
step 26950, loss=0.73  (1934.2  examples/sec;  0.066  sec/batch)
step 26960, loss=0.65  (1995.3  examples/sec;  0.064  sec/batch)
step 26970, loss=0.78  (1924.6  examples/sec;  0.067  sec/batch)
step 26980, loss=0.59  (1939.5  examples/sec;  0.066  sec/batch)
step 26990, loss=0.86  (1929.7  examples/sec;  0.066  sec/batch)
step 27000, loss=0.70  (1829.5  examples/sec;  0.070  sec/batch)
step 27010, loss=0.69  (1866.0  examples/sec;  0.069  sec/batch)
step 27020, loss=0.72  (1902.6  examples/sec;  0.067  sec/batch)
step 27030, loss=0.73  (1859.6  examples/sec;  0.069  sec/batch)
step 27040, loss=0.74  (1838.7  examples/sec;  0.070  sec/batch)
step 27050, loss=0.74  (1851.2  examples/sec;  0.069  sec/batch)
step 27060, loss=0.61  (1915.5  examples/sec;  0.067  sec/batch)
step 27070, loss=0.65  (1872.9  examples/sec;  0.068  sec/batch)
step 27080, loss=0.70  (1953.7  examples/sec;  0.066  sec/batch)
step 27090, loss=0.87  (1897.9  examples/sec;  0.067  sec/batch)
step 27100, loss=0.67  (1895.3  examples/sec;  0.068  sec/batch)
step 27110, loss=0.73  (2007.8  examples/sec;  0.064  sec/batch)
step 27120, loss=0.71  (1954.8  examples/sec;  0.065  sec/batch)
step 27130, loss=0.68  (1719.1  examples/sec;  0.074  sec/batch)
step 27140, loss=0.79  (1973.3  examples/sec;  0.065  sec/batch)
step 27150, loss=0.75  (1956.3  examples/sec;  0.065  sec/batch)
step 27160, loss=0.63  (1833.4  examples/sec;  0.070  sec/batch)
step 27170, loss=0.84  (1907.7  examples/sec;  0.067  sec/batch)
step 27180, loss=0.80  (1869.5  examples/sec;  0.068  sec/batch)
step 27190, loss=0.66  (1674.2  examples/sec;  0.076  sec/batch)
step 27200, loss=0.87  (1854.2  examples/sec;  0.069  sec/batch)
step 27210, loss=0.60  (1922.1  examples/sec;  0.067  sec/batch)
step 27220, loss=0.64  (1963.8  examples/sec;  0.065  sec/batch)
step 27230, loss=0.75  (1839.5  examples/sec;  0.070  sec/batch)
step 27240, loss=0.64  (1956.7  examples/sec;  0.065  sec/batch)
step 27250, loss=0.68  (1973.1  examples/sec;  0.065  sec/batch)
step 27260, loss=0.72  (1876.7  examples/sec;  0.068  sec/batch)
step 27270, loss=0.67  (1923.7  examples/sec;  0.067  sec/batch)
step 27280, loss=0.71  (1929.7  examples/sec;  0.066  sec/batch)
step 27290, loss=0.68  (1940.4  examples/sec;  0.066  sec/batch)
step 27300, loss=0.81  (1923.6  examples/sec;  0.067  sec/batch)
step 27310, loss=0.63  (1942.4  examples/sec;  0.066  sec/batch)
step 27320, loss=0.64  (1916.3  examples/sec;  0.067  sec/batch)
step 27330, loss=0.65  (1896.2  examples/sec;  0.068  sec/batch)
step 27340, loss=0.62  (1938.0  examples/sec;  0.066  sec/batch)
step 27350, loss=0.66  (1915.8  examples/sec;  0.067  sec/batch)
step 27360, loss=0.63  (1792.0  examples/sec;  0.071  sec/batch)
step 27370, loss=0.82  (1723.6  examples/sec;  0.074  sec/batch)
step 27380, loss=0.67  (1889.5  examples/sec;  0.068  sec/batch)
step 27390, loss=0.56  (1880.1  examples/sec;  0.068  sec/batch)
step 27400, loss=0.93  (1898.6  examples/sec;  0.067  sec/batch)
step 27410, loss=0.69  (1926.9  examples/sec;  0.066  sec/batch)
step 27420, loss=0.81  (1965.9  examples/sec;  0.065  sec/batch)
step 27430, loss=0.75  (1963.8  examples/sec;  0.065  sec/batch)
step 27440, loss=0.65  (1932.4  examples/sec;  0.066  sec/batch)
step 27450, loss=0.67  (1949.1  examples/sec;  0.066  sec/batch)
step 27460, loss=0.68  (1931.0  examples/sec;  0.066  sec/batch)
step 27470, loss=0.66  (1857.5  examples/sec;  0.069  sec/batch)
step 27480, loss=0.66  (1896.7  examples/sec;  0.067  sec/batch)
step 27490, loss=0.68  (1928.5  examples/sec;  0.066  sec/batch)
step 27500, loss=0.65  (1943.1  examples/sec;  0.066  sec/batch)
step 27510, loss=0.67  (1693.1  examples/sec;  0.076  sec/batch)
step 27520, loss=0.78  (1873.4  examples/sec;  0.068  sec/batch)
step 27530, loss=0.74  (1844.4  examples/sec;  0.069  sec/batch)
step 27540, loss=0.65  (1862.0  examples/sec;  0.069  sec/batch)
step 27550, loss=0.78  (1874.1  examples/sec;  0.068  sec/batch)
step 27560, loss=0.69  (1898.6  examples/sec;  0.067  sec/batch)
step 27570, loss=0.76  (1896.5  examples/sec;  0.067  sec/batch)
step 27580, loss=0.67  (1946.9  examples/sec;  0.066  sec/batch)
step 27590, loss=0.76  (1824.7  examples/sec;  0.070  sec/batch)
step 27600, loss=0.61  (1863.7  examples/sec;  0.069  sec/batch)
step 27610, loss=0.74  (1918.3  examples/sec;  0.067  sec/batch)
step 27620, loss=0.65  (1831.4  examples/sec;  0.070  sec/batch)
step 27630, loss=0.93  (1878.5  examples/sec;  0.068  sec/batch)
step 27640, loss=0.71  (1988.1  examples/sec;  0.064  sec/batch)
step 27650, loss=0.59  (1847.8  examples/sec;  0.069  sec/batch)
step 27660, loss=0.67  (1711.8  examples/sec;  0.075  sec/batch)
step 27670, loss=0.72  (1905.7  examples/sec;  0.067  sec/batch)
step 27680, loss=0.71  (1902.2  examples/sec;  0.067  sec/batch)
step 27690, loss=0.75  (1925.5  examples/sec;  0.066  sec/batch)
step 27700, loss=0.69  (1851.2  examples/sec;  0.069  sec/batch)
step 27710, loss=0.67  (1959.7  examples/sec;  0.065  sec/batch)
step 27720, loss=0.68  (2010.7  examples/sec;  0.064  sec/batch)
step 27730, loss=0.67  (1865.8  examples/sec;  0.069  sec/batch)
step 27740, loss=0.56  (1889.8  examples/sec;  0.068  sec/batch)
step 27750, loss=0.70  (1958.4  examples/sec;  0.065  sec/batch)
step 27760, loss=0.85  (1929.3  examples/sec;  0.066  sec/batch)
step 27770, loss=0.66  (1843.3  examples/sec;  0.069  sec/batch)
step 27780, loss=0.63  (1992.4  examples/sec;  0.064  sec/batch)
step 27790, loss=0.71  (1837.9  examples/sec;  0.070  sec/batch)
step 27800, loss=0.81  (1697.8  examples/sec;  0.075  sec/batch)
step 27810, loss=0.91  (1764.5  examples/sec;  0.073  sec/batch)
step 27820, loss=0.63  (1815.7  examples/sec;  0.070  sec/batch)
step 27830, loss=0.66  (1925.6  examples/sec;  0.066  sec/batch)
step 27840, loss=0.62  (2003.8  examples/sec;  0.064  sec/batch)
step 27850, loss=0.78  (1968.9  examples/sec;  0.065  sec/batch)
step 27860, loss=0.68  (1882.5  examples/sec;  0.068  sec/batch)
step 27870, loss=0.69  (1969.4  examples/sec;  0.065  sec/batch)
step 27880, loss=0.77  (1950.1  examples/sec;  0.066  sec/batch)
step 27890, loss=0.61  (1889.5  examples/sec;  0.068  sec/batch)
step 27900, loss=0.65  (1853.6  examples/sec;  0.069  sec/batch)
step 27910, loss=0.78  (1915.4  examples/sec;  0.067  sec/batch)
step 27920, loss=0.61  (1968.6  examples/sec;  0.065  sec/batch)
step 27930, loss=0.82  (1828.7  examples/sec;  0.070  sec/batch)
step 27940, loss=0.70  (1859.6  examples/sec;  0.069  sec/batch)
step 27950, loss=0.69  (1959.3  examples/sec;  0.065  sec/batch)
step 27960, loss=0.76  (1979.5  examples/sec;  0.065  sec/batch)
step 27970, loss=0.67  (1973.1  examples/sec;  0.065  sec/batch)
step 27980, loss=0.65  (1918.4  examples/sec;  0.067  sec/batch)
step 27990, loss=0.64  (1823.8  examples/sec;  0.070  sec/batch)
step 28000, loss=0.70  (1891.4  examples/sec;  0.068  sec/batch)
step 28010, loss=0.73  (1876.6  examples/sec;  0.068  sec/batch)
step 28020, loss=0.59  (1989.2  examples/sec;  0.064  sec/batch)
step 28030, loss=0.61  (1919.6  examples/sec;  0.067  sec/batch)
step 28040, loss=0.83  (1526.9  examples/sec;  0.084  sec/batch)
step 28050, loss=0.61  (1932.2  examples/sec;  0.066  sec/batch)
step 28060, loss=0.67  (1874.0  examples/sec;  0.068  sec/batch)
step 28070, loss=0.76  (1770.5  examples/sec;  0.072  sec/batch)
step 28080, loss=0.63  (1828.4  examples/sec;  0.070  sec/batch)
step 28090, loss=0.70  (1886.1  examples/sec;  0.068  sec/batch)
step 28100, loss=0.74  (1881.7  examples/sec;  0.068  sec/batch)
step 28110, loss=0.57  (1928.8  examples/sec;  0.066  sec/batch)
step 28120, loss=0.88  (1967.4  examples/sec;  0.065  sec/batch)
step 28130, loss=0.55  (1917.6  examples/sec;  0.067  sec/batch)
step 28140, loss=0.68  (1890.5  examples/sec;  0.068  sec/batch)
step 28150, loss=0.62  (1938.8  examples/sec;  0.066  sec/batch)
step 28160, loss=0.84  (1994.3  examples/sec;  0.064  sec/batch)
step 28170, loss=0.78  (1820.1  examples/sec;  0.070  sec/batch)
step 28180, loss=0.75  (1943.2  examples/sec;  0.066  sec/batch)
step 28190, loss=0.63  (1878.5  examples/sec;  0.068  sec/batch)
step 28200, loss=0.62  (1857.1  examples/sec;  0.069  sec/batch)
step 28210, loss=0.72  (1918.8  examples/sec;  0.067  sec/batch)
step 28220, loss=0.74  (1768.3  examples/sec;  0.072  sec/batch)
step 28230, loss=0.73  (1824.6  examples/sec;  0.070  sec/batch)
step 28240, loss=0.75  (1982.6  examples/sec;  0.065  sec/batch)
step 28250, loss=0.72  (1977.2  examples/sec;  0.065  sec/batch)
step 28260, loss=0.79  (1898.9  examples/sec;  0.067  sec/batch)
step 28270, loss=0.75  (1903.7  examples/sec;  0.067  sec/batch)
step 28280, loss=0.70  (1704.1  examples/sec;  0.075  sec/batch)
step 28290, loss=0.66  (1871.0  examples/sec;  0.068  sec/batch)
step 28300, loss=0.85  (1920.0  examples/sec;  0.067  sec/batch)
step 28310, loss=0.64  (1888.3  examples/sec;  0.068  sec/batch)
step 28320, loss=0.64  (1766.7  examples/sec;  0.072  sec/batch)
step 28330, loss=0.63  (1928.9  examples/sec;  0.066  sec/batch)
step 28340, loss=0.89  (1909.8  examples/sec;  0.067  sec/batch)
step 28350, loss=0.73  (2005.0  examples/sec;  0.064  sec/batch)
step 28360, loss=0.82  (1928.7  examples/sec;  0.066  sec/batch)
step 28370, loss=0.74  (1824.2  examples/sec;  0.070  sec/batch)
step 28380, loss=0.79  (1875.3  examples/sec;  0.068  sec/batch)
step 28390, loss=0.81  (1755.4  examples/sec;  0.073  sec/batch)
step 28400, loss=0.57  (1861.5  examples/sec;  0.069  sec/batch)
step 28410, loss=0.87  (1906.5  examples/sec;  0.067  sec/batch)
step 28420, loss=0.79  (1923.6  examples/sec;  0.067  sec/batch)
step 28430, loss=0.76  (1979.3  examples/sec;  0.065  sec/batch)
step 28440, loss=0.87  (1857.5  examples/sec;  0.069  sec/batch)
step 28450, loss=0.81  (1722.2  examples/sec;  0.074  sec/batch)
step 28460, loss=0.67  (1862.8  examples/sec;  0.069  sec/batch)
step 28470, loss=0.88  (1878.0  examples/sec;  0.068  sec/batch)
step 28480, loss=0.61  (1899.8  examples/sec;  0.067  sec/batch)
step 28490, loss=0.71  (1996.1  examples/sec;  0.064  sec/batch)
step 28500, loss=0.86  (1930.8  examples/sec;  0.066  sec/batch)
step 28510, loss=0.89  (1834.9  examples/sec;  0.070  sec/batch)
step 28520, loss=0.86  (1836.0  examples/sec;  0.070  sec/batch)
step 28530, loss=0.77  (1952.0  examples/sec;  0.066  sec/batch)
step 28540, loss=0.78  (1759.9  examples/sec;  0.073  sec/batch)
step 28550, loss=0.77  (1980.1  examples/sec;  0.065  sec/batch)
step 28560, loss=0.76  (1864.1  examples/sec;  0.069  sec/batch)
step 28570, loss=0.69  (1870.5  examples/sec;  0.068  sec/batch)
step 28580, loss=0.62  (1714.1  examples/sec;  0.075  sec/batch)
step 28590, loss=0.78  (1961.5  examples/sec;  0.065  sec/batch)
step 28600, loss=0.75  (1897.4  examples/sec;  0.067  sec/batch)
step 28610, loss=0.56  (1989.7  examples/sec;  0.064  sec/batch)
step 28620, loss=0.65  (1905.7  examples/sec;  0.067  sec/batch)
step 28630, loss=0.54  (1935.8  examples/sec;  0.066  sec/batch)
step 28640, loss=0.80  (1911.0  examples/sec;  0.067  sec/batch)
step 28650, loss=0.67  (1899.0  examples/sec;  0.067  sec/batch)
step 28660, loss=0.85  (1973.4  examples/sec;  0.065  sec/batch)
step 28670, loss=0.69  (1968.2  examples/sec;  0.065  sec/batch)
step 28680, loss=0.70  (1840.7  examples/sec;  0.070  sec/batch)
step 28690, loss=0.58  (1909.7  examples/sec;  0.067  sec/batch)
step 28700, loss=0.62  (1870.5  examples/sec;  0.068  sec/batch)
step 28710, loss=0.72  (1628.0  examples/sec;  0.079  sec/batch)
step 28720, loss=0.76  (1995.4  examples/sec;  0.064  sec/batch)
step 28730, loss=0.71  (1927.7  examples/sec;  0.066  sec/batch)
step 28740, loss=0.77  (1978.2  examples/sec;  0.065  sec/batch)
step 28750, loss=0.78  (1868.2  examples/sec;  0.069  sec/batch)
step 28760, loss=0.61  (1909.6  examples/sec;  0.067  sec/batch)
step 28770, loss=0.87  (1955.0  examples/sec;  0.065  sec/batch)
step 28780, loss=0.84  (1867.7  examples/sec;  0.069  sec/batch)
step 28790, loss=0.71  (1855.5  examples/sec;  0.069  sec/batch)
step 28800, loss=0.78  (1875.9  examples/sec;  0.068  sec/batch)
step 28810, loss=0.72  (1908.1  examples/sec;  0.067  sec/batch)
step 28820, loss=0.62  (1861.1  examples/sec;  0.069  sec/batch)
step 28830, loss=0.68  (1927.2  examples/sec;  0.066  sec/batch)
step 28840, loss=0.78  (1983.0  examples/sec;  0.065  sec/batch)
step 28850, loss=0.63  (1973.8  examples/sec;  0.065  sec/batch)
step 28860, loss=0.74  (1916.2  examples/sec;  0.067  sec/batch)
step 28870, loss=0.67  (1934.1  examples/sec;  0.066  sec/batch)
step 28880, loss=0.65  (1886.1  examples/sec;  0.068  sec/batch)
step 28890, loss=0.71  (1900.9  examples/sec;  0.067  sec/batch)
step 28900, loss=0.65  (1867.6  examples/sec;  0.069  sec/batch)
step 28910, loss=0.60  (1924.4  examples/sec;  0.067  sec/batch)
step 28920, loss=0.65  (1926.5  examples/sec;  0.066  sec/batch)
step 28930, loss=0.75  (1851.0  examples/sec;  0.069  sec/batch)
step 28940, loss=0.75  (1921.0  examples/sec;  0.067  sec/batch)
step 28950, loss=0.75  (1915.7  examples/sec;  0.067  sec/batch)
step 28960, loss=0.69  (1900.2  examples/sec;  0.067  sec/batch)
step 28970, loss=0.71  (1883.7  examples/sec;  0.068  sec/batch)
step 28980, loss=0.55  (1882.8  examples/sec;  0.068  sec/batch)
step 28990, loss=0.77  (1899.0  examples/sec;  0.067  sec/batch)
step 29000, loss=0.73  (1776.5  examples/sec;  0.072  sec/batch)
step 29010, loss=0.77  (1909.3  examples/sec;  0.067  sec/batch)
step 29020, loss=0.59  (1925.2  examples/sec;  0.066  sec/batch)
step 29030, loss=0.63  (1963.9  examples/sec;  0.065  sec/batch)
step 29040, loss=0.65  (1859.4  examples/sec;  0.069  sec/batch)
step 29050, loss=0.67  (1824.2  examples/sec;  0.070  sec/batch)
step 29060, loss=0.71  (1868.1  examples/sec;  0.069  sec/batch)
step 29070, loss=0.58  (1917.3  examples/sec;  0.067  sec/batch)
step 29080, loss=0.75  (1853.3  examples/sec;  0.069  sec/batch)
step 29090, loss=0.55  (1917.4  examples/sec;  0.067  sec/batch)
step 29100, loss=0.68  (1849.2  examples/sec;  0.069  sec/batch)
step 29110, loss=0.90  (1854.4  examples/sec;  0.069  sec/batch)
step 29120, loss=0.88  (1854.2  examples/sec;  0.069  sec/batch)
step 29130, loss=0.72  (1911.7  examples/sec;  0.067  sec/batch)
step 29140, loss=0.69  (1903.1  examples/sec;  0.067  sec/batch)
step 29150, loss=0.61  (1899.5  examples/sec;  0.067  sec/batch)
step 29160, loss=0.79  (1886.1  examples/sec;  0.068  sec/batch)
step 29170, loss=0.66  (1986.8  examples/sec;  0.064  sec/batch)
step 29180, loss=0.54  (1954.3  examples/sec;  0.065  sec/batch)
step 29190, loss=0.60  (1859.5  examples/sec;  0.069  sec/batch)
step 29200, loss=0.77  (1841.2  examples/sec;  0.070  sec/batch)
step 29210, loss=0.72  (1898.2  examples/sec;  0.067  sec/batch)
step 29220, loss=0.67  (1945.3  examples/sec;  0.066  sec/batch)
step 29230, loss=0.65  (1936.0  examples/sec;  0.066  sec/batch)
step 29240, loss=0.82  (1930.5  examples/sec;  0.066  sec/batch)
step 29250, loss=0.58  (1962.7  examples/sec;  0.065  sec/batch)
step 29260, loss=0.65  (1844.5  examples/sec;  0.069  sec/batch)
step 29270, loss=0.80  (1860.0  examples/sec;  0.069  sec/batch)
step 29280, loss=0.84  (1878.4  examples/sec;  0.068  sec/batch)
step 29290, loss=0.84  (1889.5  examples/sec;  0.068  sec/batch)
step 29300, loss=0.70  (1932.3  examples/sec;  0.066  sec/batch)
step 29310, loss=0.75  (1884.5  examples/sec;  0.068  sec/batch)
step 29320, loss=0.66  (1918.4  examples/sec;  0.067  sec/batch)
step 29330, loss=0.75  (1922.4  examples/sec;  0.067  sec/batch)
step 29340, loss=0.75  (1954.5  examples/sec;  0.065  sec/batch)
step 29350, loss=0.72  (1949.1  examples/sec;  0.066  sec/batch)
step 29360, loss=0.68  (1908.8  examples/sec;  0.067  sec/batch)
step 29370, loss=0.72  (1901.0  examples/sec;  0.067  sec/batch)
step 29380, loss=0.56  (1971.7  examples/sec;  0.065  sec/batch)
step 29390, loss=0.57  (1857.8  examples/sec;  0.069  sec/batch)
step 29400, loss=0.59  (1921.9  examples/sec;  0.067  sec/batch)
step 29410, loss=0.56  (1893.4  examples/sec;  0.068  sec/batch)
step 29420, loss=0.73  (1911.0  examples/sec;  0.067  sec/batch)
step 29430, loss=0.84  (1931.8  examples/sec;  0.066  sec/batch)
step 29440, loss=0.69  (1892.4  examples/sec;  0.068  sec/batch)
step 29450, loss=0.67  (1666.0  examples/sec;  0.077  sec/batch)
step 29460, loss=0.82  (1834.3  examples/sec;  0.070  sec/batch)
step 29470, loss=0.70  (1925.3  examples/sec;  0.066  sec/batch)
step 29480, loss=0.55  (1918.8  examples/sec;  0.067  sec/batch)
step 29490, loss=0.64  (1863.6  examples/sec;  0.069  sec/batch)
step 29500, loss=0.72  (1904.1  examples/sec;  0.067  sec/batch)
step 29510, loss=0.67  (1838.4  examples/sec;  0.070  sec/batch)
step 29520, loss=0.50  (1927.4  examples/sec;  0.066  sec/batch)
step 29530, loss=0.63  (1907.9  examples/sec;  0.067  sec/batch)
step 29540, loss=0.85  (1888.7  examples/sec;  0.068  sec/batch)
step 29550, loss=0.58  (1933.3  examples/sec;  0.066  sec/batch)
step 29560, loss=0.75  (1824.3  examples/sec;  0.070  sec/batch)
step 29570, loss=0.83  (1894.9  examples/sec;  0.068  sec/batch)
step 29580, loss=0.77  (1871.4  examples/sec;  0.068  sec/batch)
step 29590, loss=0.71  (1936.2  examples/sec;  0.066  sec/batch)
step 29600, loss=0.72  (1881.1  examples/sec;  0.068  sec/batch)
step 29610, loss=0.63  (1985.9  examples/sec;  0.064  sec/batch)
step 29620, loss=0.56  (1964.4  examples/sec;  0.065  sec/batch)
step 29630, loss=0.87  (1924.4  examples/sec;  0.067  sec/batch)
step 29640, loss=0.62  (1803.9  examples/sec;  0.071  sec/batch)
step 29650, loss=0.77  (1852.7  examples/sec;  0.069  sec/batch)
step 29660, loss=0.70  (1857.8  examples/sec;  0.069  sec/batch)
step 29670, loss=0.78  (1948.3  examples/sec;  0.066  sec/batch)
step 29680, loss=0.64  (1845.1  examples/sec;  0.069  sec/batch)
step 29690, loss=0.69  (1931.1  examples/sec;  0.066  sec/batch)
step 29700, loss=0.80  (1906.3  examples/sec;  0.067  sec/batch)
step 29710, loss=0.70  (1961.6  examples/sec;  0.065  sec/batch)
step 29720, loss=0.73  (1865.0  examples/sec;  0.069  sec/batch)
step 29730, loss=0.61  (1905.6  examples/sec;  0.067  sec/batch)
step 29740, loss=0.65  (1918.9  examples/sec;  0.067  sec/batch)
step 29750, loss=0.69  (1845.8  examples/sec;  0.069  sec/batch)
step 29760, loss=0.59  (1867.9  examples/sec;  0.069  sec/batch)
step 29770, loss=0.69  (1851.5  examples/sec;  0.069  sec/batch)
step 29780, loss=0.65  (1880.3  examples/sec;  0.068  sec/batch)
step 29790, loss=0.77  (1839.4  examples/sec;  0.070  sec/batch)
step 29800, loss=0.66  (1827.3  examples/sec;  0.070  sec/batch)
step 29810, loss=0.65  (1889.9  examples/sec;  0.068  sec/batch)
step 29820, loss=0.74  (1929.1  examples/sec;  0.066  sec/batch)
step 29830, loss=0.66  (1735.5  examples/sec;  0.074  sec/batch)
step 29840, loss=0.53  (2004.3  examples/sec;  0.064  sec/batch)
step 29850, loss=0.60  (1922.0  examples/sec;  0.067  sec/batch)
step 29860, loss=0.71  (1954.6  examples/sec;  0.065  sec/batch)
step 29870, loss=0.79  (1918.0  examples/sec;  0.067  sec/batch)
step 29880, loss=0.77  (1907.1  examples/sec;  0.067  sec/batch)
step 29890, loss=0.72  (1887.4  examples/sec;  0.068  sec/batch)
step 29900, loss=0.70  (1743.1  examples/sec;  0.073  sec/batch)
step 29910, loss=0.66  (1950.3  examples/sec;  0.066  sec/batch)
step 29920, loss=0.92  (1901.7  examples/sec;  0.067  sec/batch)
step 29930, loss=0.68  (1907.5  examples/sec;  0.067  sec/batch)
step 29940, loss=0.76  (1964.5  examples/sec;  0.065  sec/batch)
step 29950, loss=0.81  (1941.7  examples/sec;  0.066  sec/batch)
step 29960, loss=0.62  (1988.2  examples/sec;  0.064  sec/batch)
step 29970, loss=0.81  (1927.3  examples/sec;  0.066  sec/batch)
step 29980, loss=0.74  (1839.6  examples/sec;  0.070  sec/batch)
step 29990, loss=0.62  (1946.2  examples/sec;  0.066  sec/batch)
step 30000, loss=0.58  (1940.1  examples/sec;  0.066  sec/batch)
step 30010, loss=0.64  (1952.7  examples/sec;  0.066  sec/batch)
step 30020, loss=0.57  (1897.5  examples/sec;  0.067  sec/batch)
step 30030, loss=0.67  (1870.4  examples/sec;  0.068  sec/batch)
step 30040, loss=0.64  (2008.2  examples/sec;  0.064  sec/batch)
step 30050, loss=0.73  (1915.1  examples/sec;  0.067  sec/batch)
step 30060, loss=0.74  (1790.6  examples/sec;  0.071  sec/batch)
step 30070, loss=0.71  (1895.7  examples/sec;  0.068  sec/batch)
step 30080, loss=0.86  (1966.8  examples/sec;  0.065  sec/batch)
step 30090, loss=0.67  (1938.0  examples/sec;  0.066  sec/batch)
step 30100, loss=0.70  (1901.1  examples/sec;  0.067  sec/batch)
step 30110, loss=0.77  (1858.5  examples/sec;  0.069  sec/batch)
step 30120, loss=0.58  (1901.0  examples/sec;  0.067  sec/batch)
step 30130, loss=0.59  (1564.3  examples/sec;  0.082  sec/batch)
step 30140, loss=0.70  (1872.3  examples/sec;  0.068  sec/batch)
step 30150, loss=0.58  (1891.5  examples/sec;  0.068  sec/batch)
step 30160, loss=0.58  (1900.0  examples/sec;  0.067  sec/batch)
step 30170, loss=0.66  (1894.8  examples/sec;  0.068  sec/batch)
step 30180, loss=0.76  (1944.2  examples/sec;  0.066  sec/batch)
step 30190, loss=0.78  (1788.1  examples/sec;  0.072  sec/batch)
step 30200, loss=0.78  (1880.0  examples/sec;  0.068  sec/batch)
step 30210, loss=0.76  (1946.0  examples/sec;  0.066  sec/batch)
step 30220, loss=0.65  (1816.6  examples/sec;  0.070  sec/batch)
step 30230, loss=0.81  (1986.2  examples/sec;  0.064  sec/batch)
step 30240, loss=0.64  (1918.9  examples/sec;  0.067  sec/batch)
step 30250, loss=0.91  (1898.4  examples/sec;  0.067  sec/batch)
step 30260, loss=0.63  (1901.0  examples/sec;  0.067  sec/batch)
step 30270, loss=0.55  (1862.7  examples/sec;  0.069  sec/batch)
step 30280, loss=0.72  (1830.3  examples/sec;  0.070  sec/batch)
step 30290, loss=0.59  (1791.0  examples/sec;  0.071  sec/batch)
step 30300, loss=0.72  (1894.4  examples/sec;  0.068  sec/batch)
step 30310, loss=0.62  (2003.0  examples/sec;  0.064  sec/batch)
step 30320, loss=0.60  (1855.2  examples/sec;  0.069  sec/batch)
step 30330, loss=0.74  (1927.9  examples/sec;  0.066  sec/batch)
step 30340, loss=0.76  (1959.0  examples/sec;  0.065  sec/batch)
step 30350, loss=0.75  (1648.6  examples/sec;  0.078  sec/batch)
step 30360, loss=0.78  (1900.5  examples/sec;  0.067  sec/batch)
step 30370, loss=0.66  (1947.1  examples/sec;  0.066  sec/batch)
step 30380, loss=0.70  (1919.9  examples/sec;  0.067  sec/batch)
step 30390, loss=0.64  (1905.7  examples/sec;  0.067  sec/batch)
step 30400, loss=0.68  (1944.8  examples/sec;  0.066  sec/batch)
step 30410, loss=0.60  (1906.1  examples/sec;  0.067  sec/batch)
step 30420, loss=0.65  (1877.8  examples/sec;  0.068  sec/batch)
step 30430, loss=0.49  (1887.1  examples/sec;  0.068  sec/batch)
step 30440, loss=0.72  (1895.8  examples/sec;  0.068  sec/batch)
step 30450, loss=0.62  (1923.2  examples/sec;  0.067  sec/batch)
step 30460, loss=0.64  (1851.3  examples/sec;  0.069  sec/batch)
step 30470, loss=0.69  (1849.3  examples/sec;  0.069  sec/batch)
step 30480, loss=0.78  (1756.0  examples/sec;  0.073  sec/batch)
step 30490, loss=0.86  (1889.0  examples/sec;  0.068  sec/batch)
step 30500, loss=0.68  (1921.8  examples/sec;  0.067  sec/batch)
step 30510, loss=0.85  (1928.9  examples/sec;  0.066  sec/batch)
step 30520, loss=0.62  (1660.0  examples/sec;  0.077  sec/batch)
step 30530, loss=0.58  (1811.4  examples/sec;  0.071  sec/batch)
step 30540, loss=0.67  (1903.0  examples/sec;  0.067  sec/batch)
step 30550, loss=0.65  (1817.0  examples/sec;  0.070  sec/batch)
step 30560, loss=0.75  (1976.3  examples/sec;  0.065  sec/batch)
step 30570, loss=0.66  (1865.3  examples/sec;  0.069  sec/batch)
step 30580, loss=0.78  (1862.3  examples/sec;  0.069  sec/batch)
step 30590, loss=0.76  (1999.8  examples/sec;  0.064  sec/batch)
step 30600, loss=0.82  (1873.6  examples/sec;  0.068  sec/batch)
step 30610, loss=0.82  (1905.6  examples/sec;  0.067  sec/batch)
step 30620, loss=0.76  (1843.1  examples/sec;  0.069  sec/batch)
step 30630, loss=0.84  (1847.2  examples/sec;  0.069  sec/batch)
step 30640, loss=0.58  (1878.8  examples/sec;  0.068  sec/batch)
step 30650, loss=0.74  (1827.2  examples/sec;  0.070  sec/batch)
step 30660, loss=0.54  (1902.6  examples/sec;  0.067  sec/batch)
step 30670, loss=0.73  (1914.7  examples/sec;  0.067  sec/batch)
step 30680, loss=0.66  (1965.9  examples/sec;  0.065  sec/batch)
step 30690, loss=0.76  (1961.4  examples/sec;  0.065  sec/batch)
step 30700, loss=0.72  (1959.3  examples/sec;  0.065  sec/batch)
step 30710, loss=0.71  (1857.4  examples/sec;  0.069  sec/batch)
step 30720, loss=0.86  (1924.0  examples/sec;  0.067  sec/batch)
step 30730, loss=0.80  (1877.2  examples/sec;  0.068  sec/batch)
step 30740, loss=0.71  (1875.9  examples/sec;  0.068  sec/batch)
step 30750, loss=0.67  (1841.4  examples/sec;  0.070  sec/batch)
step 30760, loss=0.87  (1948.5  examples/sec;  0.066  sec/batch)
step 30770, loss=0.75  (2066.1  examples/sec;  0.062  sec/batch)
step 30780, loss=0.69  (1879.2  examples/sec;  0.068  sec/batch)
step 30790, loss=0.61  (2009.2  examples/sec;  0.064  sec/batch)
step 30800, loss=0.66  (1643.6  examples/sec;  0.078  sec/batch)
step 30810, loss=0.61  (1904.3  examples/sec;  0.067  sec/batch)
step 30820, loss=0.79  (1908.4  examples/sec;  0.067  sec/batch)
step 30830, loss=0.71  (1863.0  examples/sec;  0.069  sec/batch)
step 30840, loss=0.89  (1887.7  examples/sec;  0.068  sec/batch)
step 30850, loss=0.84  (1838.4  examples/sec;  0.070  sec/batch)
step 30860, loss=0.68  (2013.0  examples/sec;  0.064  sec/batch)
step 30870, loss=0.54  (1933.2  examples/sec;  0.066  sec/batch)
step 30880, loss=0.57  (1894.8  examples/sec;  0.068  sec/batch)
step 30890, loss=0.63  (1860.2  examples/sec;  0.069  sec/batch)
step 30900, loss=0.68  (1919.9  examples/sec;  0.067  sec/batch)
step 30910, loss=0.85  (1938.4  examples/sec;  0.066  sec/batch)
step 30920, loss=0.72  (1854.1  examples/sec;  0.069  sec/batch)
step 30930, loss=0.72  (1898.0  examples/sec;  0.067  sec/batch)
step 30940, loss=0.66  (1884.5  examples/sec;  0.068  sec/batch)
step 30950, loss=0.76  (1881.7  examples/sec;  0.068  sec/batch)
step 30960, loss=0.76  (1897.8  examples/sec;  0.067  sec/batch)
step 30970, loss=0.69  (1957.8  examples/sec;  0.065  sec/batch)
step 30980, loss=0.61  (1895.9  examples/sec;  0.068  sec/batch)
step 30990, loss=0.72  (1904.0  examples/sec;  0.067  sec/batch)
step 31000, loss=0.80  (1883.8  examples/sec;  0.068  sec/batch)
step 31010, loss=0.61  (1851.9  examples/sec;  0.069  sec/batch)
step 31020, loss=0.68  (1933.8  examples/sec;  0.066  sec/batch)
step 31030, loss=0.65  (1896.9  examples/sec;  0.067  sec/batch)
step 31040, loss=0.70  (1954.9  examples/sec;  0.065  sec/batch)
step 31050, loss=0.70  (1977.1  examples/sec;  0.065  sec/batch)
step 31060, loss=0.78  (1883.1  examples/sec;  0.068  sec/batch)
step 31070, loss=0.61  (1905.0  examples/sec;  0.067  sec/batch)
step 31080, loss=0.75  (2004.1  examples/sec;  0.064  sec/batch)
step 31090, loss=0.69  (1804.6  examples/sec;  0.071  sec/batch)
step 31100, loss=0.68  (1900.2  examples/sec;  0.067  sec/batch)
step 31110, loss=0.70  (1865.4  examples/sec;  0.069  sec/batch)
step 31120, loss=0.92  (1922.4  examples/sec;  0.067  sec/batch)
step 31130, loss=0.73  (1938.5  examples/sec;  0.066  sec/batch)
step 31140, loss=0.67  (1880.2  examples/sec;  0.068  sec/batch)
step 31150, loss=0.76  (1912.4  examples/sec;  0.067  sec/batch)
step 31160, loss=0.70  (1913.9  examples/sec;  0.067  sec/batch)
step 31170, loss=0.57  (1885.1  examples/sec;  0.068  sec/batch)
step 31180, loss=0.71  (1902.8  examples/sec;  0.067  sec/batch)
step 31190, loss=0.68  (1890.8  examples/sec;  0.068  sec/batch)
step 31200, loss=0.60  (1911.0  examples/sec;  0.067  sec/batch)
step 31210, loss=0.75  (1889.5  examples/sec;  0.068  sec/batch)
step 31220, loss=0.77  (1798.2  examples/sec;  0.071  sec/batch)
step 31230, loss=0.59  (1861.0  examples/sec;  0.069  sec/batch)
step 31240, loss=0.67  (1878.4  examples/sec;  0.068  sec/batch)
step 31250, loss=0.69  (1854.3  examples/sec;  0.069  sec/batch)
step 31260, loss=0.75  (2019.6  examples/sec;  0.063  sec/batch)
step 31270, loss=0.72  (1873.2  examples/sec;  0.068  sec/batch)
step 31280, loss=0.77  (1909.0  examples/sec;  0.067  sec/batch)
step 31290, loss=0.75  (1916.8  examples/sec;  0.067  sec/batch)
step 31300, loss=0.74  (1877.9  examples/sec;  0.068  sec/batch)
step 31310, loss=0.80  (1893.8  examples/sec;  0.068  sec/batch)
step 31320, loss=0.79  (1959.7  examples/sec;  0.065  sec/batch)
step 31330, loss=0.71  (1864.7  examples/sec;  0.069  sec/batch)
step 31340, loss=0.70  (1886.7  examples/sec;  0.068  sec/batch)
step 31350, loss=0.56  (1885.7  examples/sec;  0.068  sec/batch)
step 31360, loss=0.70  (1937.2  examples/sec;  0.066  sec/batch)
step 31370, loss=0.55  (1805.9  examples/sec;  0.071  sec/batch)
step 31380, loss=0.86  (1760.4  examples/sec;  0.073  sec/batch)
step 31390, loss=0.69  (1894.5  examples/sec;  0.068  sec/batch)
step 31400, loss=0.84  (1824.0  examples/sec;  0.070  sec/batch)
step 31410, loss=0.73  (1943.0  examples/sec;  0.066  sec/batch)
step 31420, loss=0.87  (1956.0  examples/sec;  0.065  sec/batch)
step 31430, loss=0.59  (1928.1  examples/sec;  0.066  sec/batch)
step 31440, loss=0.85  (2066.8  examples/sec;  0.062  sec/batch)
step 31450, loss=0.78  (1922.7  examples/sec;  0.067  sec/batch)
step 31460, loss=0.56  (1866.2  examples/sec;  0.069  sec/batch)
step 31470, loss=0.72  (1951.2  examples/sec;  0.066  sec/batch)
step 31480, loss=0.65  (1904.8  examples/sec;  0.067  sec/batch)
step 31490, loss=0.62  (1903.0  examples/sec;  0.067  sec/batch)
step 31500, loss=0.74  (1854.0  examples/sec;  0.069  sec/batch)
step 31510, loss=0.68  (1848.5  examples/sec;  0.069  sec/batch)
step 31520, loss=0.73  (1996.6  examples/sec;  0.064  sec/batch)
step 31530, loss=0.60  (1919.9  examples/sec;  0.067  sec/batch)
step 31540, loss=0.72  (1893.4  examples/sec;  0.068  sec/batch)
step 31550, loss=0.73  (1909.6  examples/sec;  0.067  sec/batch)
step 31560, loss=0.72  (1960.5  examples/sec;  0.065  sec/batch)
step 31570, loss=0.70  (1925.3  examples/sec;  0.066  sec/batch)
step 31580, loss=0.65  (1899.3  examples/sec;  0.067  sec/batch)
step 31590, loss=0.57  (1764.3  examples/sec;  0.073  sec/batch)
step 31600, loss=0.57  (1974.1  examples/sec;  0.065  sec/batch)
step 31610, loss=0.86  (1858.5  examples/sec;  0.069  sec/batch)
step 31620, loss=0.80  (1971.0  examples/sec;  0.065  sec/batch)
step 31630, loss=0.62  (1857.5  examples/sec;  0.069  sec/batch)
step 31640, loss=0.71  (1880.0  examples/sec;  0.068  sec/batch)
step 31650, loss=0.70  (1611.4  examples/sec;  0.079  sec/batch)
step 31660, loss=0.57  (1876.0  examples/sec;  0.068  sec/batch)
step 31670, loss=0.65  (1846.9  examples/sec;  0.069  sec/batch)
step 31680, loss=0.75  (1883.8  examples/sec;  0.068  sec/batch)
step 31690, loss=0.69  (1836.7  examples/sec;  0.070  sec/batch)
step 31700, loss=0.65  (1922.5  examples/sec;  0.067  sec/batch)
step 31710, loss=0.67  (1913.9  examples/sec;  0.067  sec/batch)
step 31720, loss=0.85  (1925.5  examples/sec;  0.066  sec/batch)
step 31730, loss=0.71  (1900.4  examples/sec;  0.067  sec/batch)
step 31740, loss=0.63  (1958.1  examples/sec;  0.065  sec/batch)
step 31750, loss=0.72  (1925.8  examples/sec;  0.066  sec/batch)
step 31760, loss=0.83  (1906.2  examples/sec;  0.067  sec/batch)
step 31770, loss=0.70  (1921.6  examples/sec;  0.067  sec/batch)
step 31780, loss=0.62  (1836.4  examples/sec;  0.070  sec/batch)
step 31790, loss=0.84  (1908.3  examples/sec;  0.067  sec/batch)
step 31800, loss=0.52  (1905.0  examples/sec;  0.067  sec/batch)
step 31810, loss=0.62  (1899.7  examples/sec;  0.067  sec/batch)
step 31820, loss=0.65  (1929.2  examples/sec;  0.066  sec/batch)
step 31830, loss=0.61  (1906.2  examples/sec;  0.067  sec/batch)
step 31840, loss=0.63  (2019.5  examples/sec;  0.063  sec/batch)
step 31850, loss=0.57  (1923.9  examples/sec;  0.067  sec/batch)
step 31860, loss=0.77  (1885.7  examples/sec;  0.068  sec/batch)
step 31870, loss=0.84  (1986.7  examples/sec;  0.064  sec/batch)
step 31880, loss=0.63  (1901.2  examples/sec;  0.067  sec/batch)
step 31890, loss=0.61  (1896.5  examples/sec;  0.067  sec/batch)
step 31900, loss=0.65  (1949.1  examples/sec;  0.066  sec/batch)
step 31910, loss=0.62  (1873.5  examples/sec;  0.068  sec/batch)
step 31920, loss=0.75  (1850.1  examples/sec;  0.069  sec/batch)
step 31930, loss=0.64  (1900.9  examples/sec;  0.067  sec/batch)
step 31940, loss=0.62  (1946.4  examples/sec;  0.066  sec/batch)
step 31950, loss=0.71  (1844.3  examples/sec;  0.069  sec/batch)
step 31960, loss=0.88  (1851.3  examples/sec;  0.069  sec/batch)
step 31970, loss=0.63  (1888.9  examples/sec;  0.068  sec/batch)
step 31980, loss=0.91  (1905.6  examples/sec;  0.067  sec/batch)
step 31990, loss=0.71  (1859.4  examples/sec;  0.069  sec/batch)
step 32000, loss=0.86  (1851.6  examples/sec;  0.069  sec/batch)
step 32010, loss=0.67  (1918.9  examples/sec;  0.067  sec/batch)
step 32020, loss=0.70  (1907.7  examples/sec;  0.067  sec/batch)
step 32030, loss=0.57  (1875.1  examples/sec;  0.068  sec/batch)
step 32040, loss=0.61  (1645.6  examples/sec;  0.078  sec/batch)
step 32050, loss=0.62  (1904.9  examples/sec;  0.067  sec/batch)
step 32060, loss=0.71  (1852.8  examples/sec;  0.069  sec/batch)
step 32070, loss=0.42  (1988.6  examples/sec;  0.064  sec/batch)
step 32080, loss=0.65  (1901.6  examples/sec;  0.067  sec/batch)
step 32090, loss=0.62  (1836.7  examples/sec;  0.070  sec/batch)
step 32100, loss=0.60  (1890.4  examples/sec;  0.068  sec/batch)
step 32110, loss=0.55  (1941.4  examples/sec;  0.066  sec/batch)
step 32120, loss=0.62  (1872.1  examples/sec;  0.068  sec/batch)
step 32130, loss=0.70  (1835.2  examples/sec;  0.070  sec/batch)
step 32140, loss=0.77  (1821.0  examples/sec;  0.070  sec/batch)
step 32150, loss=0.81  (1881.4  examples/sec;  0.068  sec/batch)
step 32160, loss=0.64  (1884.1  examples/sec;  0.068  sec/batch)
step 32170, loss=0.67  (1893.2  examples/sec;  0.068  sec/batch)
step 32180, loss=0.66  (1759.5  examples/sec;  0.073  sec/batch)
step 32190, loss=0.75  (1883.7  examples/sec;  0.068  sec/batch)
step 32200, loss=0.56  (1932.0  examples/sec;  0.066  sec/batch)
step 32210, loss=0.57  (1904.9  examples/sec;  0.067  sec/batch)
step 32220, loss=0.72  (1864.7  examples/sec;  0.069  sec/batch)
step 32230, loss=0.66  (1860.6  examples/sec;  0.069  sec/batch)
step 32240, loss=0.69  (1888.8  examples/sec;  0.068  sec/batch)
step 32250, loss=0.49  (1903.1  examples/sec;  0.067  sec/batch)
step 32260, loss=0.59  (1875.2  examples/sec;  0.068  sec/batch)
step 32270, loss=0.63  (1926.3  examples/sec;  0.066  sec/batch)
step 32280, loss=0.68  (1837.3  examples/sec;  0.070  sec/batch)
step 32290, loss=0.85  (1937.8  examples/sec;  0.066  sec/batch)
step 32300, loss=0.54  (1908.1  examples/sec;  0.067  sec/batch)
step 32310, loss=0.77  (1946.9  examples/sec;  0.066  sec/batch)
step 32320, loss=0.82  (1880.4  examples/sec;  0.068  sec/batch)
step 32330, loss=0.56  (1867.7  examples/sec;  0.069  sec/batch)
step 32340, loss=0.70  (1788.5  examples/sec;  0.072  sec/batch)
step 32350, loss=0.78  (1848.6  examples/sec;  0.069  sec/batch)
step 32360, loss=0.77  (1999.4  examples/sec;  0.064  sec/batch)
step 32370, loss=0.79  (1856.1  examples/sec;  0.069  sec/batch)
step 32380, loss=0.64  (1864.9  examples/sec;  0.069  sec/batch)
step 32390, loss=0.73  (1934.6  examples/sec;  0.066  sec/batch)
step 32400, loss=0.99  (2049.1  examples/sec;  0.062  sec/batch)
step 32410, loss=0.69  (1817.0  examples/sec;  0.070  sec/batch)
step 32420, loss=0.72  (1889.1  examples/sec;  0.068  sec/batch)
step 32430, loss=0.69  (1963.7  examples/sec;  0.065  sec/batch)
step 32440, loss=0.68  (1897.8  examples/sec;  0.067  sec/batch)
step 32450, loss=0.84  (1931.8  examples/sec;  0.066  sec/batch)
step 32460, loss=0.56  (1586.5  examples/sec;  0.081  sec/batch)
step 32470, loss=0.74  (1694.4  examples/sec;  0.076  sec/batch)
step 32480, loss=0.84  (1953.9  examples/sec;  0.066  sec/batch)
step 32490, loss=0.60  (1699.4  examples/sec;  0.075  sec/batch)
step 32500, loss=0.66  (1845.5  examples/sec;  0.069  sec/batch)
step 32510, loss=0.74  (1923.9  examples/sec;  0.067  sec/batch)
step 32520, loss=0.72  (1976.9  examples/sec;  0.065  sec/batch)
step 32530, loss=0.68  (1969.9  examples/sec;  0.065  sec/batch)
step 32540, loss=0.66  (1939.4  examples/sec;  0.066  sec/batch)
step 32550, loss=0.64  (1998.8  examples/sec;  0.064  sec/batch)
step 32560, loss=0.84  (1981.8  examples/sec;  0.065  sec/batch)
step 32570, loss=0.72  (1886.9  examples/sec;  0.068  sec/batch)
step 32580, loss=0.72  (1912.6  examples/sec;  0.067  sec/batch)
step 32590, loss=0.67  (1927.1  examples/sec;  0.066  sec/batch)
step 32600, loss=0.84  (1887.5  examples/sec;  0.068  sec/batch)
step 32610, loss=0.68  (1939.8  examples/sec;  0.066  sec/batch)
step 32620, loss=0.73  (1961.0  examples/sec;  0.065  sec/batch)
step 32630, loss=0.70  (1776.7  examples/sec;  0.072  sec/batch)
step 32640, loss=0.72  (1923.2  examples/sec;  0.067  sec/batch)
step 32650, loss=0.78  (1879.6  examples/sec;  0.068  sec/batch)
step 32660, loss=0.61  (1912.6  examples/sec;  0.067  sec/batch)
step 32670, loss=0.76  (1766.3  examples/sec;  0.072  sec/batch)
step 32680, loss=0.63  (1945.4  examples/sec;  0.066  sec/batch)
step 32690, loss=0.70  (1723.1  examples/sec;  0.074  sec/batch)
step 32700, loss=0.62  (1906.7  examples/sec;  0.067  sec/batch)
step 32710, loss=0.74  (1910.7  examples/sec;  0.067  sec/batch)
step 32720, loss=0.72  (1914.9  examples/sec;  0.067  sec/batch)
step 32730, loss=0.76  (1895.8  examples/sec;  0.068  sec/batch)
step 32740, loss=0.70  (1880.9  examples/sec;  0.068  sec/batch)
step 32750, loss=0.71  (1738.6  examples/sec;  0.074  sec/batch)
step 32760, loss=0.71  (1883.4  examples/sec;  0.068  sec/batch)
step 32770, loss=0.79  (1900.8  examples/sec;  0.067  sec/batch)
step 32780, loss=1.01  (1896.4  examples/sec;  0.067  sec/batch)
step 32790, loss=0.63  (1914.1  examples/sec;  0.067  sec/batch)
step 32800, loss=0.67  (1854.4  examples/sec;  0.069  sec/batch)
step 32810, loss=0.77  (1879.8  examples/sec;  0.068  sec/batch)
step 32820, loss=0.69  (1558.8  examples/sec;  0.082  sec/batch)
step 32830, loss=0.74  (1863.2  examples/sec;  0.069  sec/batch)
step 32840, loss=0.86  (1861.9  examples/sec;  0.069  sec/batch)
step 32850, loss=0.78  (2017.5  examples/sec;  0.063  sec/batch)
step 32860, loss=0.70  (1985.2  examples/sec;  0.064  sec/batch)
step 32870, loss=0.70  (1905.2  examples/sec;  0.067  sec/batch)
step 32880, loss=0.69  (1980.5  examples/sec;  0.065  sec/batch)
step 32890, loss=0.56  (1895.2  examples/sec;  0.068  sec/batch)
step 32900, loss=0.65  (1941.7  examples/sec;  0.066  sec/batch)
step 32910, loss=0.80  (1931.7  examples/sec;  0.066  sec/batch)
step 32920, loss=0.72  (1903.4  examples/sec;  0.067  sec/batch)
step 32930, loss=0.72  (1799.7  examples/sec;  0.071  sec/batch)
step 32940, loss=0.75  (1891.9  examples/sec;  0.068  sec/batch)
step 32950, loss=0.56  (1886.2  examples/sec;  0.068  sec/batch)
step 32960, loss=0.94  (1860.8  examples/sec;  0.069  sec/batch)
step 32970, loss=0.73  (1980.4  examples/sec;  0.065  sec/batch)
step 32980, loss=0.72  (1741.4  examples/sec;  0.074  sec/batch)
step 32990, loss=0.65  (1928.6  examples/sec;  0.066  sec/batch)
step 33000, loss=0.73  (1945.5  examples/sec;  0.066  sec/batch)
step 33010, loss=0.69  (1862.7  examples/sec;  0.069  sec/batch)
step 33020, loss=0.66  (1904.8  examples/sec;  0.067  sec/batch)
step 33030, loss=0.71  (1836.4  examples/sec;  0.070  sec/batch)
step 33040, loss=0.77  (1932.1  examples/sec;  0.066  sec/batch)
step 33050, loss=0.57  (1930.4  examples/sec;  0.066  sec/batch)
step 33060, loss=0.83  (1926.6  examples/sec;  0.066  sec/batch)
step 33070, loss=0.76  (1930.4  examples/sec;  0.066  sec/batch)
step 33080, loss=0.61  (1798.1  examples/sec;  0.071  sec/batch)
step 33090, loss=0.66  (1919.5  examples/sec;  0.067  sec/batch)
step 33100, loss=0.45  (1984.4  examples/sec;  0.065  sec/batch)
step 33110, loss=0.96  (1860.3  examples/sec;  0.069  sec/batch)
step 33120, loss=0.62  (1819.7  examples/sec;  0.070  sec/batch)
step 33130, loss=0.62  (1881.4  examples/sec;  0.068  sec/batch)
step 33140, loss=0.65  (1880.0  examples/sec;  0.068  sec/batch)
step 33150, loss=0.56  (1760.2  examples/sec;  0.073  sec/batch)
step 33160, loss=0.73  (1894.5  examples/sec;  0.068  sec/batch)
step 33170, loss=0.68  (1979.5  examples/sec;  0.065  sec/batch)
step 33180, loss=0.54  (1870.8  examples/sec;  0.068  sec/batch)
step 33190, loss=0.82  (1859.4  examples/sec;  0.069  sec/batch)
step 33200, loss=0.69  (1992.7  examples/sec;  0.064  sec/batch)
step 33210, loss=0.65  (1890.6  examples/sec;  0.068  sec/batch)
step 33220, loss=0.74  (1900.2  examples/sec;  0.067  sec/batch)
step 33230, loss=0.65  (1896.0  examples/sec;  0.068  sec/batch)
step 33240, loss=0.67  (1984.9  examples/sec;  0.064  sec/batch)
step 33250, loss=0.68  (1929.6  examples/sec;  0.066  sec/batch)
step 33260, loss=0.70  (1950.7  examples/sec;  0.066  sec/batch)
step 33270, loss=0.79  (1853.7  examples/sec;  0.069  sec/batch)
step 33280, loss=0.53  (1875.0  examples/sec;  0.068  sec/batch)
step 33290, loss=0.96  (1949.6  examples/sec;  0.066  sec/batch)
step 33300, loss=0.86  (1848.6  examples/sec;  0.069  sec/batch)
step 33310, loss=0.73  (1950.4  examples/sec;  0.066  sec/batch)
step 33320, loss=0.59  (1789.2  examples/sec;  0.072  sec/batch)
step 33330, loss=0.59  (1860.0  examples/sec;  0.069  sec/batch)
step 33340, loss=0.76  (1948.9  examples/sec;  0.066  sec/batch)
step 33350, loss=0.77  (1961.7  examples/sec;  0.065  sec/batch)
step 33360, loss=0.72  (1886.0  examples/sec;  0.068  sec/batch)
step 33370, loss=0.54  (1863.5  examples/sec;  0.069  sec/batch)
step 33380, loss=0.57  (1885.5  examples/sec;  0.068  sec/batch)
step 33390, loss=0.74  (1943.7  examples/sec;  0.066  sec/batch)
step 33400, loss=0.76  (1910.4  examples/sec;  0.067  sec/batch)
step 33410, loss=0.73  (1969.9  examples/sec;  0.065  sec/batch)
step 33420, loss=0.65  (1828.2  examples/sec;  0.070  sec/batch)
step 33430, loss=0.55  (1930.9  examples/sec;  0.066  sec/batch)
step 33440, loss=0.53  (1874.5  examples/sec;  0.068  sec/batch)
step 33450, loss=0.60  (1936.7  examples/sec;  0.066  sec/batch)
step 33460, loss=0.55  (1884.0  examples/sec;  0.068  sec/batch)
step 33470, loss=0.65  (1570.9  examples/sec;  0.081  sec/batch)
step 33480, loss=0.67  (1902.6  examples/sec;  0.067  sec/batch)
step 33490, loss=0.69  (1957.8  examples/sec;  0.065  sec/batch)
step 33500, loss=0.80  (1879.8  examples/sec;  0.068  sec/batch)
step 33510, loss=0.65  (1769.9  examples/sec;  0.072  sec/batch)
step 33520, loss=0.65  (1941.9  examples/sec;  0.066  sec/batch)
step 33530, loss=0.57  (1853.9  examples/sec;  0.069  sec/batch)
step 33540, loss=0.61  (1928.7  examples/sec;  0.066  sec/batch)
step 33550, loss=0.74  (1942.1  examples/sec;  0.066  sec/batch)
step 33560, loss=0.68  (1877.1  examples/sec;  0.068  sec/batch)
step 33570, loss=0.71  (1966.8  examples/sec;  0.065  sec/batch)
step 33580, loss=0.73  (1869.5  examples/sec;  0.068  sec/batch)
step 33590, loss=0.69  (1988.8  examples/sec;  0.064  sec/batch)
step 33600, loss=0.75  (1693.0  examples/sec;  0.076  sec/batch)
step 33610, loss=0.53  (1908.7  examples/sec;  0.067  sec/batch)
step 33620, loss=0.54  (1926.9  examples/sec;  0.066  sec/batch)
step 33630, loss=0.64  (1696.8  examples/sec;  0.075  sec/batch)
step 33640, loss=0.70  (1926.9  examples/sec;  0.066  sec/batch)
step 33650, loss=0.75  (1627.2  examples/sec;  0.079  sec/batch)
step 33660, loss=0.63  (1885.6  examples/sec;  0.068  sec/batch)
step 33670, loss=0.64  (1999.6  examples/sec;  0.064  sec/batch)
step 33680, loss=0.69  (1884.5  examples/sec;  0.068  sec/batch)
step 33690, loss=0.60  (1950.5  examples/sec;  0.066  sec/batch)
step 33700, loss=0.64  (1970.8  examples/sec;  0.065  sec/batch)
step 33710, loss=0.71  (1948.5  examples/sec;  0.066  sec/batch)
step 33720, loss=0.79  (1985.9  examples/sec;  0.064  sec/batch)
step 33730, loss=0.68  (1878.3  examples/sec;  0.068  sec/batch)
step 33740, loss=0.57  (1920.2  examples/sec;  0.067  sec/batch)
step 33750, loss=0.60  (1942.2  examples/sec;  0.066  sec/batch)
step 33760, loss=0.58  (1837.2  examples/sec;  0.070  sec/batch)
step 33770, loss=0.76  (1966.5  examples/sec;  0.065  sec/batch)
step 33780, loss=0.70  (1921.8  examples/sec;  0.067  sec/batch)
step 33790, loss=0.68  (1854.2  examples/sec;  0.069  sec/batch)
step 33800, loss=0.48  (1930.8  examples/sec;  0.066  sec/batch)
step 33810, loss=0.81  (1902.3  examples/sec;  0.067  sec/batch)
step 33820, loss=0.53  (1966.6  examples/sec;  0.065  sec/batch)
step 33830, loss=0.57  (1950.8  examples/sec;  0.066  sec/batch)
step 33840, loss=0.79  (1868.3  examples/sec;  0.069  sec/batch)
step 33850, loss=0.75  (1916.5  examples/sec;  0.067  sec/batch)
step 33860, loss=0.83  (1939.2  examples/sec;  0.066  sec/batch)
step 33870, loss=0.68  (1761.7  examples/sec;  0.073  sec/batch)
step 33880, loss=0.64  (1837.3  examples/sec;  0.070  sec/batch)
step 33890, loss=0.62  (1918.9  examples/sec;  0.067  sec/batch)
step 33900, loss=0.84  (1948.3  examples/sec;  0.066  sec/batch)
step 33910, loss=0.69  (1961.0  examples/sec;  0.065  sec/batch)
step 33920, loss=0.81  (1917.7  examples/sec;  0.067  sec/batch)
step 33930, loss=0.68  (1919.8  examples/sec;  0.067  sec/batch)
step 33940, loss=0.63  (1818.7  examples/sec;  0.070  sec/batch)
step 33950, loss=0.70  (1912.2  examples/sec;  0.067  sec/batch)
step 33960, loss=0.73  (1899.0  examples/sec;  0.067  sec/batch)
step 33970, loss=0.85  (1878.6  examples/sec;  0.068  sec/batch)
step 33980, loss=0.67  (2009.5  examples/sec;  0.064  sec/batch)
step 33990, loss=0.83  (1909.2  examples/sec;  0.067  sec/batch)
step 34000, loss=0.66  (1945.1  examples/sec;  0.066  sec/batch)
step 34010, loss=0.66  (1935.0  examples/sec;  0.066  sec/batch)
step 34020, loss=0.70  (1916.7  examples/sec;  0.067  sec/batch)
step 34030, loss=0.77  (1916.9  examples/sec;  0.067  sec/batch)
step 34040, loss=0.60  (1876.7  examples/sec;  0.068  sec/batch)
step 34050, loss=0.75  (1918.1  examples/sec;  0.067  sec/batch)
step 34060, loss=0.64  (1926.3  examples/sec;  0.066  sec/batch)
step 34070, loss=0.69  (1894.3  examples/sec;  0.068  sec/batch)
step 34080, loss=0.60  (1948.2  examples/sec;  0.066  sec/batch)
step 34090, loss=0.77  (1895.0  examples/sec;  0.068  sec/batch)
step 34100, loss=0.60  (1850.5  examples/sec;  0.069  sec/batch)
step 34110, loss=0.67  (1901.1  examples/sec;  0.067  sec/batch)
step 34120, loss=0.78  (1951.0  examples/sec;  0.066  sec/batch)
step 34130, loss=0.69  (1909.4  examples/sec;  0.067  sec/batch)
step 34140, loss=0.73  (1874.4  examples/sec;  0.068  sec/batch)
step 34150, loss=0.89  (1915.1  examples/sec;  0.067  sec/batch)
step 34160, loss=0.60  (1865.0  examples/sec;  0.069  sec/batch)
step 34170, loss=0.63  (1889.6  examples/sec;  0.068  sec/batch)
step 34180, loss=0.75  (1922.9  examples/sec;  0.067  sec/batch)
step 34190, loss=0.79  (1979.5  examples/sec;  0.065  sec/batch)
step 34200, loss=0.64  (1941.6  examples/sec;  0.066  sec/batch)
step 34210, loss=0.73  (1974.0  examples/sec;  0.065  sec/batch)
step 34220, loss=0.74  (1857.5  examples/sec;  0.069  sec/batch)
step 34230, loss=0.83  (1874.4  examples/sec;  0.068  sec/batch)
step 34240, loss=0.75  (1964.8  examples/sec;  0.065  sec/batch)
step 34250, loss=0.73  (1875.9  examples/sec;  0.068  sec/batch)
step 34260, loss=0.71  (1940.6  examples/sec;  0.066  sec/batch)
step 34270, loss=0.53  (1919.4  examples/sec;  0.067  sec/batch)
step 34280, loss=0.76  (1932.3  examples/sec;  0.066  sec/batch)
step 34290, loss=0.74  (1969.9  examples/sec;  0.065  sec/batch)
step 34300, loss=0.81  (1931.5  examples/sec;  0.066  sec/batch)
step 34310, loss=0.72  (1798.2  examples/sec;  0.071  sec/batch)
step 34320, loss=0.58  (1902.0  examples/sec;  0.067  sec/batch)
step 34330, loss=0.69  (1943.0  examples/sec;  0.066  sec/batch)
step 34340, loss=0.78  (1947.7  examples/sec;  0.066  sec/batch)
step 34350, loss=0.60  (1813.7  examples/sec;  0.071  sec/batch)
step 34360, loss=0.71  (1893.5  examples/sec;  0.068  sec/batch)
step 34370, loss=0.72  (1933.6  examples/sec;  0.066  sec/batch)
step 34380, loss=0.91  (1855.6  examples/sec;  0.069  sec/batch)
step 34390, loss=0.69  (1877.1  examples/sec;  0.068  sec/batch)
step 34400, loss=0.68  (1921.8  examples/sec;  0.067  sec/batch)
step 34410, loss=0.54  (1899.9  examples/sec;  0.067  sec/batch)
step 34420, loss=0.61  (1888.2  examples/sec;  0.068  sec/batch)
step 34430, loss=0.61  (1920.2  examples/sec;  0.067  sec/batch)
step 34440, loss=0.65  (1838.7  examples/sec;  0.070  sec/batch)
step 34450, loss=0.71  (1866.8  examples/sec;  0.069  sec/batch)
step 34460, loss=0.71  (1941.7  examples/sec;  0.066  sec/batch)
step 34470, loss=0.59  (1871.2  examples/sec;  0.068  sec/batch)
step 34480, loss=0.68  (1955.7  examples/sec;  0.065  sec/batch)
step 34490, loss=0.67  (1734.3  examples/sec;  0.074  sec/batch)
step 34500, loss=0.65  (1871.3  examples/sec;  0.068  sec/batch)
step 34510, loss=0.63  (1789.0  examples/sec;  0.072  sec/batch)
step 34520, loss=0.71  (1965.7  examples/sec;  0.065  sec/batch)
step 34530, loss=0.71  (1919.5  examples/sec;  0.067  sec/batch)
step 34540, loss=0.67  (1888.4  examples/sec;  0.068  sec/batch)
step 34550, loss=0.63  (1977.4  examples/sec;  0.065  sec/batch)
step 34560, loss=0.62  (1873.9  examples/sec;  0.068  sec/batch)
step 34570, loss=0.54  (1822.3  examples/sec;  0.070  sec/batch)
step 34580, loss=0.93  (1939.6  examples/sec;  0.066  sec/batch)
step 34590, loss=0.70  (1938.6  examples/sec;  0.066  sec/batch)
step 34600, loss=0.91  (1863.3  examples/sec;  0.069  sec/batch)
step 34610, loss=0.61  (2072.8  examples/sec;  0.062  sec/batch)
step 34620, loss=0.75  (1904.4  examples/sec;  0.067  sec/batch)
step 34630, loss=0.63  (1896.8  examples/sec;  0.067  sec/batch)
step 34640, loss=0.64  (1905.1  examples/sec;  0.067  sec/batch)
step 34650, loss=0.73  (1936.5  examples/sec;  0.066  sec/batch)
step 34660, loss=0.65  (1867.5  examples/sec;  0.069  sec/batch)
step 34670, loss=0.88  (1920.4  examples/sec;  0.067  sec/batch)
step 34680, loss=0.63  (1900.5  examples/sec;  0.067  sec/batch)
step 34690, loss=0.72  (1841.5  examples/sec;  0.070  sec/batch)
step 34700, loss=0.57  (1853.3  examples/sec;  0.069  sec/batch)
step 34710, loss=0.54  (1947.8  examples/sec;  0.066  sec/batch)
step 34720, loss=0.76  (1817.0  examples/sec;  0.070  sec/batch)
step 34730, loss=0.74  (1918.6  examples/sec;  0.067  sec/batch)
step 34740, loss=0.51  (1907.3  examples/sec;  0.067  sec/batch)
step 34750, loss=0.74  (1911.6  examples/sec;  0.067  sec/batch)
step 34760, loss=0.63  (1897.4  examples/sec;  0.067  sec/batch)
step 34770, loss=0.56  (1827.9  examples/sec;  0.070  sec/batch)
step 34780, loss=0.60  (1987.5  examples/sec;  0.064  sec/batch)
step 34790, loss=0.73  (1980.5  examples/sec;  0.065  sec/batch)
step 34800, loss=0.61  (1862.9  examples/sec;  0.069  sec/batch)
step 34810, loss=0.71  (1968.2  examples/sec;  0.065  sec/batch)
step 34820, loss=0.73  (1962.6  examples/sec;  0.065  sec/batch)
step 34830, loss=0.59  (1851.1  examples/sec;  0.069  sec/batch)
step 34840, loss=0.82  (1845.0  examples/sec;  0.069  sec/batch)
step 34850, loss=0.80  (2027.4  examples/sec;  0.063  sec/batch)
step 34860, loss=0.57  (2007.9  examples/sec;  0.064  sec/batch)
step 34870, loss=0.54  (1827.6  examples/sec;  0.070  sec/batch)
step 34880, loss=0.79  (1917.8  examples/sec;  0.067  sec/batch)
step 34890, loss=0.70  (1907.4  examples/sec;  0.067  sec/batch)
step 34900, loss=0.66  (1897.1  examples/sec;  0.067  sec/batch)
step 34910, loss=0.79  (1620.6  examples/sec;  0.079  sec/batch)
step 34920, loss=0.65  (1938.6  examples/sec;  0.066  sec/batch)
step 34930, loss=0.77  (1882.3  examples/sec;  0.068  sec/batch)
step 34940, loss=0.75  (1923.1  examples/sec;  0.067  sec/batch)
step 34950, loss=0.78  (1861.5  examples/sec;  0.069  sec/batch)
step 34960, loss=0.67  (1954.9  examples/sec;  0.065  sec/batch)
step 34970, loss=0.84  (1791.6  examples/sec;  0.071  sec/batch)
step 34980, loss=0.67  (1930.5  examples/sec;  0.066  sec/batch)
step 34990, loss=0.72  (1899.4  examples/sec;  0.067  sec/batch)
step 35000, loss=0.81  (2012.9  examples/sec;  0.064  sec/batch)
step 35010, loss=0.44  (1911.2  examples/sec;  0.067  sec/batch)
step 35020, loss=0.75  (1886.3  examples/sec;  0.068  sec/batch)
step 35030, loss=0.64  (1860.6  examples/sec;  0.069  sec/batch)
step 35040, loss=0.54  (1860.0  examples/sec;  0.069  sec/batch)
step 35050, loss=0.51  (1918.0  examples/sec;  0.067  sec/batch)
step 35060, loss=0.63  (1970.7  examples/sec;  0.065  sec/batch)
step 35070, loss=0.68  (1735.5  examples/sec;  0.074  sec/batch)
step 35080, loss=0.65  (1806.6  examples/sec;  0.071  sec/batch)
step 35090, loss=0.82  (1903.6  examples/sec;  0.067  sec/batch)
step 35100, loss=0.78  (1848.1  examples/sec;  0.069  sec/batch)
step 35110, loss=0.81  (1836.6  examples/sec;  0.070  sec/batch)
step 35120, loss=0.68  (1890.1  examples/sec;  0.068  sec/batch)
step 35130, loss=0.73  (1895.5  examples/sec;  0.068  sec/batch)
step 35140, loss=0.61  (1925.4  examples/sec;  0.066  sec/batch)
step 35150, loss=0.69  (1875.1  examples/sec;  0.068  sec/batch)
step 35160, loss=0.59  (1995.3  examples/sec;  0.064  sec/batch)
step 35170, loss=0.72  (1938.5  examples/sec;  0.066  sec/batch)
step 35180, loss=0.64  (1942.6  examples/sec;  0.066  sec/batch)
step 35190, loss=0.80  (1931.3  examples/sec;  0.066  sec/batch)
step 35200, loss=0.71  (1946.8  examples/sec;  0.066  sec/batch)
step 35210, loss=0.86  (1956.9  examples/sec;  0.065  sec/batch)
step 35220, loss=0.68  (1880.6  examples/sec;  0.068  sec/batch)
step 35230, loss=0.57  (1869.9  examples/sec;  0.068  sec/batch)
step 35240, loss=0.69  (1897.7  examples/sec;  0.067  sec/batch)
step 35250, loss=0.64  (1902.9  examples/sec;  0.067  sec/batch)
step 35260, loss=0.69  (1915.5  examples/sec;  0.067  sec/batch)
step 35270, loss=0.69  (1886.8  examples/sec;  0.068  sec/batch)
step 35280, loss=0.68  (1953.2  examples/sec;  0.066  sec/batch)
step 35290, loss=0.71  (1896.5  examples/sec;  0.067  sec/batch)
step 35300, loss=0.59  (1955.4  examples/sec;  0.065  sec/batch)
step 35310, loss=0.84  (1882.9  examples/sec;  0.068  sec/batch)
step 35320, loss=0.59  (1976.0  examples/sec;  0.065  sec/batch)
step 35330, loss=0.68  (1926.1  examples/sec;  0.066  sec/batch)
step 35340, loss=0.60  (1920.0  examples/sec;  0.067  sec/batch)
step 35350, loss=0.72  (1985.2  examples/sec;  0.064  sec/batch)
step 35360, loss=0.65  (1885.4  examples/sec;  0.068  sec/batch)
step 35370, loss=0.70  (1905.5  examples/sec;  0.067  sec/batch)
step 35380, loss=0.67  (2025.8  examples/sec;  0.063  sec/batch)
step 35390, loss=0.65  (1673.5  examples/sec;  0.076  sec/batch)
step 35400, loss=0.64  (1748.8  examples/sec;  0.073  sec/batch)
step 35410, loss=0.81  (1978.3  examples/sec;  0.065  sec/batch)
step 35420, loss=0.79  (1914.4  examples/sec;  0.067  sec/batch)
step 35430, loss=0.62  (1839.6  examples/sec;  0.070  sec/batch)
step 35440, loss=0.68  (1907.7  examples/sec;  0.067  sec/batch)
step 35450, loss=0.73  (1819.3  examples/sec;  0.070  sec/batch)
step 35460, loss=0.68  (1902.3  examples/sec;  0.067  sec/batch)
step 35470, loss=0.76  (1848.5  examples/sec;  0.069  sec/batch)
step 35480, loss=0.57  (1920.9  examples/sec;  0.067  sec/batch)
step 35490, loss=0.71  (1906.0  examples/sec;  0.067  sec/batch)
step 35500, loss=0.61  (1974.3  examples/sec;  0.065  sec/batch)
step 35510, loss=0.85  (1896.4  examples/sec;  0.067  sec/batch)
step 35520, loss=0.78  (1929.1  examples/sec;  0.066  sec/batch)
step 35530, loss=0.61  (1903.9  examples/sec;  0.067  sec/batch)
step 35540, loss=0.74  (1976.2  examples/sec;  0.065  sec/batch)
step 35550, loss=1.01  (1873.9  examples/sec;  0.068  sec/batch)
step 35560, loss=0.66  (1871.0  examples/sec;  0.068  sec/batch)
step 35570, loss=0.61  (1877.3  examples/sec;  0.068  sec/batch)
step 35580, loss=0.56  (2033.8  examples/sec;  0.063  sec/batch)
step 35590, loss=0.57  (1664.2  examples/sec;  0.077  sec/batch)
step 35600, loss=0.62  (1878.9  examples/sec;  0.068  sec/batch)
step 35610, loss=0.63  (1953.5  examples/sec;  0.066  sec/batch)
step 35620, loss=0.53  (1822.7  examples/sec;  0.070  sec/batch)
step 35630, loss=0.80  (1683.0  examples/sec;  0.076  sec/batch)
step 35640, loss=0.64  (1929.0  examples/sec;  0.066  sec/batch)
step 35650, loss=0.71  (1780.3  examples/sec;  0.072  sec/batch)
step 35660, loss=0.76  (1899.3  examples/sec;  0.067  sec/batch)
step 35670, loss=0.61  (1862.1  examples/sec;  0.069  sec/batch)
step 35680, loss=0.77  (1904.7  examples/sec;  0.067  sec/batch)
step 35690, loss=0.65  (1851.2  examples/sec;  0.069  sec/batch)
step 35700, loss=0.66  (1897.1  examples/sec;  0.067  sec/batch)
step 35710, loss=0.73  (1855.7  examples/sec;  0.069  sec/batch)
step 35720, loss=0.77  (2033.3  examples/sec;  0.063  sec/batch)
step 35730, loss=0.65  (1887.5  examples/sec;  0.068  sec/batch)
step 35740, loss=0.77  (1772.6  examples/sec;  0.072  sec/batch)
step 35750, loss=0.69  (1874.0  examples/sec;  0.068  sec/batch)
step 35760, loss=0.76  (1889.8  examples/sec;  0.068  sec/batch)
step 35770, loss=0.69  (1894.3  examples/sec;  0.068  sec/batch)
step 35780, loss=0.66  (1971.2  examples/sec;  0.065  sec/batch)
step 35790, loss=0.48  (1906.2  examples/sec;  0.067  sec/batch)
step 35800, loss=0.90  (1922.8  examples/sec;  0.067  sec/batch)
step 35810, loss=0.62  (1927.3  examples/sec;  0.066  sec/batch)
step 35820, loss=0.79  (1772.5  examples/sec;  0.072  sec/batch)
step 35830, loss=0.63  (1872.9  examples/sec;  0.068  sec/batch)
step 35840, loss=0.71  (1930.5  examples/sec;  0.066  sec/batch)
step 35850, loss=0.62  (1923.3  examples/sec;  0.067  sec/batch)
step 35860, loss=0.75  (1649.3  examples/sec;  0.078  sec/batch)
step 35870, loss=0.68  (1934.8  examples/sec;  0.066  sec/batch)
step 35880, loss=0.66  (2030.6  examples/sec;  0.063  sec/batch)
step 35890, loss=0.51  (1883.1  examples/sec;  0.068  sec/batch)
step 35900, loss=0.68  (1860.2  examples/sec;  0.069  sec/batch)
step 35910, loss=0.69  (1909.0  examples/sec;  0.067  sec/batch)
step 35920, loss=0.77  (1940.4  examples/sec;  0.066  sec/batch)
step 35930, loss=0.87  (1857.1  examples/sec;  0.069  sec/batch)
step 35940, loss=0.67  (1863.1  examples/sec;  0.069  sec/batch)
step 35950, loss=0.84  (1851.8  examples/sec;  0.069  sec/batch)
step 35960, loss=0.70  (1901.4  examples/sec;  0.067  sec/batch)
step 35970, loss=0.61  (1827.5  examples/sec;  0.070  sec/batch)
step 35980, loss=0.60  (1840.0  examples/sec;  0.070  sec/batch)
step 35990, loss=0.64  (1822.2  examples/sec;  0.070  sec/batch)
step 36000, loss=0.65  (1994.0  examples/sec;  0.064  sec/batch)
step 36010, loss=0.69  (1830.3  examples/sec;  0.070  sec/batch)
step 36020, loss=0.77  (1841.6  examples/sec;  0.070  sec/batch)
step 36030, loss=0.57  (1885.2  examples/sec;  0.068  sec/batch)
step 36040, loss=0.63  (1891.0  examples/sec;  0.068  sec/batch)
step 36050, loss=0.69  (1937.5  examples/sec;  0.066  sec/batch)
step 36060, loss=0.56  (1919.6  examples/sec;  0.067  sec/batch)
step 36070, loss=0.83  (1952.2  examples/sec;  0.066  sec/batch)
step 36080, loss=0.65  (1916.9  examples/sec;  0.067  sec/batch)
step 36090, loss=0.74  (1873.7  examples/sec;  0.068  sec/batch)
step 36100, loss=0.78  (1899.3  examples/sec;  0.067  sec/batch)
step 36110, loss=0.68  (1886.3  examples/sec;  0.068  sec/batch)
step 36120, loss=0.71  (1787.6  examples/sec;  0.072  sec/batch)
step 36130, loss=0.69  (1877.2  examples/sec;  0.068  sec/batch)
step 36140, loss=0.56  (1869.3  examples/sec;  0.068  sec/batch)
step 36150, loss=0.63  (1852.3  examples/sec;  0.069  sec/batch)
step 36160, loss=0.55  (1844.2  examples/sec;  0.069  sec/batch)
step 36170, loss=0.70  (1895.8  examples/sec;  0.068  sec/batch)
step 36180, loss=0.80  (1889.1  examples/sec;  0.068  sec/batch)
step 36190, loss=0.77  (1869.3  examples/sec;  0.068  sec/batch)
step 36200, loss=0.73  (1893.1  examples/sec;  0.068  sec/batch)
step 36210, loss=0.65  (1911.3  examples/sec;  0.067  sec/batch)
step 36220, loss=0.67  (1924.3  examples/sec;  0.067  sec/batch)
step 36230, loss=0.63  (1885.8  examples/sec;  0.068  sec/batch)
step 36240, loss=0.63  (1941.6  examples/sec;  0.066  sec/batch)
step 36250, loss=0.63  (1883.5  examples/sec;  0.068  sec/batch)
step 36260, loss=0.71  (1973.7  examples/sec;  0.065  sec/batch)
step 36270, loss=0.72  (1931.0  examples/sec;  0.066  sec/batch)
step 36280, loss=0.59  (1874.0  examples/sec;  0.068  sec/batch)
step 36290, loss=0.60  (1893.6  examples/sec;  0.068  sec/batch)
step 36300, loss=0.59  (1819.4  examples/sec;  0.070  sec/batch)
step 36310, loss=0.66  (1920.3  examples/sec;  0.067  sec/batch)
step 36320, loss=0.64  (1915.3  examples/sec;  0.067  sec/batch)
step 36330, loss=0.76  (1904.8  examples/sec;  0.067  sec/batch)
step 36340, loss=0.72  (1930.0  examples/sec;  0.066  sec/batch)
step 36350, loss=0.73  (1955.7  examples/sec;  0.065  sec/batch)
step 36360, loss=0.55  (1860.1  examples/sec;  0.069  sec/batch)
step 36370, loss=0.62  (1995.8  examples/sec;  0.064  sec/batch)
step 36380, loss=0.76  (1938.8  examples/sec;  0.066  sec/batch)
step 36390, loss=0.59  (1947.3  examples/sec;  0.066  sec/batch)
step 36400, loss=0.73  (1915.1  examples/sec;  0.067  sec/batch)
step 36410, loss=0.64  (1823.3  examples/sec;  0.070  sec/batch)
step 36420, loss=0.62  (1881.9  examples/sec;  0.068  sec/batch)
step 36430, loss=0.52  (1874.5  examples/sec;  0.068  sec/batch)
step 36440, loss=0.68  (1921.8  examples/sec;  0.067  sec/batch)
step 36450, loss=0.63  (1932.7  examples/sec;  0.066  sec/batch)
step 36460, loss=0.68  (1813.2  examples/sec;  0.071  sec/batch)
step 36470, loss=0.70  (1191.6  examples/sec;  0.107  sec/batch)
step 36480, loss=0.65  (1895.5  examples/sec;  0.068  sec/batch)
step 36490, loss=0.73  (1906.9  examples/sec;  0.067  sec/batch)
step 36500, loss=0.79  (1890.7  examples/sec;  0.068  sec/batch)
step 36510, loss=0.60  (1896.5  examples/sec;  0.067  sec/batch)
step 36520, loss=0.73  (1897.2  examples/sec;  0.067  sec/batch)
step 36530, loss=0.64  (1896.3  examples/sec;  0.067  sec/batch)
step 36540, loss=0.56  (1912.1  examples/sec;  0.067  sec/batch)
step 36550, loss=0.65  (1894.5  examples/sec;  0.068  sec/batch)
step 36560, loss=0.76  (1870.4  examples/sec;  0.068  sec/batch)
step 36570, loss=0.59  (1901.0  examples/sec;  0.067  sec/batch)
step 36580, loss=0.97  (1882.5  examples/sec;  0.068  sec/batch)
step 36590, loss=0.76  (1919.5  examples/sec;  0.067  sec/batch)
step 36600, loss=0.71  (1819.4  examples/sec;  0.070  sec/batch)
step 36610, loss=0.64  (1820.7  examples/sec;  0.070  sec/batch)
step 36620, loss=0.71  (1851.7  examples/sec;  0.069  sec/batch)
step 36630, loss=0.61  (1832.4  examples/sec;  0.070  sec/batch)
step 36640, loss=0.87  (1889.7  examples/sec;  0.068  sec/batch)
step 36650, loss=0.81  (1901.2  examples/sec;  0.067  sec/batch)
step 36660, loss=0.71  (1963.0  examples/sec;  0.065  sec/batch)
step 36670, loss=0.64  (1728.3  examples/sec;  0.074  sec/batch)
step 36680, loss=0.64  (1969.5  examples/sec;  0.065  sec/batch)
step 36690, loss=0.65  (1874.4  examples/sec;  0.068  sec/batch)
step 36700, loss=0.68  (1883.8  examples/sec;  0.068  sec/batch)
step 36710, loss=0.71  (1838.9  examples/sec;  0.070  sec/batch)
step 36720, loss=0.62  (1935.3  examples/sec;  0.066  sec/batch)
step 36730, loss=0.73  (1900.5  examples/sec;  0.067  sec/batch)
step 36740, loss=0.64  (1884.3  examples/sec;  0.068  sec/batch)
step 36750, loss=0.67  (1697.5  examples/sec;  0.075  sec/batch)
step 36760, loss=0.64  (1914.8  examples/sec;  0.067  sec/batch)
step 36770, loss=0.69  (1916.0  examples/sec;  0.067  sec/batch)
step 36780, loss=0.60  (1837.1  examples/sec;  0.070  sec/batch)
step 36790, loss=0.68  (1876.5  examples/sec;  0.068  sec/batch)
step 36800, loss=0.62  (1929.1  examples/sec;  0.066  sec/batch)
step 36810, loss=0.63  (1915.6  examples/sec;  0.067  sec/batch)
step 36820, loss=0.62  (1939.7  examples/sec;  0.066  sec/batch)
step 36830, loss=0.67  (1841.7  examples/sec;  0.069  sec/batch)
step 36840, loss=0.62  (1796.0  examples/sec;  0.071  sec/batch)
step 36850, loss=0.84  (1945.5  examples/sec;  0.066  sec/batch)
step 36860, loss=0.83  (1970.7  examples/sec;  0.065  sec/batch)
step 36870, loss=0.67  (1933.9  examples/sec;  0.066  sec/batch)
step 36880, loss=0.55  (1933.4  examples/sec;  0.066  sec/batch)
step 36890, loss=0.79  (1926.8  examples/sec;  0.066  sec/batch)
step 36900, loss=0.66  (1647.6  examples/sec;  0.078  sec/batch)
step 36910, loss=0.59  (1851.0  examples/sec;  0.069  sec/batch)
step 36920, loss=0.66  (1883.0  examples/sec;  0.068  sec/batch)
step 36930, loss=0.69  (1876.9  examples/sec;  0.068  sec/batch)
step 36940, loss=0.76  (1901.5  examples/sec;  0.067  sec/batch)
step 36950, loss=0.77  (1834.9  examples/sec;  0.070  sec/batch)
step 36960, loss=0.64  (1871.5  examples/sec;  0.068  sec/batch)
step 36970, loss=0.56  (1844.2  examples/sec;  0.069  sec/batch)
step 36980, loss=0.69  (1834.3  examples/sec;  0.070  sec/batch)
step 36990, loss=0.77  (1972.9  examples/sec;  0.065  sec/batch)
step 37000, loss=0.90  (1816.0  examples/sec;  0.070  sec/batch)
step 37010, loss=0.64  (1819.3  examples/sec;  0.070  sec/batch)
step 37020, loss=0.69  (1812.4  examples/sec;  0.071  sec/batch)
step 37030, loss=0.57  (1930.5  examples/sec;  0.066  sec/batch)
step 37040, loss=0.57  (1934.8  examples/sec;  0.066  sec/batch)
step 37050, loss=0.62  (1858.8  examples/sec;  0.069  sec/batch)
step 37060, loss=0.55  (1881.4  examples/sec;  0.068  sec/batch)
step 37070, loss=0.75  (1922.0  examples/sec;  0.067  sec/batch)
step 37080, loss=0.69  (1884.4  examples/sec;  0.068  sec/batch)
step 37090, loss=0.64  (1882.4  examples/sec;  0.068  sec/batch)
step 37100, loss=0.63  (1880.0  examples/sec;  0.068  sec/batch)
step 37110, loss=0.67  (1924.8  examples/sec;  0.066  sec/batch)
step 37120, loss=0.64  (1887.0  examples/sec;  0.068  sec/batch)
step 37130, loss=0.65  (1921.7  examples/sec;  0.067  sec/batch)
step 37140, loss=0.69  (1939.7  examples/sec;  0.066  sec/batch)
step 37150, loss=0.54  (1923.6  examples/sec;  0.067  sec/batch)
step 37160, loss=0.61  (1850.5  examples/sec;  0.069  sec/batch)
step 37170, loss=0.72  (1848.6  examples/sec;  0.069  sec/batch)
step 37180, loss=0.65  (2010.9  examples/sec;  0.064  sec/batch)
step 37190, loss=0.77  (1887.4  examples/sec;  0.068  sec/batch)
step 37200, loss=0.70  (1964.3  examples/sec;  0.065  sec/batch)
step 37210, loss=0.75  (1853.8  examples/sec;  0.069  sec/batch)
step 37220, loss=0.67  (1974.4  examples/sec;  0.065  sec/batch)
step 37230, loss=0.67  (1857.9  examples/sec;  0.069  sec/batch)
step 37240, loss=0.72  (1875.5  examples/sec;  0.068  sec/batch)
step 37250, loss=0.60  (1928.8  examples/sec;  0.066  sec/batch)
step 37260, loss=0.65  (1896.6  examples/sec;  0.067  sec/batch)
step 37270, loss=0.65  (1922.3  examples/sec;  0.067  sec/batch)
step 37280, loss=0.55  (1980.6  examples/sec;  0.065  sec/batch)
step 37290, loss=0.59  (1941.0  examples/sec;  0.066  sec/batch)
step 37300, loss=0.75  (1873.6  examples/sec;  0.068  sec/batch)
step 37310, loss=0.81  (1799.3  examples/sec;  0.071  sec/batch)
step 37320, loss=0.52  (2004.0  examples/sec;  0.064  sec/batch)
step 37330, loss=0.53  (1875.2  examples/sec;  0.068  sec/batch)
step 37340, loss=0.69  (1887.8  examples/sec;  0.068  sec/batch)
step 37350, loss=0.62  (1891.3  examples/sec;  0.068  sec/batch)
step 37360, loss=0.64  (1861.5  examples/sec;  0.069  sec/batch)
step 37370, loss=0.68  (1877.4  examples/sec;  0.068  sec/batch)
step 37380, loss=0.67  (1909.6  examples/sec;  0.067  sec/batch)
step 37390, loss=0.64  (1819.0  examples/sec;  0.070  sec/batch)
step 37400, loss=0.58  (1923.0  examples/sec;  0.067  sec/batch)
step 37410, loss=0.80  (1967.7  examples/sec;  0.065  sec/batch)
step 37420, loss=0.64  (1860.0  examples/sec;  0.069  sec/batch)
step 37430, loss=0.69  (1981.1  examples/sec;  0.065  sec/batch)
step 37440, loss=0.70  (1916.7  examples/sec;  0.067  sec/batch)
step 37450, loss=0.68  (1938.6  examples/sec;  0.066  sec/batch)
step 37460, loss=0.51  (1934.7  examples/sec;  0.066  sec/batch)
step 37470, loss=0.61  (1835.4  examples/sec;  0.070  sec/batch)
step 37480, loss=0.74  (1825.5  examples/sec;  0.070  sec/batch)
step 37490, loss=0.87  (1809.3  examples/sec;  0.071  sec/batch)
step 37500, loss=0.70  (1909.9  examples/sec;  0.067  sec/batch)
step 37510, loss=0.66  (1793.9  examples/sec;  0.071  sec/batch)
step 37520, loss=0.61  (1873.3  examples/sec;  0.068  sec/batch)
step 37530, loss=0.82  (1907.5  examples/sec;  0.067  sec/batch)
step 37540, loss=0.69  (1902.6  examples/sec;  0.067  sec/batch)
step 37550, loss=0.67  (1807.0  examples/sec;  0.071  sec/batch)
step 37560, loss=0.69  (1945.3  examples/sec;  0.066  sec/batch)
step 37570, loss=0.67  (1892.1  examples/sec;  0.068  sec/batch)
step 37580, loss=0.80  (1922.2  examples/sec;  0.067  sec/batch)
step 37590, loss=0.59  (1911.9  examples/sec;  0.067  sec/batch)
step 37600, loss=0.71  (1930.1  examples/sec;  0.066  sec/batch)
step 37610, loss=0.76  (1948.2  examples/sec;  0.066  sec/batch)
step 37620, loss=0.75  (1980.7  examples/sec;  0.065  sec/batch)
step 37630, loss=0.66  (1909.5  examples/sec;  0.067  sec/batch)
step 37640, loss=0.74  (1809.7  examples/sec;  0.071  sec/batch)
step 37650, loss=0.64  (1849.6  examples/sec;  0.069  sec/batch)
step 37660, loss=0.71  (1862.4  examples/sec;  0.069  sec/batch)
step 37670, loss=0.85  (1879.9  examples/sec;  0.068  sec/batch)
step 37680, loss=0.65  (1945.1  examples/sec;  0.066  sec/batch)
step 37690, loss=0.75  (1955.5  examples/sec;  0.065  sec/batch)
step 37700, loss=0.68  (1797.8  examples/sec;  0.071  sec/batch)
step 37710, loss=0.75  (1878.8  examples/sec;  0.068  sec/batch)
step 37720, loss=0.47  (1900.1  examples/sec;  0.067  sec/batch)
step 37730, loss=0.60  (1913.0  examples/sec;  0.067  sec/batch)
step 37740, loss=0.57  (1914.5  examples/sec;  0.067  sec/batch)
step 37750, loss=0.92  (1886.5  examples/sec;  0.068  sec/batch)
step 37760, loss=0.58  (1935.6  examples/sec;  0.066  sec/batch)
step 37770, loss=0.78  (1880.3  examples/sec;  0.068  sec/batch)
step 37780, loss=0.60  (1920.6  examples/sec;  0.067  sec/batch)
step 37790, loss=0.67  (1903.7  examples/sec;  0.067  sec/batch)
step 37800, loss=0.51  (1968.4  examples/sec;  0.065  sec/batch)
step 37810, loss=0.64  (1726.2  examples/sec;  0.074  sec/batch)
step 37820, loss=0.74  (1913.6  examples/sec;  0.067  sec/batch)
step 37830, loss=0.68  (1931.4  examples/sec;  0.066  sec/batch)
step 37840, loss=0.63  (1889.4  examples/sec;  0.068  sec/batch)
step 37850, loss=0.71  (1960.0  examples/sec;  0.065  sec/batch)
step 37860, loss=0.61  (1901.1  examples/sec;  0.067  sec/batch)
step 37870, loss=0.72  (1912.1  examples/sec;  0.067  sec/batch)
step 37880, loss=0.59  (1815.6  examples/sec;  0.071  sec/batch)
step 37890, loss=0.87  (1884.2  examples/sec;  0.068  sec/batch)
step 37900, loss=0.67  (1990.5  examples/sec;  0.064  sec/batch)
step 37910, loss=0.57  (1859.8  examples/sec;  0.069  sec/batch)
step 37920, loss=0.79  (1655.0  examples/sec;  0.077  sec/batch)
step 37930, loss=0.67  (1938.7  examples/sec;  0.066  sec/batch)
step 37940, loss=0.68  (1999.3  examples/sec;  0.064  sec/batch)
step 37950, loss=0.63  (1977.1  examples/sec;  0.065  sec/batch)
step 37960, loss=0.75  (1797.5  examples/sec;  0.071  sec/batch)
step 37970, loss=0.67  (1870.9  examples/sec;  0.068  sec/batch)
step 37980, loss=0.57  (1759.8  examples/sec;  0.073  sec/batch)
step 37990, loss=0.69  (1896.5  examples/sec;  0.067  sec/batch)
step 38000, loss=0.54  (1910.4  examples/sec;  0.067  sec/batch)
step 38010, loss=0.77  (1849.6  examples/sec;  0.069  sec/batch)
step 38020, loss=0.79  (1941.4  examples/sec;  0.066  sec/batch)
step 38030, loss=0.62  (1846.6  examples/sec;  0.069  sec/batch)
step 38040, loss=0.59  (1952.7  examples/sec;  0.066  sec/batch)
step 38050, loss=0.79  (1872.5  examples/sec;  0.068  sec/batch)
step 38060, loss=0.70  (1870.7  examples/sec;  0.068  sec/batch)
step 38070, loss=0.80  (1918.6  examples/sec;  0.067  sec/batch)
step 38080, loss=0.59  (1856.4  examples/sec;  0.069  sec/batch)
step 38090, loss=0.78  (1930.0  examples/sec;  0.066  sec/batch)
step 38100, loss=0.65  (1944.4  examples/sec;  0.066  sec/batch)
step 38110, loss=0.81  (1865.6  examples/sec;  0.069  sec/batch)
step 38120, loss=0.88  (1861.6  examples/sec;  0.069  sec/batch)
step 38130, loss=0.56  (1827.5  examples/sec;  0.070  sec/batch)
step 38140, loss=0.60  (1973.8  examples/sec;  0.065  sec/batch)
step 38150, loss=0.67  (1907.1  examples/sec;  0.067  sec/batch)
step 38160, loss=0.75  (1686.5  examples/sec;  0.076  sec/batch)
step 38170, loss=0.68  (2020.5  examples/sec;  0.063  sec/batch)
step 38180, loss=0.68  (1890.1  examples/sec;  0.068  sec/batch)
step 38190, loss=0.65  (1916.0  examples/sec;  0.067  sec/batch)
step 38200, loss=0.76  (1920.6  examples/sec;  0.067  sec/batch)
step 38210, loss=0.67  (1967.3  examples/sec;  0.065  sec/batch)
step 38220, loss=0.58  (1941.9  examples/sec;  0.066  sec/batch)
step 38230, loss=0.64  (1855.9  examples/sec;  0.069  sec/batch)
step 38240, loss=0.77  (1887.7  examples/sec;  0.068  sec/batch)
step 38250, loss=0.61  (1955.9  examples/sec;  0.065  sec/batch)
step 38260, loss=0.63  (1833.5  examples/sec;  0.070  sec/batch)
step 38270, loss=0.57  (1884.6  examples/sec;  0.068  sec/batch)
step 38280, loss=0.55  (1857.4  examples/sec;  0.069  sec/batch)
step 38290, loss=0.69  (1976.0  examples/sec;  0.065  sec/batch)
step 38300, loss=0.85  (1969.2  examples/sec;  0.065  sec/batch)
step 38310, loss=0.53  (1868.8  examples/sec;  0.068  sec/batch)
step 38320, loss=0.75  (1739.4  examples/sec;  0.074  sec/batch)
step 38330, loss=0.68  (1886.6  examples/sec;  0.068  sec/batch)
step 38340, loss=0.47  (1911.2  examples/sec;  0.067  sec/batch)
step 38350, loss=0.76  (1836.3  examples/sec;  0.070  sec/batch)
step 38360, loss=0.58  (1909.0  examples/sec;  0.067  sec/batch)
step 38370, loss=0.60  (1875.3  examples/sec;  0.068  sec/batch)
step 38380, loss=0.73  (1880.2  examples/sec;  0.068  sec/batch)
step 38390, loss=0.66  (1890.7  examples/sec;  0.068  sec/batch)
step 38400, loss=0.66  (1878.4  examples/sec;  0.068  sec/batch)
step 38410, loss=0.55  (1798.9  examples/sec;  0.071  sec/batch)
step 38420, loss=0.68  (1844.2  examples/sec;  0.069  sec/batch)
step 38430, loss=0.71  (1836.9  examples/sec;  0.070  sec/batch)
step 38440, loss=0.90  (1908.3  examples/sec;  0.067  sec/batch)
step 38450, loss=0.56  (1920.8  examples/sec;  0.067  sec/batch)
step 38460, loss=0.46  (1829.3  examples/sec;  0.070  sec/batch)
step 38470, loss=0.62  (1979.8  examples/sec;  0.065  sec/batch)
step 38480, loss=0.72  (1952.6  examples/sec;  0.066  sec/batch)
step 38490, loss=0.62  (1984.3  examples/sec;  0.065  sec/batch)
step 38500, loss=0.54  (1837.9  examples/sec;  0.070  sec/batch)
step 38510, loss=0.75  (1869.2  examples/sec;  0.068  sec/batch)
step 38520, loss=0.65  (1951.6  examples/sec;  0.066  sec/batch)
step 38530, loss=0.64  (1944.8  examples/sec;  0.066  sec/batch)
step 38540, loss=0.50  (1860.6  examples/sec;  0.069  sec/batch)
step 38550, loss=0.65  (1825.8  examples/sec;  0.070  sec/batch)
step 38560, loss=0.60  (1995.9  examples/sec;  0.064  sec/batch)
step 38570, loss=0.71  (1976.6  examples/sec;  0.065  sec/batch)
step 38580, loss=0.64  (1930.2  examples/sec;  0.066  sec/batch)
step 38590, loss=0.81  (1868.9  examples/sec;  0.068  sec/batch)
step 38600, loss=0.72  (1955.0  examples/sec;  0.065  sec/batch)
step 38610, loss=0.62  (1942.1  examples/sec;  0.066  sec/batch)
step 38620, loss=0.68  (1859.4  examples/sec;  0.069  sec/batch)
step 38630, loss=0.79  (1916.7  examples/sec;  0.067  sec/batch)
step 38640, loss=0.72  (1814.3  examples/sec;  0.071  sec/batch)
step 38650, loss=0.72  (1869.9  examples/sec;  0.068  sec/batch)
step 38660, loss=0.67  (1881.3  examples/sec;  0.068  sec/batch)
step 38670, loss=0.64  (1864.5  examples/sec;  0.069  sec/batch)
step 38680, loss=0.71  (1882.9  examples/sec;  0.068  sec/batch)
step 38690, loss=0.57  (1922.0  examples/sec;  0.067  sec/batch)
step 38700, loss=0.62  (1915.0  examples/sec;  0.067  sec/batch)
step 38710, loss=0.66  (1849.6  examples/sec;  0.069  sec/batch)
step 38720, loss=0.63  (1923.7  examples/sec;  0.067  sec/batch)
step 38730, loss=0.63  (1836.2  examples/sec;  0.070  sec/batch)
step 38740, loss=0.67  (1947.2  examples/sec;  0.066  sec/batch)
step 38750, loss=0.64  (1912.0  examples/sec;  0.067  sec/batch)
step 38760, loss=0.59  (2011.7  examples/sec;  0.064  sec/batch)
step 38770, loss=0.56  (1883.8  examples/sec;  0.068  sec/batch)
step 38780, loss=0.65  (1893.8  examples/sec;  0.068  sec/batch)
step 38790, loss=0.59  (1823.5  examples/sec;  0.070  sec/batch)
step 38800, loss=0.79  (2021.8  examples/sec;  0.063  sec/batch)
step 38810, loss=0.58  (1945.3  examples/sec;  0.066  sec/batch)
step 38820, loss=0.58  (1957.9  examples/sec;  0.065  sec/batch)
step 38830, loss=0.81  (1612.7  examples/sec;  0.079  sec/batch)
step 38840, loss=0.70  (1931.2  examples/sec;  0.066  sec/batch)
step 38850, loss=0.68  (1883.6  examples/sec;  0.068  sec/batch)
step 38860, loss=0.60  (1872.8  examples/sec;  0.068  sec/batch)
step 38870, loss=0.58  (1916.9  examples/sec;  0.067  sec/batch)
step 38880, loss=0.61  (1901.2  examples/sec;  0.067  sec/batch)
step 38890, loss=0.67  (1940.7  examples/sec;  0.066  sec/batch)
step 38900, loss=0.69  (1820.1  examples/sec;  0.070  sec/batch)
step 38910, loss=0.61  (1858.5  examples/sec;  0.069  sec/batch)
step 38920, loss=0.81  (1864.2  examples/sec;  0.069  sec/batch)
step 38930, loss=0.52  (1771.4  examples/sec;  0.072  sec/batch)
step 38940, loss=0.56  (1848.7  examples/sec;  0.069  sec/batch)
step 38950, loss=0.82  (1859.6  examples/sec;  0.069  sec/batch)
step 38960, loss=0.62  (1887.2  examples/sec;  0.068  sec/batch)
step 38970, loss=0.57  (1839.9  examples/sec;  0.070  sec/batch)
step 38980, loss=0.83  (1894.3  examples/sec;  0.068  sec/batch)
step 38990, loss=0.63  (1875.1  examples/sec;  0.068  sec/batch)
step 39000, loss=0.77  (1938.3  examples/sec;  0.066  sec/batch)
step 39010, loss=0.61  (1870.9  examples/sec;  0.068  sec/batch)
step 39020, loss=0.58  (1922.2  examples/sec;  0.067  sec/batch)
step 39030, loss=0.73  (1944.5  examples/sec;  0.066  sec/batch)
step 39040, loss=0.73  (1826.4  examples/sec;  0.070  sec/batch)
step 39050, loss=0.91  (1702.0  examples/sec;  0.075  sec/batch)
step 39060, loss=0.63  (1896.1  examples/sec;  0.068  sec/batch)
step 39070, loss=0.70  (1855.8  examples/sec;  0.069  sec/batch)
step 39080, loss=0.70  (1834.8  examples/sec;  0.070  sec/batch)
step 39090, loss=0.71  (1896.0  examples/sec;  0.068  sec/batch)
step 39100, loss=0.59  (1888.0  examples/sec;  0.068  sec/batch)
step 39110, loss=0.60  (1865.5  examples/sec;  0.069  sec/batch)
step 39120, loss=0.69  (1796.6  examples/sec;  0.071  sec/batch)
step 39130, loss=0.65  (1425.4  examples/sec;  0.090  sec/batch)
step 39140, loss=0.64  (1157.4  examples/sec;  0.111  sec/batch)
step 39150, loss=0.74  (1883.7  examples/sec;  0.068  sec/batch)
step 39160, loss=0.67  (1918.1  examples/sec;  0.067  sec/batch)
step 39170, loss=0.61  (1815.3  examples/sec;  0.071  sec/batch)
step 39180, loss=0.70  (1931.0  examples/sec;  0.066  sec/batch)
step 39190, loss=0.72  (1820.7  examples/sec;  0.070  sec/batch)
step 39200, loss=0.67  (1858.0  examples/sec;  0.069  sec/batch)
step 39210, loss=0.50  (1690.7  examples/sec;  0.076  sec/batch)
step 39220, loss=0.73  (1979.4  examples/sec;  0.065  sec/batch)
step 39230, loss=0.66  (1944.7  examples/sec;  0.066  sec/batch)
step 39240, loss=0.80  (1665.0  examples/sec;  0.077  sec/batch)
step 39250, loss=0.61  (1862.5  examples/sec;  0.069  sec/batch)
step 39260, loss=0.70  (1945.9  examples/sec;  0.066  sec/batch)
step 39270, loss=0.76  (1895.8  examples/sec;  0.068  sec/batch)
step 39280, loss=0.56  (1878.5  examples/sec;  0.068  sec/batch)
step 39290, loss=0.76  (1881.1  examples/sec;  0.068  sec/batch)
step 39300, loss=0.68  (1914.1  examples/sec;  0.067  sec/batch)
step 39310, loss=0.73  (1915.7  examples/sec;  0.067  sec/batch)
step 39320, loss=0.60  (2034.9  examples/sec;  0.063  sec/batch)
step 39330, loss=0.56  (1926.7  examples/sec;  0.066  sec/batch)
step 39340, loss=0.67  (1879.4  examples/sec;  0.068  sec/batch)
step 39350, loss=0.78  (1974.5  examples/sec;  0.065  sec/batch)
step 39360, loss=0.75  (1921.2  examples/sec;  0.067  sec/batch)
step 39370, loss=0.69  (1900.9  examples/sec;  0.067  sec/batch)
step 39380, loss=0.59  (1897.6  examples/sec;  0.067  sec/batch)
step 39390, loss=0.61  (1904.9  examples/sec;  0.067  sec/batch)
step 39400, loss=0.52  (1980.3  examples/sec;  0.065  sec/batch)
step 39410, loss=0.65  (1934.8  examples/sec;  0.066  sec/batch)
step 39420, loss=0.83  (1889.1  examples/sec;  0.068  sec/batch)
step 39430, loss=0.68  (1898.6  examples/sec;  0.067  sec/batch)
step 39440, loss=0.69  (1856.6  examples/sec;  0.069  sec/batch)
step 39450, loss=0.63  (1946.9  examples/sec;  0.066  sec/batch)
step 39460, loss=0.68  (1903.3  examples/sec;  0.067  sec/batch)
step 39470, loss=0.76  (1847.6  examples/sec;  0.069  sec/batch)
step 39480, loss=0.60  (1892.3  examples/sec;  0.068  sec/batch)
step 39490, loss=0.73  (1852.3  examples/sec;  0.069  sec/batch)
step 39500, loss=0.69  (1838.2  examples/sec;  0.070  sec/batch)
step 39510, loss=0.75  (1902.4  examples/sec;  0.067  sec/batch)
step 39520, loss=0.80  (1951.0  examples/sec;  0.066  sec/batch)
step 39530, loss=0.63  (1767.7  examples/sec;  0.072  sec/batch)
step 39540, loss=0.65  (1920.2  examples/sec;  0.067  sec/batch)
step 39550, loss=0.58  (1909.7  examples/sec;  0.067  sec/batch)
step 39560, loss=0.60  (1945.1  examples/sec;  0.066  sec/batch)
step 39570, loss=0.64  (1954.5  examples/sec;  0.065  sec/batch)
step 39580, loss=0.59  (1873.6  examples/sec;  0.068  sec/batch)
step 39590, loss=0.63  (1899.6  examples/sec;  0.067  sec/batch)
step 39600, loss=0.67  (1932.0  examples/sec;  0.066  sec/batch)
step 39610, loss=0.64  (1857.7  examples/sec;  0.069  sec/batch)
step 39620, loss=0.68  (1887.8  examples/sec;  0.068  sec/batch)
step 39630, loss=0.63  (1844.8  examples/sec;  0.069  sec/batch)
step 39640, loss=0.75  (1891.2  examples/sec;  0.068  sec/batch)
step 39650, loss=0.71  (1932.7  examples/sec;  0.066  sec/batch)
step 39660, loss=0.61  (1936.2  examples/sec;  0.066  sec/batch)
step 39670, loss=0.61  (1884.3  examples/sec;  0.068  sec/batch)
step 39680, loss=0.59  (1894.5  examples/sec;  0.068  sec/batch)
step 39690, loss=0.67  (1890.5  examples/sec;  0.068  sec/batch)
step 39700, loss=0.74  (1949.6  examples/sec;  0.066  sec/batch)
step 39710, loss=0.64  (1963.3  examples/sec;  0.065  sec/batch)
step 39720, loss=0.78  (1784.1  examples/sec;  0.072  sec/batch)
step 39730, loss=0.77  (1944.2  examples/sec;  0.066  sec/batch)
step 39740, loss=0.61  (1817.1  examples/sec;  0.070  sec/batch)
step 39750, loss=0.58  (1915.2  examples/sec;  0.067  sec/batch)
step 39760, loss=0.68  (1877.6  examples/sec;  0.068  sec/batch)
step 39770, loss=0.62  (1964.5  examples/sec;  0.065  sec/batch)
step 39780, loss=0.64  (1868.9  examples/sec;  0.068  sec/batch)
step 39790, loss=0.74  (1896.0  examples/sec;  0.068  sec/batch)
step 39800, loss=0.70  (1890.0  examples/sec;  0.068  sec/batch)
step 39810, loss=0.66  (1974.8  examples/sec;  0.065  sec/batch)
step 39820, loss=0.50  (1843.2  examples/sec;  0.069  sec/batch)
step 39830, loss=0.64  (1852.4  examples/sec;  0.069  sec/batch)
step 39840, loss=0.77  (1900.0  examples/sec;  0.067  sec/batch)
step 39850, loss=0.71  (1967.4  examples/sec;  0.065  sec/batch)
step 39860, loss=0.70  (1869.6  examples/sec;  0.068  sec/batch)
step 39870, loss=0.57  (1914.8  examples/sec;  0.067  sec/batch)
step 39880, loss=0.76  (1924.4  examples/sec;  0.067  sec/batch)
step 39890, loss=0.60  (1801.1  examples/sec;  0.071  sec/batch)
step 39900, loss=0.66  (1803.5  examples/sec;  0.071  sec/batch)
step 39910, loss=0.52  (1932.1  examples/sec;  0.066  sec/batch)
step 39920, loss=0.79  (1894.8  examples/sec;  0.068  sec/batch)
step 39930, loss=0.62  (1715.1  examples/sec;  0.075  sec/batch)
step 39940, loss=0.73  (1999.9  examples/sec;  0.064  sec/batch)
step 39950, loss=0.56  (1924.5  examples/sec;  0.067  sec/batch)
step 39960, loss=0.55  (2030.3  examples/sec;  0.063  sec/batch)
step 39970, loss=0.61  (1942.5  examples/sec;  0.066  sec/batch)
step 39980, loss=0.80  (1944.5  examples/sec;  0.066  sec/batch)
step 39990, loss=0.76  (1933.7  examples/sec;  0.066  sec/batch)
step 40000, loss=0.58  (1891.7  examples/sec;  0.068  sec/batch)
step 40010, loss=0.78  (1974.1  examples/sec;  0.065  sec/batch)
step 40020, loss=0.80  (1912.4  examples/sec;  0.067  sec/batch)
step 40030, loss=0.67  (1965.8  examples/sec;  0.065  sec/batch)
step 40040, loss=0.81  (1866.5  examples/sec;  0.069  sec/batch)
step 40050, loss=0.73  (1915.6  examples/sec;  0.067  sec/batch)
step 40060, loss=0.67  (1833.3  examples/sec;  0.070  sec/batch)
step 40070, loss=0.68  (1964.8  examples/sec;  0.065  sec/batch)
step 40080, loss=0.69  (1937.3  examples/sec;  0.066  sec/batch)
step 40090, loss=0.71  (1951.3  examples/sec;  0.066  sec/batch)
step 40100, loss=0.63  (1892.6  examples/sec;  0.068  sec/batch)
step 40110, loss=0.77  (1944.6  examples/sec;  0.066  sec/batch)
step 40120, loss=0.57  (1936.0  examples/sec;  0.066  sec/batch)
step 40130, loss=0.76  (1883.9  examples/sec;  0.068  sec/batch)
step 40140, loss=0.59  (1952.2  examples/sec;  0.066  sec/batch)
step 40150, loss=0.76  (1865.6  examples/sec;  0.069  sec/batch)
step 40160, loss=0.71  (1856.7  examples/sec;  0.069  sec/batch)
step 40170, loss=0.56  (1967.5  examples/sec;  0.065  sec/batch)
step 40180, loss=0.61  (1859.7  examples/sec;  0.069  sec/batch)
step 40190, loss=0.62  (1869.7  examples/sec;  0.068  sec/batch)
step 40200, loss=0.54  (1852.7  examples/sec;  0.069  sec/batch)
step 40210, loss=0.58  (1954.6  examples/sec;  0.065  sec/batch)
step 40220, loss=0.82  (1892.9  examples/sec;  0.068  sec/batch)
step 40230, loss=0.66  (1861.8  examples/sec;  0.069  sec/batch)
step 40240, loss=0.64  (1963.3  examples/sec;  0.065  sec/batch)
step 40250, loss=0.62  (1893.9  examples/sec;  0.068  sec/batch)
step 40260, loss=0.62  (1802.7  examples/sec;  0.071  sec/batch)
step 40270, loss=0.66  (1879.1  examples/sec;  0.068  sec/batch)
step 40280, loss=0.67  (1995.6  examples/sec;  0.064  sec/batch)
step 40290, loss=0.77  (1934.7  examples/sec;  0.066  sec/batch)
step 40300, loss=0.77  (1906.3  examples/sec;  0.067  sec/batch)
step 40310, loss=0.78  (1914.7  examples/sec;  0.067  sec/batch)
step 40320, loss=0.57  (1881.0  examples/sec;  0.068  sec/batch)
step 40330, loss=0.60  (1880.3  examples/sec;  0.068  sec/batch)
step 40340, loss=0.58  (1913.6  examples/sec;  0.067  sec/batch)
step 40350, loss=0.71  (1801.1  examples/sec;  0.071  sec/batch)
step 40360, loss=0.55  (1900.9  examples/sec;  0.067  sec/batch)
step 40370, loss=0.84  (1937.5  examples/sec;  0.066  sec/batch)
step 40380, loss=0.82  (1899.9  examples/sec;  0.067  sec/batch)
step 40390, loss=0.64  (1925.4  examples/sec;  0.066  sec/batch)
step 40400, loss=0.65  (1921.7  examples/sec;  0.067  sec/batch)
step 40410, loss=0.60  (1964.3  examples/sec;  0.065  sec/batch)
step 40420, loss=0.66  (1872.1  examples/sec;  0.068  sec/batch)
step 40430, loss=0.87  (1940.3  examples/sec;  0.066  sec/batch)
step 40440, loss=0.70  (1927.9  examples/sec;  0.066  sec/batch)
step 40450, loss=0.62  (1932.7  examples/sec;  0.066  sec/batch)
step 40460, loss=0.74  (1823.1  examples/sec;  0.070  sec/batch)
step 40470, loss=0.48  (1919.0  examples/sec;  0.067  sec/batch)
step 40480, loss=0.62  (1961.5  examples/sec;  0.065  sec/batch)
step 40490, loss=0.67  (1287.0  examples/sec;  0.099  sec/batch)
step 40500, loss=0.84  (1552.4  examples/sec;  0.082  sec/batch)
step 40510, loss=0.70  (1291.6  examples/sec;  0.099  sec/batch)
step 40520, loss=0.60  (1873.9  examples/sec;  0.068  sec/batch)
step 40530, loss=0.67  (2042.2  examples/sec;  0.063  sec/batch)
step 40540, loss=0.63  (1892.9  examples/sec;  0.068  sec/batch)
step 40550, loss=0.67  (1873.2  examples/sec;  0.068  sec/batch)
step 40560, loss=0.60  (1815.0  examples/sec;  0.071  sec/batch)
step 40570, loss=0.47  (1914.3  examples/sec;  0.067  sec/batch)
step 40580, loss=0.67  (1876.5  examples/sec;  0.068  sec/batch)
step 40590, loss=0.63  (1691.2  examples/sec;  0.076  sec/batch)
step 40600, loss=0.79  (1949.6  examples/sec;  0.066  sec/batch)
step 40610, loss=0.60  (1916.3  examples/sec;  0.067  sec/batch)
step 40620, loss=0.70  (1934.5  examples/sec;  0.066  sec/batch)
step 40630, loss=0.64  (1903.3  examples/sec;  0.067  sec/batch)
step 40640, loss=0.52  (1842.0  examples/sec;  0.069  sec/batch)
step 40650, loss=0.55  (1902.4  examples/sec;  0.067  sec/batch)
step 40660, loss=0.84  (1890.2  examples/sec;  0.068  sec/batch)
step 40670, loss=0.60  (1890.6  examples/sec;  0.068  sec/batch)
step 40680, loss=0.57  (1880.2  examples/sec;  0.068  sec/batch)
step 40690, loss=0.69  (1901.2  examples/sec;  0.067  sec/batch)
step 40700, loss=0.69  (1783.1  examples/sec;  0.072  sec/batch)
step 40710, loss=0.64  (1938.2  examples/sec;  0.066  sec/batch)
step 40720, loss=0.70  (1757.0  examples/sec;  0.073  sec/batch)
step 40730, loss=0.61  (1850.2  examples/sec;  0.069  sec/batch)
step 40740, loss=0.69  (1959.9  examples/sec;  0.065  sec/batch)
step 40750, loss=0.59  (1910.9  examples/sec;  0.067  sec/batch)
step 40760, loss=0.62  (1821.9  examples/sec;  0.070  sec/batch)
step 40770, loss=0.63  (1916.7  examples/sec;  0.067  sec/batch)
step 40780, loss=0.60  (1909.8  examples/sec;  0.067  sec/batch)
step 40790, loss=0.76  (1892.0  examples/sec;  0.068  sec/batch)
step 40800, loss=0.73  (1910.2  examples/sec;  0.067  sec/batch)
step 40810, loss=0.75  (1897.8  examples/sec;  0.067  sec/batch)
step 40820, loss=0.74  (1905.1  examples/sec;  0.067  sec/batch)
step 40830, loss=0.74  (1937.4  examples/sec;  0.066  sec/batch)
step 40840, loss=0.82  (1837.5  examples/sec;  0.070  sec/batch)
step 40850, loss=0.63  (1896.2  examples/sec;  0.068  sec/batch)
step 40860, loss=0.64  (1849.7  examples/sec;  0.069  sec/batch)
step 40870, loss=0.91  (1951.3  examples/sec;  0.066  sec/batch)
step 40880, loss=0.72  (1935.6  examples/sec;  0.066  sec/batch)
step 40890, loss=0.51  (1847.0  examples/sec;  0.069  sec/batch)
step 40900, loss=0.71  (1972.6  examples/sec;  0.065  sec/batch)
step 40910, loss=0.60  (1866.9  examples/sec;  0.069  sec/batch)
step 40920, loss=0.57  (1832.3  examples/sec;  0.070  sec/batch)
step 40930, loss=0.82  (1946.9  examples/sec;  0.066  sec/batch)
step 40940, loss=0.72  (1870.3  examples/sec;  0.068  sec/batch)
step 40950, loss=0.60  (1873.6  examples/sec;  0.068  sec/batch)
step 40960, loss=0.49  (1857.0  examples/sec;  0.069  sec/batch)
step 40970, loss=0.50  (1909.7  examples/sec;  0.067  sec/batch)
step 40980, loss=0.63  (1949.2  examples/sec;  0.066  sec/batch)
step 40990, loss=0.72  (1904.5  examples/sec;  0.067  sec/batch)
step 41000, loss=0.78  (1848.9  examples/sec;  0.069  sec/batch)
step 41010, loss=0.64  (1927.4  examples/sec;  0.066  sec/batch)
step 41020, loss=0.65  (1959.8  examples/sec;  0.065  sec/batch)
step 41030, loss=0.79  (1920.0  examples/sec;  0.067  sec/batch)
step 41040, loss=0.64  (1881.2  examples/sec;  0.068  sec/batch)
step 41050, loss=0.64  (1828.5  examples/sec;  0.070  sec/batch)
step 41060, loss=0.61  (1833.9  examples/sec;  0.070  sec/batch)
step 41070, loss=0.58  (1971.7  examples/sec;  0.065  sec/batch)
step 41080, loss=0.75  (1955.7  examples/sec;  0.065  sec/batch)
step 41090, loss=0.82  (1923.0  examples/sec;  0.067  sec/batch)
step 41100, loss=0.65  (1918.5  examples/sec;  0.067  sec/batch)
step 41110, loss=0.61  (1963.1  examples/sec;  0.065  sec/batch)
step 41120, loss=0.74  (1982.4  examples/sec;  0.065  sec/batch)
step 41130, loss=0.78  (1845.6  examples/sec;  0.069  sec/batch)
step 41140, loss=0.57  (1839.1  examples/sec;  0.070  sec/batch)
step 41150, loss=0.72  (1827.4  examples/sec;  0.070  sec/batch)
step 41160, loss=0.75  (1956.8  examples/sec;  0.065  sec/batch)
step 41170, loss=0.67  (1917.6  examples/sec;  0.067  sec/batch)
step 41180, loss=0.76  (1889.6  examples/sec;  0.068  sec/batch)
step 41190, loss=0.63  (1875.6  examples/sec;  0.068  sec/batch)
step 41200, loss=0.64  (1930.3  examples/sec;  0.066  sec/batch)
step 41210, loss=0.82  (1930.5  examples/sec;  0.066  sec/batch)
step 41220, loss=0.77  (1871.8  examples/sec;  0.068  sec/batch)
step 41230, loss=0.61  (1916.7  examples/sec;  0.067  sec/batch)
step 41240, loss=0.75  (1931.5  examples/sec;  0.066  sec/batch)
step 41250, loss=0.61  (1795.2  examples/sec;  0.071  sec/batch)
step 41260, loss=0.56  (1903.7  examples/sec;  0.067  sec/batch)
step 41270, loss=0.63  (2092.8  examples/sec;  0.061  sec/batch)
step 41280, loss=0.59  (1634.6  examples/sec;  0.078  sec/batch)
step 41290, loss=0.65  (1826.0  examples/sec;  0.070  sec/batch)
step 41300, loss=0.74  (1904.4  examples/sec;  0.067  sec/batch)
step 41310, loss=0.53  (1956.3  examples/sec;  0.065  sec/batch)
step 41320, loss=0.56  (1880.4  examples/sec;  0.068  sec/batch)
step 41330, loss=0.61  (1956.3  examples/sec;  0.065  sec/batch)
step 41340, loss=0.69  (1927.5  examples/sec;  0.066  sec/batch)
step 41350, loss=0.42  (1881.3  examples/sec;  0.068  sec/batch)
step 41360, loss=0.74  (1964.7  examples/sec;  0.065  sec/batch)
step 41370, loss=0.77  (1929.7  examples/sec;  0.066  sec/batch)
step 41380, loss=0.57  (1918.6  examples/sec;  0.067  sec/batch)
step 41390, loss=0.71  (1940.5  examples/sec;  0.066  sec/batch)
step 41400, loss=0.71  (1972.4  examples/sec;  0.065  sec/batch)
step 41410, loss=0.70  (1906.4  examples/sec;  0.067  sec/batch)
step 41420, loss=0.65  (1849.1  examples/sec;  0.069  sec/batch)
step 41430, loss=0.78  (1968.9  examples/sec;  0.065  sec/batch)
step 41440, loss=0.74  (1931.3  examples/sec;  0.066  sec/batch)
step 41450, loss=0.68  (1881.4  examples/sec;  0.068  sec/batch)
step 41460, loss=0.72  (1900.1  examples/sec;  0.067  sec/batch)
step 41470, loss=0.74  (1850.3  examples/sec;  0.069  sec/batch)
step 41480, loss=0.61  (1967.8  examples/sec;  0.065  sec/batch)
step 41490, loss=0.65  (1919.2  examples/sec;  0.067  sec/batch)
step 41500, loss=0.62  (1846.7  examples/sec;  0.069  sec/batch)
step 41510, loss=0.56  (1863.1  examples/sec;  0.069  sec/batch)
step 41520, loss=0.57  (1898.0  examples/sec;  0.067  sec/batch)
step 41530, loss=0.73  (1859.5  examples/sec;  0.069  sec/batch)
step 41540, loss=0.57  (1869.7  examples/sec;  0.068  sec/batch)
step 41550, loss=0.72  (1907.8  examples/sec;  0.067  sec/batch)
step 41560, loss=0.63  (1864.8  examples/sec;  0.069  sec/batch)
step 41570, loss=0.73  (1885.8  examples/sec;  0.068  sec/batch)
step 41580, loss=0.57  (1969.5  examples/sec;  0.065  sec/batch)
step 41590, loss=0.65  (1930.0  examples/sec;  0.066  sec/batch)
step 41600, loss=0.55  (1879.2  examples/sec;  0.068  sec/batch)
step 41610, loss=0.51  (2002.6  examples/sec;  0.064  sec/batch)
step 41620, loss=0.79  (1869.4  examples/sec;  0.068  sec/batch)
step 41630, loss=0.60  (1871.9  examples/sec;  0.068  sec/batch)
step 41640, loss=0.81  (2030.3  examples/sec;  0.063  sec/batch)
step 41650, loss=0.58  (1851.7  examples/sec;  0.069  sec/batch)
step 41660, loss=0.67  (1919.6  examples/sec;  0.067  sec/batch)
step 41670, loss=0.61  (1922.4  examples/sec;  0.067  sec/batch)
step 41680, loss=0.56  (1925.4  examples/sec;  0.066  sec/batch)
step 41690, loss=0.59  (1911.6  examples/sec;  0.067  sec/batch)
step 41700, loss=0.67  (1851.7  examples/sec;  0.069  sec/batch)
step 41710, loss=0.73  (1872.9  examples/sec;  0.068  sec/batch)
step 41720, loss=0.81  (1930.9  examples/sec;  0.066  sec/batch)
step 41730, loss=0.60  (1833.1  examples/sec;  0.070  sec/batch)
step 41740, loss=0.71  (1831.2  examples/sec;  0.070  sec/batch)
step 41750, loss=0.58  (1903.9  examples/sec;  0.067  sec/batch)
step 41760, loss=0.54  (1883.6  examples/sec;  0.068  sec/batch)
step 41770, loss=0.60  (1875.6  examples/sec;  0.068  sec/batch)
step 41780, loss=0.63  (1740.6  examples/sec;  0.074  sec/batch)
step 41790, loss=0.69  (1753.6  examples/sec;  0.073  sec/batch)
step 41800, loss=0.62  (1899.4  examples/sec;  0.067  sec/batch)
step 41810, loss=0.46  (1983.5  examples/sec;  0.065  sec/batch)
step 41820, loss=0.63  (1910.4  examples/sec;  0.067  sec/batch)
step 41830, loss=0.63  (1801.0  examples/sec;  0.071  sec/batch)
step 41840, loss=0.66  (1889.7  examples/sec;  0.068  sec/batch)
step 41850, loss=0.91  (1864.8  examples/sec;  0.069  sec/batch)
step 41860, loss=0.64  (1910.0  examples/sec;  0.067  sec/batch)
step 41870, loss=0.73  (2041.1  examples/sec;  0.063  sec/batch)
step 41880, loss=0.58  (1955.8  examples/sec;  0.065  sec/batch)
step 41890, loss=0.61  (1933.7  examples/sec;  0.066  sec/batch)
step 41900, loss=0.64  (1841.1  examples/sec;  0.070  sec/batch)
step 41910, loss=0.64  (1889.3  examples/sec;  0.068  sec/batch)
step 41920, loss=0.51  (1893.2  examples/sec;  0.068  sec/batch)
step 41930, loss=0.56  (1925.4  examples/sec;  0.066  sec/batch)
step 41940, loss=0.68  (1798.6  examples/sec;  0.071  sec/batch)
step 41950, loss=0.58  (1847.1  examples/sec;  0.069  sec/batch)
step 41960, loss=0.87  (1910.1  examples/sec;  0.067  sec/batch)
step 41970, loss=0.67  (1888.9  examples/sec;  0.068  sec/batch)
step 41980, loss=0.61  (1772.7  examples/sec;  0.072  sec/batch)
step 41990, loss=0.72  (1948.4  examples/sec;  0.066  sec/batch)
step 42000, loss=0.57  (1816.7  examples/sec;  0.070  sec/batch)
step 42010, loss=0.66  (1762.1  examples/sec;  0.073  sec/batch)
step 42020, loss=0.69  (1920.9  examples/sec;  0.067  sec/batch)
step 42030, loss=0.66  (1819.5  examples/sec;  0.070  sec/batch)
step 42040, loss=0.70  (1938.0  examples/sec;  0.066  sec/batch)
step 42050, loss=0.67  (1909.3  examples/sec;  0.067  sec/batch)
step 42060, loss=0.58  (1874.1  examples/sec;  0.068  sec/batch)
step 42070, loss=0.67  (1924.1  examples/sec;  0.067  sec/batch)
step 42080, loss=0.65  (1891.9  examples/sec;  0.068  sec/batch)
step 42090, loss=0.62  (1881.4  examples/sec;  0.068  sec/batch)
step 42100, loss=0.52  (2000.4  examples/sec;  0.064  sec/batch)
step 42110, loss=0.63  (1920.6  examples/sec;  0.067  sec/batch)
step 42120, loss=0.80  (1854.7  examples/sec;  0.069  sec/batch)
step 42130, loss=0.55  (1849.5  examples/sec;  0.069  sec/batch)
step 42140, loss=0.68  (1880.9  examples/sec;  0.068  sec/batch)
step 42150, loss=0.78  (1886.9  examples/sec;  0.068  sec/batch)
step 42160, loss=0.59  (1943.6  examples/sec;  0.066  sec/batch)
step 42170, loss=0.66  (1862.4  examples/sec;  0.069  sec/batch)
step 42180, loss=0.54  (1931.7  examples/sec;  0.066  sec/batch)
step 42190, loss=0.64  (1963.6  examples/sec;  0.065  sec/batch)
step 42200, loss=0.65  (1906.8  examples/sec;  0.067  sec/batch)
step 42210, loss=0.64  (1840.5  examples/sec;  0.070  sec/batch)
step 42220, loss=0.70  (1949.1  examples/sec;  0.066  sec/batch)
step 42230, loss=0.62  (1888.1  examples/sec;  0.068  sec/batch)
step 42240, loss=0.78  (1781.2  examples/sec;  0.072  sec/batch)
step 42250, loss=0.70  (1957.5  examples/sec;  0.065  sec/batch)
step 42260, loss=0.84  (1884.2  examples/sec;  0.068  sec/batch)
step 42270, loss=0.67  (1893.2  examples/sec;  0.068  sec/batch)
step 42280, loss=0.62  (1904.7  examples/sec;  0.067  sec/batch)
step 42290, loss=0.69  (1841.9  examples/sec;  0.069  sec/batch)
step 42300, loss=0.68  (1895.3  examples/sec;  0.068  sec/batch)
step 42310, loss=0.73  (1922.9  examples/sec;  0.067  sec/batch)
step 42320, loss=0.66  (1927.9  examples/sec;  0.066  sec/batch)
step 42330, loss=0.56  (1891.0  examples/sec;  0.068  sec/batch)
step 42340, loss=0.67  (1953.9  examples/sec;  0.066  sec/batch)
step 42350, loss=0.68  (1910.5  examples/sec;  0.067  sec/batch)
step 42360, loss=0.72  (1940.9  examples/sec;  0.066  sec/batch)
step 42370, loss=0.63  (1942.5  examples/sec;  0.066  sec/batch)
step 42380, loss=0.78  (1675.5  examples/sec;  0.076  sec/batch)
step 42390, loss=0.60  (1919.6  examples/sec;  0.067  sec/batch)
step 42400, loss=0.73  (1859.9  examples/sec;  0.069  sec/batch)
step 42410, loss=0.61  (1881.2  examples/sec;  0.068  sec/batch)
step 42420, loss=0.53  (1850.6  examples/sec;  0.069  sec/batch)
step 42430, loss=0.69  (1915.7  examples/sec;  0.067  sec/batch)
step 42440, loss=0.66  (1846.9  examples/sec;  0.069  sec/batch)
step 42450, loss=0.72  (1920.6  examples/sec;  0.067  sec/batch)
step 42460, loss=0.64  (1858.0  examples/sec;  0.069  sec/batch)
step 42470, loss=0.67  (1856.2  examples/sec;  0.069  sec/batch)
step 42480, loss=0.75  (1883.1  examples/sec;  0.068  sec/batch)
step 42490, loss=0.59  (1891.7  examples/sec;  0.068  sec/batch)
step 42500, loss=0.66  (1875.6  examples/sec;  0.068  sec/batch)
step 42510, loss=0.75  (1907.9  examples/sec;  0.067  sec/batch)
step 42520, loss=0.69  (1940.8  examples/sec;  0.066  sec/batch)
step 42530, loss=0.67  (1944.8  examples/sec;  0.066  sec/batch)
step 42540, loss=0.65  (1927.7  examples/sec;  0.066  sec/batch)
step 42550, loss=0.58  (1972.8  examples/sec;  0.065  sec/batch)
step 42560, loss=0.55  (1815.1  examples/sec;  0.071  sec/batch)
step 42570, loss=0.61  (1891.2  examples/sec;  0.068  sec/batch)
step 42580, loss=0.75  (1902.7  examples/sec;  0.067  sec/batch)
step 42590, loss=0.66  (1951.6  examples/sec;  0.066  sec/batch)
step 42600, loss=0.86  (1873.2  examples/sec;  0.068  sec/batch)
step 42610, loss=0.64  (1841.4  examples/sec;  0.070  sec/batch)
step 42620, loss=0.74  (1838.8  examples/sec;  0.070  sec/batch)
step 42630, loss=0.78  (1772.8  examples/sec;  0.072  sec/batch)
step 42640, loss=0.62  (1881.0  examples/sec;  0.068  sec/batch)
step 42650, loss=0.56  (1914.4  examples/sec;  0.067  sec/batch)
step 42660, loss=0.63  (1903.1  examples/sec;  0.067  sec/batch)
step 42670, loss=0.79  (1853.8  examples/sec;  0.069  sec/batch)
step 42680, loss=0.73  (1909.4  examples/sec;  0.067  sec/batch)
step 42690, loss=0.58  (1961.0  examples/sec;  0.065  sec/batch)
step 42700, loss=0.56  (1903.1  examples/sec;  0.067  sec/batch)
step 42710, loss=0.65  (1878.1  examples/sec;  0.068  sec/batch)
step 42720, loss=0.66  (1930.8  examples/sec;  0.066  sec/batch)
step 42730, loss=0.75  (1876.2  examples/sec;  0.068  sec/batch)
step 42740, loss=0.63  (1853.8  examples/sec;  0.069  sec/batch)
step 42750, loss=0.66  (1964.1  examples/sec;  0.065  sec/batch)
step 42760, loss=0.72  (1848.9  examples/sec;  0.069  sec/batch)
step 42770, loss=0.83  (1987.1  examples/sec;  0.064  sec/batch)
step 42780, loss=0.64  (2001.4  examples/sec;  0.064  sec/batch)
step 42790, loss=0.80  (1920.9  examples/sec;  0.067  sec/batch)
step 42800, loss=0.61  (1898.1  examples/sec;  0.067  sec/batch)
step 42810, loss=0.51  (1947.7  examples/sec;  0.066  sec/batch)
step 42820, loss=0.61  (1983.6  examples/sec;  0.065  sec/batch)
step 42830, loss=0.68  (1923.3  examples/sec;  0.067  sec/batch)
step 42840, loss=0.54  (2008.4  examples/sec;  0.064  sec/batch)
step 42850, loss=0.63  (1996.4  examples/sec;  0.064  sec/batch)
step 42860, loss=0.61  (1830.3  examples/sec;  0.070  sec/batch)
step 42870, loss=0.77  (1854.0  examples/sec;  0.069  sec/batch)
step 42880, loss=0.76  (1985.5  examples/sec;  0.064  sec/batch)
step 42890, loss=0.58  (1940.1  examples/sec;  0.066  sec/batch)
step 42900, loss=0.85  (1964.6  examples/sec;  0.065  sec/batch)
step 42910, loss=0.59  (1767.8  examples/sec;  0.072  sec/batch)
step 42920, loss=0.68  (1942.5  examples/sec;  0.066  sec/batch)
step 42930, loss=0.77  (1912.8  examples/sec;  0.067  sec/batch)
step 42940, loss=0.71  (1912.8  examples/sec;  0.067  sec/batch)
step 42950, loss=0.70  (1940.7  examples/sec;  0.066  sec/batch)
step 42960, loss=0.67  (1961.9  examples/sec;  0.065  sec/batch)
step 42970, loss=0.69  (1869.2  examples/sec;  0.068  sec/batch)
step 42980, loss=0.67  (1865.1  examples/sec;  0.069  sec/batch)
step 42990, loss=0.56  (1925.6  examples/sec;  0.066  sec/batch)
step 43000, loss=0.62  (1953.1  examples/sec;  0.066  sec/batch)
step 43010, loss=0.55  (1966.5  examples/sec;  0.065  sec/batch)
step 43020, loss=0.81  (1975.4  examples/sec;  0.065  sec/batch)
step 43030, loss=0.79  (1868.4  examples/sec;  0.069  sec/batch)
step 43040, loss=0.53  (1949.5  examples/sec;  0.066  sec/batch)
step 43050, loss=0.68  (1881.1  examples/sec;  0.068  sec/batch)
step 43060, loss=0.67  (1992.0  examples/sec;  0.064  sec/batch)
step 43070, loss=0.70  (1837.0  examples/sec;  0.070  sec/batch)
step 43080, loss=0.73  (1923.1  examples/sec;  0.067  sec/batch)
step 43090, loss=0.76  (1901.7  examples/sec;  0.067  sec/batch)
step 43100, loss=0.66  (1912.4  examples/sec;  0.067  sec/batch)
step 43110, loss=0.56  (1956.8  examples/sec;  0.065  sec/batch)
step 43120, loss=0.52  (1923.7  examples/sec;  0.067  sec/batch)
step 43130, loss=0.74  (1835.3  examples/sec;  0.070  sec/batch)
step 43140, loss=0.74  (1838.8  examples/sec;  0.070  sec/batch)
step 43150, loss=0.46  (1872.3  examples/sec;  0.068  sec/batch)
step 43160, loss=0.77  (1863.0  examples/sec;  0.069  sec/batch)
step 43170, loss=0.65  (1884.9  examples/sec;  0.068  sec/batch)
step 43180, loss=0.69  (1946.3  examples/sec;  0.066  sec/batch)
step 43190, loss=0.67  (1966.2  examples/sec;  0.065  sec/batch)
step 43200, loss=0.72  (1940.3  examples/sec;  0.066  sec/batch)
step 43210, loss=0.78  (1873.3  examples/sec;  0.068  sec/batch)
step 43220, loss=0.60  (1949.0  examples/sec;  0.066  sec/batch)
step 43230, loss=0.62  (1846.6  examples/sec;  0.069  sec/batch)
step 43240, loss=0.66  (1881.7  examples/sec;  0.068  sec/batch)
step 43250, loss=0.62  (1784.5  examples/sec;  0.072  sec/batch)
step 43260, loss=0.49  (1912.6  examples/sec;  0.067  sec/batch)
step 43270, loss=0.62  (1978.0  examples/sec;  0.065  sec/batch)
step 43280, loss=0.62  (1858.7  examples/sec;  0.069  sec/batch)
step 43290, loss=0.66  (1963.4  examples/sec;  0.065  sec/batch)
step 43300, loss=0.59  (1888.3  examples/sec;  0.068  sec/batch)
step 43310, loss=0.70  (1638.6  examples/sec;  0.078  sec/batch)
step 43320, loss=0.67  (1833.9  examples/sec;  0.070  sec/batch)
step 43330, loss=0.84  (1899.2  examples/sec;  0.067  sec/batch)
step 43340, loss=0.64  (1888.5  examples/sec;  0.068  sec/batch)
step 43350, loss=0.58  (1868.7  examples/sec;  0.068  sec/batch)
step 43360, loss=0.71  (1876.9  examples/sec;  0.068  sec/batch)
step 43370, loss=0.53  (1903.5  examples/sec;  0.067  sec/batch)
step 43380, loss=0.81  (1904.7  examples/sec;  0.067  sec/batch)
step 43390, loss=0.65  (1821.5  examples/sec;  0.070  sec/batch)
step 43400, loss=0.67  (1947.9  examples/sec;  0.066  sec/batch)
step 43410, loss=0.72  (1869.3  examples/sec;  0.068  sec/batch)
step 43420, loss=0.55  (1830.3  examples/sec;  0.070  sec/batch)
step 43430, loss=0.63  (1822.7  examples/sec;  0.070  sec/batch)
step 43440, loss=0.64  (1823.1  examples/sec;  0.070  sec/batch)
step 43450, loss=0.86  (1943.4  examples/sec;  0.066  sec/batch)
step 43460, loss=0.62  (1834.1  examples/sec;  0.070  sec/batch)
step 43470, loss=0.63  (1854.5  examples/sec;  0.069  sec/batch)
step 43480, loss=0.66  (1944.0  examples/sec;  0.066  sec/batch)
step 43490, loss=0.67  (1825.8  examples/sec;  0.070  sec/batch)
step 43500, loss=0.79  (1964.8  examples/sec;  0.065  sec/batch)
step 43510, loss=0.72  (1899.5  examples/sec;  0.067  sec/batch)
step 43520, loss=0.74  (1927.1  examples/sec;  0.066  sec/batch)
step 43530, loss=0.54  (1994.6  examples/sec;  0.064  sec/batch)
step 43540, loss=0.54  (1910.4  examples/sec;  0.067  sec/batch)
step 43550, loss=0.67  (1856.9  examples/sec;  0.069  sec/batch)
step 43560, loss=0.56  (1893.6  examples/sec;  0.068  sec/batch)
step 43570, loss=0.54  (1779.8  examples/sec;  0.072  sec/batch)
step 43580, loss=0.59  (1897.3  examples/sec;  0.067  sec/batch)
step 43590, loss=0.73  (1928.6  examples/sec;  0.066  sec/batch)
step 43600, loss=0.54  (1927.9  examples/sec;  0.066  sec/batch)
step 43610, loss=0.71  (1821.2  examples/sec;  0.070  sec/batch)
step 43620, loss=0.54  (1690.2  examples/sec;  0.076  sec/batch)
step 43630, loss=0.63  (1985.0  examples/sec;  0.064  sec/batch)
step 43640, loss=0.67  (1909.2  examples/sec;  0.067  sec/batch)
step 43650, loss=0.56  (1947.7  examples/sec;  0.066  sec/batch)
step 43660, loss=0.63  (1862.6  examples/sec;  0.069  sec/batch)
step 43670, loss=0.56  (1853.8  examples/sec;  0.069  sec/batch)
step 43680, loss=0.57  (1890.9  examples/sec;  0.068  sec/batch)
step 43690, loss=0.72  (1921.9  examples/sec;  0.067  sec/batch)
step 43700, loss=0.86  (1886.6  examples/sec;  0.068  sec/batch)
step 43710, loss=0.76  (1780.1  examples/sec;  0.072  sec/batch)
step 43720, loss=0.65  (1593.5  examples/sec;  0.080  sec/batch)
step 43730, loss=0.65  (1978.6  examples/sec;  0.065  sec/batch)
step 43740, loss=0.64  (2002.8  examples/sec;  0.064  sec/batch)
step 43750, loss=0.62  (1788.3  examples/sec;  0.072  sec/batch)
step 43760, loss=0.83  (1853.8  examples/sec;  0.069  sec/batch)
step 43770, loss=0.68  (1886.4  examples/sec;  0.068  sec/batch)
step 43780, loss=0.67  (1899.3  examples/sec;  0.067  sec/batch)
step 43790, loss=0.66  (1702.9  examples/sec;  0.075  sec/batch)
step 43800, loss=0.71  (1942.6  examples/sec;  0.066  sec/batch)
step 43810, loss=0.77  (1884.7  examples/sec;  0.068  sec/batch)
step 43820, loss=0.67  (1853.2  examples/sec;  0.069  sec/batch)
step 43830, loss=0.62  (1850.6  examples/sec;  0.069  sec/batch)
step 43840, loss=0.67  (1872.4  examples/sec;  0.068  sec/batch)
step 43850, loss=0.74  (1925.6  examples/sec;  0.066  sec/batch)
step 43860, loss=0.68  (1836.0  examples/sec;  0.070  sec/batch)
step 43870, loss=0.58  (1929.1  examples/sec;  0.066  sec/batch)
step 43880, loss=0.81  (1942.9  examples/sec;  0.066  sec/batch)
step 43890, loss=0.80  (1827.4  examples/sec;  0.070  sec/batch)
step 43900, loss=0.63  (1815.3  examples/sec;  0.071  sec/batch)
step 43910, loss=0.65  (1810.5  examples/sec;  0.071  sec/batch)
step 43920, loss=0.73  (1844.2  examples/sec;  0.069  sec/batch)
step 43930, loss=0.57  (1962.1  examples/sec;  0.065  sec/batch)
step 43940, loss=0.65  (1853.0  examples/sec;  0.069  sec/batch)
step 43950, loss=0.68  (1973.1  examples/sec;  0.065  sec/batch)
step 43960, loss=0.81  (1916.2  examples/sec;  0.067  sec/batch)
step 43970, loss=0.66  (1854.9  examples/sec;  0.069  sec/batch)
step 43980, loss=0.77  (1659.4  examples/sec;  0.077  sec/batch)
step 43990, loss=0.57  (1908.6  examples/sec;  0.067  sec/batch)
step 44000, loss=0.72  (1955.9  examples/sec;  0.065  sec/batch)
step 44010, loss=0.68  (1929.7  examples/sec;  0.066  sec/batch)
step 44020, loss=0.71  (1893.2  examples/sec;  0.068  sec/batch)
step 44030, loss=0.68  (1876.4  examples/sec;  0.068  sec/batch)
step 44040, loss=0.71  (1839.9  examples/sec;  0.070  sec/batch)
step 44050, loss=0.73  (1904.2  examples/sec;  0.067  sec/batch)
step 44060, loss=0.75  (1918.2  examples/sec;  0.067  sec/batch)
step 44070, loss=0.71  (1868.9  examples/sec;  0.068  sec/batch)
step 44080, loss=0.68  (1946.5  examples/sec;  0.066  sec/batch)
step 44090, loss=0.69  (1960.6  examples/sec;  0.065  sec/batch)
step 44100, loss=0.89  (1920.9  examples/sec;  0.067  sec/batch)
step 44110, loss=0.59  (1892.1  examples/sec;  0.068  sec/batch)
step 44120, loss=0.72  (1878.3  examples/sec;  0.068  sec/batch)
step 44130, loss=0.56  (1888.2  examples/sec;  0.068  sec/batch)
step 44140, loss=0.62  (1929.7  examples/sec;  0.066  sec/batch)
step 44150, loss=0.60  (1890.6  examples/sec;  0.068  sec/batch)
step 44160, loss=0.68  (1904.6  examples/sec;  0.067  sec/batch)
step 44170, loss=0.81  (1913.0  examples/sec;  0.067  sec/batch)
step 44180, loss=0.75  (1860.7  examples/sec;  0.069  sec/batch)
step 44190, loss=0.65  (1828.0  examples/sec;  0.070  sec/batch)
step 44200, loss=0.64  (1981.3  examples/sec;  0.065  sec/batch)
step 44210, loss=0.73  (1800.9  examples/sec;  0.071  sec/batch)
step 44220, loss=0.58  (1931.2  examples/sec;  0.066  sec/batch)
step 44230, loss=0.71  (1974.9  examples/sec;  0.065  sec/batch)
step 44240, loss=0.61  (1917.0  examples/sec;  0.067  sec/batch)
step 44250, loss=0.62  (1968.8  examples/sec;  0.065  sec/batch)
step 44260, loss=0.67  (1916.2  examples/sec;  0.067  sec/batch)
step 44270, loss=0.62  (1945.4  examples/sec;  0.066  sec/batch)
step 44280, loss=0.74  (1944.4  examples/sec;  0.066  sec/batch)
step 44290, loss=0.61  (1962.3  examples/sec;  0.065  sec/batch)
step 44300, loss=0.79  (1879.9  examples/sec;  0.068  sec/batch)
step 44310, loss=0.57  (1880.3  examples/sec;  0.068  sec/batch)
step 44320, loss=0.66  (1849.3  examples/sec;  0.069  sec/batch)
step 44330, loss=0.67  (1979.4  examples/sec;  0.065  sec/batch)
step 44340, loss=0.65  (1836.1  examples/sec;  0.070  sec/batch)
step 44350, loss=0.61  (1961.8  examples/sec;  0.065  sec/batch)
step 44360, loss=0.73  (1910.3  examples/sec;  0.067  sec/batch)
step 44370, loss=0.70  (1944.6  examples/sec;  0.066  sec/batch)
step 44380, loss=0.49  (1877.3  examples/sec;  0.068  sec/batch)
step 44390, loss=0.70  (1859.2  examples/sec;  0.069  sec/batch)
step 44400, loss=0.57  (1885.3  examples/sec;  0.068  sec/batch)
step 44410, loss=0.78  (1835.4  examples/sec;  0.070  sec/batch)
step 44420, loss=0.81  (1884.2  examples/sec;  0.068  sec/batch)
step 44430, loss=0.85  (2005.1  examples/sec;  0.064  sec/batch)
step 44440, loss=0.64  (1905.0  examples/sec;  0.067  sec/batch)
step 44450, loss=0.71  (1904.1  examples/sec;  0.067  sec/batch)
step 44460, loss=0.80  (1915.5  examples/sec;  0.067  sec/batch)
step 44470, loss=0.85  (1619.5  examples/sec;  0.079  sec/batch)
step 44480, loss=0.62  (1811.4  examples/sec;  0.071  sec/batch)
step 44490, loss=0.66  (1818.1  examples/sec;  0.070  sec/batch)
step 44500, loss=0.60  (1915.7  examples/sec;  0.067  sec/batch)
step 44510, loss=0.74  (1863.6  examples/sec;  0.069  sec/batch)
step 44520, loss=0.83  (1923.0  examples/sec;  0.067  sec/batch)
step 44530, loss=0.64  (1900.1  examples/sec;  0.067  sec/batch)
step 44540, loss=0.87  (1935.8  examples/sec;  0.066  sec/batch)
step 44550, loss=0.80  (1925.5  examples/sec;  0.066  sec/batch)
step 44560, loss=0.56  (1934.6  examples/sec;  0.066  sec/batch)
step 44570, loss=0.68  (1892.7  examples/sec;  0.068  sec/batch)
step 44580, loss=0.58  (1889.9  examples/sec;  0.068  sec/batch)
step 44590, loss=0.66  (1971.3  examples/sec;  0.065  sec/batch)
step 44600, loss=0.68  (1951.0  examples/sec;  0.066  sec/batch)
step 44610, loss=0.61  (1916.7  examples/sec;  0.067  sec/batch)
step 44620, loss=0.79  (1690.8  examples/sec;  0.076  sec/batch)
step 44630, loss=0.78  (1918.3  examples/sec;  0.067  sec/batch)
step 44640, loss=0.64  (1901.0  examples/sec;  0.067  sec/batch)
step 44650, loss=0.85  (1877.8  examples/sec;  0.068  sec/batch)
step 44660, loss=0.52  (2048.2  examples/sec;  0.062  sec/batch)
step 44670, loss=0.64  (1936.7  examples/sec;  0.066  sec/batch)
step 44680, loss=0.74  (1793.0  examples/sec;  0.071  sec/batch)
step 44690, loss=0.67  (1901.5  examples/sec;  0.067  sec/batch)
step 44700, loss=0.57  (1885.6  examples/sec;  0.068  sec/batch)
step 44710, loss=0.69  (1811.5  examples/sec;  0.071  sec/batch)
step 44720, loss=0.62  (1913.0  examples/sec;  0.067  sec/batch)
step 44730, loss=0.63  (1802.1  examples/sec;  0.071  sec/batch)
step 44740, loss=0.65  (1908.4  examples/sec;  0.067  sec/batch)
step 44750, loss=0.71  (1897.2  examples/sec;  0.067  sec/batch)
step 44760, loss=0.65  (1592.7  examples/sec;  0.080  sec/batch)
step 44770, loss=0.76  (1975.6  examples/sec;  0.065  sec/batch)
step 44780, loss=0.72  (1887.5  examples/sec;  0.068  sec/batch)
step 44790, loss=0.56  (1912.2  examples/sec;  0.067  sec/batch)
step 44800, loss=0.64  (1971.7  examples/sec;  0.065  sec/batch)
step 44810, loss=0.70  (1981.4  examples/sec;  0.065  sec/batch)
step 44820, loss=0.84  (1892.5  examples/sec;  0.068  sec/batch)
step 44830, loss=0.61  (1788.0  examples/sec;  0.072  sec/batch)
step 44840, loss=0.66  (1891.2  examples/sec;  0.068  sec/batch)
step 44850, loss=0.63  (1964.1  examples/sec;  0.065  sec/batch)
step 44860, loss=0.63  (1902.8  examples/sec;  0.067  sec/batch)
step 44870, loss=0.61  (1953.7  examples/sec;  0.066  sec/batch)
step 44880, loss=0.62  (1820.2  examples/sec;  0.070  sec/batch)
step 44890, loss=0.61  (1869.7  examples/sec;  0.068  sec/batch)
step 44900, loss=0.57  (1848.6  examples/sec;  0.069  sec/batch)
step 44910, loss=0.65  (1884.1  examples/sec;  0.068  sec/batch)
step 44920, loss=0.65  (1895.4  examples/sec;  0.068  sec/batch)
step 44930, loss=0.59  (1898.2  examples/sec;  0.067  sec/batch)
step 44940, loss=0.60  (2018.8  examples/sec;  0.063  sec/batch)
step 44950, loss=0.47  (1898.1  examples/sec;  0.067  sec/batch)
step 44960, loss=0.53  (1964.5  examples/sec;  0.065  sec/batch)
step 44970, loss=0.65  (1969.4  examples/sec;  0.065  sec/batch)
step 44980, loss=0.60  (1925.4  examples/sec;  0.066  sec/batch)
step 44990, loss=0.76  (1861.2  examples/sec;  0.069  sec/batch)
step 45000, loss=0.67  (1804.3  examples/sec;  0.071  sec/batch)
step 45010, loss=0.75  (1967.4  examples/sec;  0.065  sec/batch)
step 45020, loss=0.69  (1921.8  examples/sec;  0.067  sec/batch)
step 45030, loss=0.57  (1984.8  examples/sec;  0.064  sec/batch)
step 45040, loss=0.66  (1911.8  examples/sec;  0.067  sec/batch)
step 45050, loss=0.77  (1793.5  examples/sec;  0.071  sec/batch)
step 45060, loss=0.55  (1889.0  examples/sec;  0.068  sec/batch)
step 45070, loss=0.80  (1997.1  examples/sec;  0.064  sec/batch)
step 45080, loss=0.56  (1833.9  examples/sec;  0.070  sec/batch)
step 45090, loss=0.56  (1952.9  examples/sec;  0.066  sec/batch)
step 45100, loss=0.55  (1976.3  examples/sec;  0.065  sec/batch)
step 45110, loss=0.70  (1874.1  examples/sec;  0.068  sec/batch)
step 45120, loss=0.56  (1883.5  examples/sec;  0.068  sec/batch)
step 45130, loss=0.73  (1961.9  examples/sec;  0.065  sec/batch)
step 45140, loss=0.64  (1958.2  examples/sec;  0.065  sec/batch)
step 45150, loss=0.63  (1881.2  examples/sec;  0.068  sec/batch)
step 45160, loss=0.71  (1905.4  examples/sec;  0.067  sec/batch)
step 45170, loss=0.72  (1941.0  examples/sec;  0.066  sec/batch)
step 45180, loss=0.57  (1875.6  examples/sec;  0.068  sec/batch)
step 45190, loss=0.52  (1674.3  examples/sec;  0.076  sec/batch)
step 45200, loss=0.59  (1826.8  examples/sec;  0.070  sec/batch)
step 45210, loss=0.66  (1903.5  examples/sec;  0.067  sec/batch)
step 45220, loss=0.82  (1880.6  examples/sec;  0.068  sec/batch)
step 45230, loss=0.58  (1871.4  examples/sec;  0.068  sec/batch)
step 45240, loss=0.60  (1908.2  examples/sec;  0.067  sec/batch)
step 45250, loss=0.74  (1809.6  examples/sec;  0.071  sec/batch)
step 45260, loss=0.70  (1879.5  examples/sec;  0.068  sec/batch)
step 45270, loss=0.64  (1930.3  examples/sec;  0.066  sec/batch)
step 45280, loss=0.78  (1961.8  examples/sec;  0.065  sec/batch)
step 45290, loss=0.44  (1950.5  examples/sec;  0.066  sec/batch)
step 45300, loss=0.69  (1863.1  examples/sec;  0.069  sec/batch)
step 45310, loss=0.69  (1849.2  examples/sec;  0.069  sec/batch)
step 45320, loss=0.53  (1823.9  examples/sec;  0.070  sec/batch)
step 45330, loss=0.64  (1871.3  examples/sec;  0.068  sec/batch)
step 45340, loss=0.63  (1877.1  examples/sec;  0.068  sec/batch)
step 45350, loss=0.80  (1927.3  examples/sec;  0.066  sec/batch)
step 45360, loss=0.86  (1842.1  examples/sec;  0.069  sec/batch)
step 45370, loss=0.64  (1917.7  examples/sec;  0.067  sec/batch)
step 45380, loss=0.75  (1902.5  examples/sec;  0.067  sec/batch)
step 45390, loss=0.58  (1935.5  examples/sec;  0.066  sec/batch)
step 45400, loss=0.58  (1865.6  examples/sec;  0.069  sec/batch)
step 45410, loss=0.65  (1890.6  examples/sec;  0.068  sec/batch)
step 45420, loss=0.75  (1902.9  examples/sec;  0.067  sec/batch)
step 45430, loss=0.70  (1849.9  examples/sec;  0.069  sec/batch)
step 45440, loss=0.52  (1862.2  examples/sec;  0.069  sec/batch)
step 45450, loss=0.63  (1838.6  examples/sec;  0.070  sec/batch)
step 45460, loss=0.58  (1659.6  examples/sec;  0.077  sec/batch)
step 45470, loss=0.68  (1844.3  examples/sec;  0.069  sec/batch)
step 45480, loss=0.71  (1981.6  examples/sec;  0.065  sec/batch)
step 45490, loss=0.56  (1720.4  examples/sec;  0.074  sec/batch)
step 45500, loss=0.62  (1872.3  examples/sec;  0.068  sec/batch)
step 45510, loss=0.56  (1868.6  examples/sec;  0.068  sec/batch)
step 45520, loss=0.62  (1937.6  examples/sec;  0.066  sec/batch)
step 45530, loss=0.65  (1867.8  examples/sec;  0.069  sec/batch)
step 45540, loss=0.61  (1896.9  examples/sec;  0.067  sec/batch)
step 45550, loss=0.65  (1876.7  examples/sec;  0.068  sec/batch)
step 45560, loss=0.88  (1901.1  examples/sec;  0.067  sec/batch)
step 45570, loss=0.67  (1913.6  examples/sec;  0.067  sec/batch)
step 45580, loss=0.62  (1935.4  examples/sec;  0.066  sec/batch)
step 45590, loss=0.66  (1878.8  examples/sec;  0.068  sec/batch)
step 45600, loss=0.61  (1896.5  examples/sec;  0.067  sec/batch)
step 45610, loss=0.80  (1710.2  examples/sec;  0.075  sec/batch)
step 45620, loss=0.57  (1902.8  examples/sec;  0.067  sec/batch)
step 45630, loss=0.73  (1971.7  examples/sec;  0.065  sec/batch)
step 45640, loss=0.71  (1586.6  examples/sec;  0.081  sec/batch)
step 45650, loss=0.64  (1934.0  examples/sec;  0.066  sec/batch)
step 45660, loss=0.63  (1814.8  examples/sec;  0.071  sec/batch)
step 45670, loss=0.78  (1967.1  examples/sec;  0.065  sec/batch)
step 45680, loss=0.64  (1830.8  examples/sec;  0.070  sec/batch)
step 45690, loss=0.64  (1784.9  examples/sec;  0.072  sec/batch)
step 45700, loss=0.68  (1870.3  examples/sec;  0.068  sec/batch)
step 45710, loss=0.65  (1894.6  examples/sec;  0.068  sec/batch)
step 45720, loss=0.68  (1972.0  examples/sec;  0.065  sec/batch)
step 45730, loss=0.59  (1977.3  examples/sec;  0.065  sec/batch)
step 45740, loss=0.62  (1999.7  examples/sec;  0.064  sec/batch)
step 45750, loss=0.64  (1944.0  examples/sec;  0.066  sec/batch)
step 45760, loss=0.59  (1962.3  examples/sec;  0.065  sec/batch)
step 45770, loss=0.59  (1922.4  examples/sec;  0.067  sec/batch)
step 45780, loss=0.56  (1985.2  examples/sec;  0.064  sec/batch)
step 45790, loss=0.57  (1981.3  examples/sec;  0.065  sec/batch)
step 45800, loss=0.66  (1913.6  examples/sec;  0.067  sec/batch)
step 45810, loss=0.66  (1917.0  examples/sec;  0.067  sec/batch)
step 45820, loss=0.45  (1884.4  examples/sec;  0.068  sec/batch)
step 45830, loss=0.52  (1975.6  examples/sec;  0.065  sec/batch)
step 45840, loss=0.71  (1857.2  examples/sec;  0.069  sec/batch)
step 45850, loss=0.64  (1970.4  examples/sec;  0.065  sec/batch)
step 45860, loss=0.74  (1881.1  examples/sec;  0.068  sec/batch)
step 45870, loss=0.67  (1962.2  examples/sec;  0.065  sec/batch)
step 45880, loss=0.67  (1903.9  examples/sec;  0.067  sec/batch)
step 45890, loss=0.64  (1951.9  examples/sec;  0.066  sec/batch)
step 45900, loss=0.59  (1905.6  examples/sec;  0.067  sec/batch)
step 45910, loss=0.66  (1911.1  examples/sec;  0.067  sec/batch)
step 45920, loss=0.70  (1855.8  examples/sec;  0.069  sec/batch)
step 45930, loss=0.56  (1638.4  examples/sec;  0.078  sec/batch)
step 45940, loss=0.73  (1928.0  examples/sec;  0.066  sec/batch)
step 45950, loss=0.57  (1840.4  examples/sec;  0.070  sec/batch)
step 45960, loss=0.79  (1958.9  examples/sec;  0.065  sec/batch)
step 45970, loss=0.66  (1902.5  examples/sec;  0.067  sec/batch)
step 45980, loss=0.53  (2004.2  examples/sec;  0.064  sec/batch)
step 45990, loss=0.63  (1894.4  examples/sec;  0.068  sec/batch)
step 46000, loss=0.70  (1838.9  examples/sec;  0.070  sec/batch)
step 46010, loss=0.66  (1863.4  examples/sec;  0.069  sec/batch)
step 46020, loss=0.64  (1943.6  examples/sec;  0.066  sec/batch)
step 46030, loss=0.63  (1878.0  examples/sec;  0.068  sec/batch)
step 46040, loss=0.80  (1857.0  examples/sec;  0.069  sec/batch)
step 46050, loss=0.53  (1876.6  examples/sec;  0.068  sec/batch)
step 46060, loss=0.60  (1925.0  examples/sec;  0.066  sec/batch)
step 46070, loss=0.61  (1874.7  examples/sec;  0.068  sec/batch)
step 46080, loss=0.57  (1896.9  examples/sec;  0.067  sec/batch)
step 46090, loss=0.42  (1912.7  examples/sec;  0.067  sec/batch)
step 46100, loss=0.67  (1873.9  examples/sec;  0.068  sec/batch)
step 46110, loss=0.73  (1935.9  examples/sec;  0.066  sec/batch)
step 46120, loss=0.47  (1918.8  examples/sec;  0.067  sec/batch)
step 46130, loss=0.69  (1867.2  examples/sec;  0.069  sec/batch)
step 46140, loss=0.68  (1918.8  examples/sec;  0.067  sec/batch)
step 46150, loss=0.66  (1914.9  examples/sec;  0.067  sec/batch)
step 46160, loss=0.75  (1893.2  examples/sec;  0.068  sec/batch)
step 46170, loss=0.57  (1899.8  examples/sec;  0.067  sec/batch)
step 46180, loss=0.69  (1892.8  examples/sec;  0.068  sec/batch)
step 46190, loss=0.60  (1961.0  examples/sec;  0.065  sec/batch)
step 46200, loss=0.68  (1940.8  examples/sec;  0.066  sec/batch)
step 46210, loss=0.62  (1834.2  examples/sec;  0.070  sec/batch)
step 46220, loss=0.67  (1867.0  examples/sec;  0.069  sec/batch)
step 46230, loss=0.77  (1910.7  examples/sec;  0.067  sec/batch)
step 46240, loss=0.84  (1457.1  examples/sec;  0.088  sec/batch)
step 46250, loss=0.80  (1947.4  examples/sec;  0.066  sec/batch)
step 46260, loss=0.69  (1793.9  examples/sec;  0.071  sec/batch)
step 46270, loss=0.80  (1902.5  examples/sec;  0.067  sec/batch)
step 46280, loss=0.79  (1977.7  examples/sec;  0.065  sec/batch)
step 46290, loss=0.67  (1781.5  examples/sec;  0.072  sec/batch)
step 46300, loss=0.55  (1844.1  examples/sec;  0.069  sec/batch)
step 46310, loss=0.60  (1935.0  examples/sec;  0.066  sec/batch)
step 46320, loss=0.81  (1862.2  examples/sec;  0.069  sec/batch)
step 46330, loss=0.62  (1903.7  examples/sec;  0.067  sec/batch)
step 46340, loss=0.72  (1834.4  examples/sec;  0.070  sec/batch)
step 46350, loss=0.59  (1950.8  examples/sec;  0.066  sec/batch)
step 46360, loss=0.68  (1912.2  examples/sec;  0.067  sec/batch)
step 46370, loss=0.72  (1820.0  examples/sec;  0.070  sec/batch)
step 46380, loss=0.70  (1816.4  examples/sec;  0.070  sec/batch)
step 46390, loss=0.60  (1903.7  examples/sec;  0.067  sec/batch)
step 46400, loss=0.64  (1790.8  examples/sec;  0.071  sec/batch)
step 46410, loss=0.53  (1866.0  examples/sec;  0.069  sec/batch)
step 46420, loss=0.56  (1940.3  examples/sec;  0.066  sec/batch)
step 46430, loss=0.61  (1899.2  examples/sec;  0.067  sec/batch)
step 46440, loss=0.73  (1927.1  examples/sec;  0.066  sec/batch)
step 46450, loss=0.55  (1892.6  examples/sec;  0.068  sec/batch)
step 46460, loss=0.69  (1976.3  examples/sec;  0.065  sec/batch)
step 46470, loss=0.65  (1700.3  examples/sec;  0.075  sec/batch)
step 46480, loss=0.58  (1778.9  examples/sec;  0.072  sec/batch)
step 46490, loss=0.79  (1865.5  examples/sec;  0.069  sec/batch)
step 46500, loss=0.65  (1845.7  examples/sec;  0.069  sec/batch)
step 46510, loss=0.61  (1998.4  examples/sec;  0.064  sec/batch)
step 46520, loss=0.62  (1856.7  examples/sec;  0.069  sec/batch)
step 46530, loss=0.71  (1934.2  examples/sec;  0.066  sec/batch)
step 46540, loss=0.85  (2001.6  examples/sec;  0.064  sec/batch)
step 46550, loss=0.70  (1855.6  examples/sec;  0.069  sec/batch)
step 46560, loss=0.74  (1887.0  examples/sec;  0.068  sec/batch)
step 46570, loss=0.70  (1887.8  examples/sec;  0.068  sec/batch)
step 46580, loss=0.72  (1896.6  examples/sec;  0.067  sec/batch)
step 46590, loss=0.65  (2015.4  examples/sec;  0.064  sec/batch)
step 46600, loss=0.65  (1888.8  examples/sec;  0.068  sec/batch)
step 46610, loss=0.71  (1885.5  examples/sec;  0.068  sec/batch)
step 46620, loss=0.85  (1914.0  examples/sec;  0.067  sec/batch)
step 46630, loss=0.53  (1959.8  examples/sec;  0.065  sec/batch)
step 46640, loss=0.79  (1891.1  examples/sec;  0.068  sec/batch)
step 46650, loss=0.69  (1908.4  examples/sec;  0.067  sec/batch)
step 46660, loss=0.52  (1926.1  examples/sec;  0.066  sec/batch)
step 46670, loss=0.57  (1877.4  examples/sec;  0.068  sec/batch)
step 46680, loss=0.68  (1976.8  examples/sec;  0.065  sec/batch)
step 46690, loss=0.65  (1850.0  examples/sec;  0.069  sec/batch)
step 46700, loss=0.81  (1687.4  examples/sec;  0.076  sec/batch)
step 46710, loss=0.58  (1944.7  examples/sec;  0.066  sec/batch)
step 46720, loss=0.51  (1958.5  examples/sec;  0.065  sec/batch)
step 46730, loss=0.58  (1559.9  examples/sec;  0.082  sec/batch)
step 46740, loss=0.51  (1883.0  examples/sec;  0.068  sec/batch)
step 46750, loss=0.66  (1931.0  examples/sec;  0.066  sec/batch)
step 46760, loss=0.73  (1930.4  examples/sec;  0.066  sec/batch)
step 46770, loss=0.56  (1922.5  examples/sec;  0.067  sec/batch)
step 46780, loss=0.64  (1856.0  examples/sec;  0.069  sec/batch)
step 46790, loss=0.67  (1669.0  examples/sec;  0.077  sec/batch)
step 46800, loss=0.80  (1944.8  examples/sec;  0.066  sec/batch)
step 46810, loss=0.75  (1878.5  examples/sec;  0.068  sec/batch)
step 46820, loss=0.71  (2019.4  examples/sec;  0.063  sec/batch)
step 46830, loss=0.74  (1895.4  examples/sec;  0.068  sec/batch)
step 46840, loss=0.75  (1904.5  examples/sec;  0.067  sec/batch)
step 46850, loss=0.69  (1915.5  examples/sec;  0.067  sec/batch)
step 46860, loss=0.69  (1867.5  examples/sec;  0.069  sec/batch)
step 46870, loss=0.52  (1971.0  examples/sec;  0.065  sec/batch)
step 46880, loss=0.53  (2017.3  examples/sec;  0.063  sec/batch)
step 46890, loss=0.88  (1851.5  examples/sec;  0.069  sec/batch)
step 46900, loss=0.73  (1911.6  examples/sec;  0.067  sec/batch)
step 46910, loss=0.58  (1808.7  examples/sec;  0.071  sec/batch)
step 46920, loss=0.68  (1883.3  examples/sec;  0.068  sec/batch)
step 46930, loss=0.77  (1911.3  examples/sec;  0.067  sec/batch)
step 46940, loss=0.64  (1870.3  examples/sec;  0.068  sec/batch)
step 46950, loss=0.68  (1854.0  examples/sec;  0.069  sec/batch)
step 46960, loss=0.81  (1908.7  examples/sec;  0.067  sec/batch)
step 46970, loss=0.57  (1923.6  examples/sec;  0.067  sec/batch)
step 46980, loss=0.76  (1978.4  examples/sec;  0.065  sec/batch)
step 46990, loss=0.71  (1828.7  examples/sec;  0.070  sec/batch)
step 47000, loss=0.81  (1813.0  examples/sec;  0.071  sec/batch)
step 47010, loss=0.74  (1886.8  examples/sec;  0.068  sec/batch)
step 47020, loss=0.57  (1917.7  examples/sec;  0.067  sec/batch)
step 47030, loss=0.48  (1804.7  examples/sec;  0.071  sec/batch)
step 47040, loss=0.73  (1915.4  examples/sec;  0.067  sec/batch)
step 47050, loss=0.87  (1859.8  examples/sec;  0.069  sec/batch)
step 47060, loss=0.73  (1933.3  examples/sec;  0.066  sec/batch)
step 47070, loss=0.77  (1820.2  examples/sec;  0.070  sec/batch)
step 47080, loss=0.72  (1914.4  examples/sec;  0.067  sec/batch)
step 47090, loss=0.52  (1936.6  examples/sec;  0.066  sec/batch)
step 47100, loss=0.68  (1965.4  examples/sec;  0.065  sec/batch)
step 47110, loss=0.69  (1912.2  examples/sec;  0.067  sec/batch)
step 47120, loss=0.49  (1934.9  examples/sec;  0.066  sec/batch)
step 47130, loss=0.71  (1841.8  examples/sec;  0.069  sec/batch)
step 47140, loss=0.70  (1936.7  examples/sec;  0.066  sec/batch)
step 47150, loss=0.60  (1856.8  examples/sec;  0.069  sec/batch)
step 47160, loss=0.65  (1794.9  examples/sec;  0.071  sec/batch)
step 47170, loss=0.76  (1833.0  examples/sec;  0.070  sec/batch)
step 47180, loss=0.60  (1883.0  examples/sec;  0.068  sec/batch)
step 47190, loss=0.67  (1914.9  examples/sec;  0.067  sec/batch)
step 47200, loss=0.70  (1791.4  examples/sec;  0.071  sec/batch)
step 47210, loss=0.74  (1815.7  examples/sec;  0.070  sec/batch)
step 47220, loss=0.65  (1835.4  examples/sec;  0.070  sec/batch)
step 47230, loss=0.67  (1835.8  examples/sec;  0.070  sec/batch)
step 47240, loss=0.63  (1869.7  examples/sec;  0.068  sec/batch)
step 47250, loss=0.87  (1962.4  examples/sec;  0.065  sec/batch)
step 47260, loss=0.75  (1906.1  examples/sec;  0.067  sec/batch)
step 47270, loss=0.64  (2013.9  examples/sec;  0.064  sec/batch)
step 47280, loss=0.76  (1699.9  examples/sec;  0.075  sec/batch)
step 47290, loss=0.65  (1816.7  examples/sec;  0.070  sec/batch)
step 47300, loss=0.70  (1849.8  examples/sec;  0.069  sec/batch)
step 47310, loss=0.70  (1731.8  examples/sec;  0.074  sec/batch)
step 47320, loss=0.63  (1918.3  examples/sec;  0.067  sec/batch)
step 47330, loss=0.67  (1804.2  examples/sec;  0.071  sec/batch)
step 47340, loss=0.66  (2008.7  examples/sec;  0.064  sec/batch)
step 47350, loss=0.61  (1902.8  examples/sec;  0.067  sec/batch)
step 47360, loss=0.74  (1867.7  examples/sec;  0.069  sec/batch)
step 47370, loss=0.62  (1932.3  examples/sec;  0.066  sec/batch)
step 47380, loss=0.69  (1886.2  examples/sec;  0.068  sec/batch)
step 47390, loss=0.60  (1748.2  examples/sec;  0.073  sec/batch)
step 47400, loss=0.66  (1904.6  examples/sec;  0.067  sec/batch)
step 47410, loss=0.68  (2014.0  examples/sec;  0.064  sec/batch)
step 47420, loss=0.75  (1961.0  examples/sec;  0.065  sec/batch)
step 47430, loss=0.50  (1952.9  examples/sec;  0.066  sec/batch)
step 47440, loss=0.63  (1908.4  examples/sec;  0.067  sec/batch)
step 47450, loss=0.95  (1856.0  examples/sec;  0.069  sec/batch)
step 47460, loss=0.62  (1938.8  examples/sec;  0.066  sec/batch)
step 47470, loss=0.61  (1822.9  examples/sec;  0.070  sec/batch)
step 47480, loss=0.72  (1979.5  examples/sec;  0.065  sec/batch)
step 47490, loss=0.56  (1898.6  examples/sec;  0.067  sec/batch)
step 47500, loss=0.64  (1892.6  examples/sec;  0.068  sec/batch)
step 47510, loss=0.61  (1857.9  examples/sec;  0.069  sec/batch)
step 47520, loss=0.68  (1901.9  examples/sec;  0.067  sec/batch)
step 47530, loss=0.74  (1869.2  examples/sec;  0.068  sec/batch)
step 47540, loss=0.68  (1918.5  examples/sec;  0.067  sec/batch)
step 47550, loss=0.57  (1699.3  examples/sec;  0.075  sec/batch)
step 47560, loss=0.54  (1877.8  examples/sec;  0.068  sec/batch)
step 47570, loss=0.67  (1833.6  examples/sec;  0.070  sec/batch)
step 47580, loss=0.62  (1893.2  examples/sec;  0.068  sec/batch)
step 47590, loss=0.69  (1978.2  examples/sec;  0.065  sec/batch)
step 47600, loss=0.65  (1963.1  examples/sec;  0.065  sec/batch)
step 47610, loss=0.67  (1965.0  examples/sec;  0.065  sec/batch)
step 47620, loss=0.66  (1941.3  examples/sec;  0.066  sec/batch)
step 47630, loss=0.51  (2018.6  examples/sec;  0.063  sec/batch)
step 47640, loss=0.59  (1698.8  examples/sec;  0.075  sec/batch)
step 47650, loss=0.56  (1883.1  examples/sec;  0.068  sec/batch)
step 47660, loss=0.71  (1909.2  examples/sec;  0.067  sec/batch)
step 47670, loss=0.75  (1934.4  examples/sec;  0.066  sec/batch)
step 47680, loss=0.69  (1925.4  examples/sec;  0.066  sec/batch)
step 47690, loss=0.63  (1800.6  examples/sec;  0.071  sec/batch)
step 47700, loss=0.67  (1725.6  examples/sec;  0.074  sec/batch)
step 47710, loss=0.63  (1845.7  examples/sec;  0.069  sec/batch)
step 47720, loss=0.53  (1950.5  examples/sec;  0.066  sec/batch)
step 47730, loss=0.59  (1842.7  examples/sec;  0.069  sec/batch)
step 47740, loss=0.73  (1841.4  examples/sec;  0.070  sec/batch)
step 47750, loss=0.66  (1921.5  examples/sec;  0.067  sec/batch)
step 47760, loss=0.75  (1906.0  examples/sec;  0.067  sec/batch)
step 47770, loss=0.75  (1930.7  examples/sec;  0.066  sec/batch)
step 47780, loss=0.55  (1917.6  examples/sec;  0.067  sec/batch)
step 47790, loss=0.62  (1885.6  examples/sec;  0.068  sec/batch)
step 47800, loss=0.80  (1860.7  examples/sec;  0.069  sec/batch)
step 47810, loss=0.56  (1872.0  examples/sec;  0.068  sec/batch)
step 47820, loss=0.62  (1850.5  examples/sec;  0.069  sec/batch)
step 47830, loss=0.67  (1949.6  examples/sec;  0.066  sec/batch)
step 47840, loss=0.61  (1994.3  examples/sec;  0.064  sec/batch)
step 47850, loss=0.63  (1927.7  examples/sec;  0.066  sec/batch)
step 47860, loss=0.67  (1836.5  examples/sec;  0.070  sec/batch)
step 47870, loss=0.58  (1903.7  examples/sec;  0.067  sec/batch)
step 47880, loss=0.75  (1882.4  examples/sec;  0.068  sec/batch)
step 47890, loss=0.66  (1950.0  examples/sec;  0.066  sec/batch)
step 47900, loss=0.60  (1783.4  examples/sec;  0.072  sec/batch)
step 47910, loss=0.66  (1862.7  examples/sec;  0.069  sec/batch)
step 47920, loss=0.57  (1903.5  examples/sec;  0.067  sec/batch)
step 47930, loss=0.49  (1919.9  examples/sec;  0.067  sec/batch)
step 47940, loss=0.69  (1920.6  examples/sec;  0.067  sec/batch)
step 47950, loss=0.57  (1915.6  examples/sec;  0.067  sec/batch)
step 47960, loss=0.90  (1863.2  examples/sec;  0.069  sec/batch)
step 47970, loss=0.60  (1908.8  examples/sec;  0.067  sec/batch)
step 47980, loss=0.66  (1882.7  examples/sec;  0.068  sec/batch)
step 47990, loss=0.73  (1916.6  examples/sec;  0.067  sec/batch)
step 48000, loss=0.73  (2008.0  examples/sec;  0.064  sec/batch)
step 48010, loss=0.75  (1917.7  examples/sec;  0.067  sec/batch)
step 48020, loss=0.74  (1861.2  examples/sec;  0.069  sec/batch)
step 48030, loss=0.56  (1912.6  examples/sec;  0.067  sec/batch)
step 48040, loss=0.67  (1886.0  examples/sec;  0.068  sec/batch)
step 48050, loss=0.65  (1904.6  examples/sec;  0.067  sec/batch)
step 48060, loss=0.72  (1959.9  examples/sec;  0.065  sec/batch)
step 48070, loss=0.80  (1888.5  examples/sec;  0.068  sec/batch)
step 48080, loss=0.73  (1896.5  examples/sec;  0.067  sec/batch)
step 48090, loss=0.63  (1904.1  examples/sec;  0.067  sec/batch)
step 48100, loss=0.74  (1846.8  examples/sec;  0.069  sec/batch)
step 48110, loss=0.63  (1922.7  examples/sec;  0.067  sec/batch)
step 48120, loss=0.56  (1893.3  examples/sec;  0.068  sec/batch)
step 48130, loss=0.66  (1868.9  examples/sec;  0.068  sec/batch)
step 48140, loss=0.68  (1851.7  examples/sec;  0.069  sec/batch)
step 48150, loss=0.78  (1941.7  examples/sec;  0.066  sec/batch)
step 48160, loss=0.55  (1813.8  examples/sec;  0.071  sec/batch)
step 48170, loss=0.66  (1671.8  examples/sec;  0.077  sec/batch)
step 48180, loss=0.72  (1846.8  examples/sec;  0.069  sec/batch)
step 48190, loss=0.50  (1997.6  examples/sec;  0.064  sec/batch)
step 48200, loss=0.53  (1831.2  examples/sec;  0.070  sec/batch)
step 48210, loss=0.66  (1877.1  examples/sec;  0.068  sec/batch)
step 48220, loss=0.68  (1604.5  examples/sec;  0.080  sec/batch)
step 48230, loss=0.67  (1877.3  examples/sec;  0.068  sec/batch)
step 48240, loss=0.53  (1943.5  examples/sec;  0.066  sec/batch)
step 48250, loss=0.64  (1844.9  examples/sec;  0.069  sec/batch)
step 48260, loss=0.64  (1861.0  examples/sec;  0.069  sec/batch)
step 48270, loss=0.65  (1843.2  examples/sec;  0.069  sec/batch)
step 48280, loss=0.52  (1937.0  examples/sec;  0.066  sec/batch)
step 48290, loss=0.58  (1907.6  examples/sec;  0.067  sec/batch)
step 48300, loss=0.68  (1865.0  examples/sec;  0.069  sec/batch)
step 48310, loss=0.59  (1899.5  examples/sec;  0.067  sec/batch)
step 48320, loss=0.77  (1997.0  examples/sec;  0.064  sec/batch)
step 48330, loss=0.71  (1897.5  examples/sec;  0.067  sec/batch)
step 48340, loss=0.58  (1843.9  examples/sec;  0.069  sec/batch)
step 48350, loss=0.56  (1905.2  examples/sec;  0.067  sec/batch)
step 48360, loss=0.65  (1943.3  examples/sec;  0.066  sec/batch)
step 48370, loss=0.67  (1910.2  examples/sec;  0.067  sec/batch)
step 48380, loss=0.74  (1920.0  examples/sec;  0.067  sec/batch)
step 48390, loss=0.71  (1941.1  examples/sec;  0.066  sec/batch)
step 48400, loss=0.72  (2014.6  examples/sec;  0.064  sec/batch)
step 48410, loss=0.60  (1854.9  examples/sec;  0.069  sec/batch)
step 48420, loss=0.63  (1903.0  examples/sec;  0.067  sec/batch)
step 48430, loss=0.60  (1931.6  examples/sec;  0.066  sec/batch)
step 48440, loss=0.57  (1860.6  examples/sec;  0.069  sec/batch)
step 48450, loss=0.55  (1944.0  examples/sec;  0.066  sec/batch)
step 48460, loss=0.63  (1905.3  examples/sec;  0.067  sec/batch)
step 48470, loss=0.59  (1920.4  examples/sec;  0.067  sec/batch)
step 48480, loss=0.56  (1867.9  examples/sec;  0.069  sec/batch)
step 48490, loss=0.61  (1976.8  examples/sec;  0.065  sec/batch)
step 48500, loss=0.59  (1868.2  examples/sec;  0.069  sec/batch)
step 48510, loss=0.73  (1874.4  examples/sec;  0.068  sec/batch)
step 48520, loss=0.73  (1934.2  examples/sec;  0.066  sec/batch)
step 48530, loss=0.58  (1833.3  examples/sec;  0.070  sec/batch)
step 48540, loss=0.68  (1866.2  examples/sec;  0.069  sec/batch)
step 48550, loss=0.69  (1837.6  examples/sec;  0.070  sec/batch)
step 48560, loss=0.70  (1887.3  examples/sec;  0.068  sec/batch)
step 48570, loss=0.68  (1969.9  examples/sec;  0.065  sec/batch)
step 48580, loss=0.78  (1965.2  examples/sec;  0.065  sec/batch)
step 48590, loss=0.63  (1942.2  examples/sec;  0.066  sec/batch)
step 48600, loss=0.66  (2020.8  examples/sec;  0.063  sec/batch)
step 48610, loss=0.70  (1966.6  examples/sec;  0.065  sec/batch)
step 48620, loss=0.63  (1958.1  examples/sec;  0.065  sec/batch)
step 48630, loss=0.62  (1861.7  examples/sec;  0.069  sec/batch)
step 48640, loss=0.54  (1828.9  examples/sec;  0.070  sec/batch)
step 48650, loss=0.62  (1886.7  examples/sec;  0.068  sec/batch)
step 48660, loss=0.66  (1877.1  examples/sec;  0.068  sec/batch)
step 48670, loss=0.78  (1934.4  examples/sec;  0.066  sec/batch)
step 48680, loss=0.60  (1833.3  examples/sec;  0.070  sec/batch)
step 48690, loss=0.68  (1946.8  examples/sec;  0.066  sec/batch)
step 48700, loss=0.65  (1828.9  examples/sec;  0.070  sec/batch)
step 48710, loss=0.72  (1855.0  examples/sec;  0.069  sec/batch)
step 48720, loss=0.59  (1904.2  examples/sec;  0.067  sec/batch)
step 48730, loss=0.67  (1926.5  examples/sec;  0.066  sec/batch)
step 48740, loss=0.70  (1938.5  examples/sec;  0.066  sec/batch)
step 48750, loss=0.71  (1825.7  examples/sec;  0.070  sec/batch)
step 48760, loss=0.75  (1790.8  examples/sec;  0.071  sec/batch)
step 48770, loss=0.62  (1913.9  examples/sec;  0.067  sec/batch)
step 48780, loss=0.43  (1854.9  examples/sec;  0.069  sec/batch)
step 48790, loss=0.63  (1881.8  examples/sec;  0.068  sec/batch)
step 48800, loss=0.71  (1802.8  examples/sec;  0.071  sec/batch)
step 48810, loss=0.57  (1848.6  examples/sec;  0.069  sec/batch)
step 48820, loss=0.54  (1959.0  examples/sec;  0.065  sec/batch)
step 48830, loss=0.61  (1909.6  examples/sec;  0.067  sec/batch)
step 48840, loss=0.58  (1930.8  examples/sec;  0.066  sec/batch)
step 48850, loss=0.69  (1880.9  examples/sec;  0.068  sec/batch)
step 48860, loss=0.64  (1893.0  examples/sec;  0.068  sec/batch)
step 48870, loss=0.71  (1865.5  examples/sec;  0.069  sec/batch)
step 48880, loss=0.50  (1860.3  examples/sec;  0.069  sec/batch)
step 48890, loss=0.54  (1744.1  examples/sec;  0.073  sec/batch)
step 48900, loss=0.48  (1839.7  examples/sec;  0.070  sec/batch)
step 48910, loss=0.64  (1825.1  examples/sec;  0.070  sec/batch)
step 48920, loss=0.65  (1826.9  examples/sec;  0.070  sec/batch)
step 48930, loss=0.61  (1144.6  examples/sec;  0.112  sec/batch)
step 48940, loss=0.62  (1955.9  examples/sec;  0.065  sec/batch)
step 48950, loss=0.79  (1818.0  examples/sec;  0.070  sec/batch)
step 48960, loss=0.71  (1298.7  examples/sec;  0.099  sec/batch)
step 48970, loss=0.62  (1890.7  examples/sec;  0.068  sec/batch)
step 48980, loss=0.69  (1871.3  examples/sec;  0.068  sec/batch)
step 48990, loss=0.64  (1895.2  examples/sec;  0.068  sec/batch)
step 49000, loss=0.62  (1817.4  examples/sec;  0.070  sec/batch)
step 49010, loss=0.53  (1894.6  examples/sec;  0.068  sec/batch)
step 49020, loss=0.75  (1831.4  examples/sec;  0.070  sec/batch)
step 49030, loss=0.57  (1878.1  examples/sec;  0.068  sec/batch)
step 49040, loss=0.77  (1944.6  examples/sec;  0.066  sec/batch)
step 49050, loss=0.50  (1943.5  examples/sec;  0.066  sec/batch)
step 49060, loss=0.65  (1880.4  examples/sec;  0.068  sec/batch)
step 49070, loss=0.50  (1869.7  examples/sec;  0.068  sec/batch)
step 49080, loss=0.52  (1899.5  examples/sec;  0.067  sec/batch)
step 49090, loss=0.53  (1974.0  examples/sec;  0.065  sec/batch)
step 49100, loss=0.62  (1985.7  examples/sec;  0.064  sec/batch)
step 49110, loss=0.58  (1880.6  examples/sec;  0.068  sec/batch)
step 49120, loss=0.85  (1907.5  examples/sec;  0.067  sec/batch)
step 49130, loss=0.74  (1879.7  examples/sec;  0.068  sec/batch)
step 49140, loss=0.64  (1954.7  examples/sec;  0.065  sec/batch)
step 49150, loss=0.57  (1873.2  examples/sec;  0.068  sec/batch)
step 49160, loss=0.62  (1885.7  examples/sec;  0.068  sec/batch)
step 49170, loss=0.61  (1891.2  examples/sec;  0.068  sec/batch)
step 49180, loss=0.67  (1961.7  examples/sec;  0.065  sec/batch)
step 49190, loss=0.81  (1911.2  examples/sec;  0.067  sec/batch)
step 49200, loss=0.63  (1940.0  examples/sec;  0.066  sec/batch)
step 49210, loss=0.58  (1873.8  examples/sec;  0.068  sec/batch)
step 49220, loss=0.77  (1900.7  examples/sec;  0.067  sec/batch)
step 49230, loss=0.77  (1949.9  examples/sec;  0.066  sec/batch)
step 49240, loss=0.88  (1801.0  examples/sec;  0.071  sec/batch)
step 49250, loss=0.63  (1929.1  examples/sec;  0.066  sec/batch)
step 49260, loss=0.56  (2002.4  examples/sec;  0.064  sec/batch)
step 49270, loss=0.71  (1883.9  examples/sec;  0.068  sec/batch)
step 49280, loss=0.78  (1913.8  examples/sec;  0.067  sec/batch)
step 49290, loss=0.65  (1909.2  examples/sec;  0.067  sec/batch)
step 49300, loss=0.52  (1859.4  examples/sec;  0.069  sec/batch)
step 49310, loss=0.74  (1889.8  examples/sec;  0.068  sec/batch)
step 49320, loss=0.63  (1878.6  examples/sec;  0.068  sec/batch)
step 49330, loss=0.59  (1962.2  examples/sec;  0.065  sec/batch)
step 49340, loss=0.55  (2027.5  examples/sec;  0.063  sec/batch)
step 49350, loss=0.62  (1891.7  examples/sec;  0.068  sec/batch)
step 49360, loss=0.58  (1976.2  examples/sec;  0.065  sec/batch)
step 49370, loss=0.69  (1551.9  examples/sec;  0.082  sec/batch)
step 49380, loss=0.60  (1890.0  examples/sec;  0.068  sec/batch)
step 49390, loss=0.52  (1904.0  examples/sec;  0.067  sec/batch)
step 49400, loss=0.60  (1951.7  examples/sec;  0.066  sec/batch)
step 49410, loss=0.73  (1893.2  examples/sec;  0.068  sec/batch)
step 49420, loss=0.56  (1880.9  examples/sec;  0.068  sec/batch)
step 49430, loss=0.67  (1884.7  examples/sec;  0.068  sec/batch)
step 49440, loss=0.70  (1882.5  examples/sec;  0.068  sec/batch)
step 49450, loss=0.71  (1768.2  examples/sec;  0.072  sec/batch)
step 49460, loss=0.56  (1534.6  examples/sec;  0.083  sec/batch)
step 49470, loss=0.69  (1924.5  examples/sec;  0.067  sec/batch)
step 49480, loss=0.66  (1874.8  examples/sec;  0.068  sec/batch)
step 49490, loss=0.64  (1899.2  examples/sec;  0.067  sec/batch)
step 49500, loss=0.86  (1916.5  examples/sec;  0.067  sec/batch)
step 49510, loss=0.56  (1875.9  examples/sec;  0.068  sec/batch)
step 49520, loss=0.65  (1818.3  examples/sec;  0.070  sec/batch)
step 49530, loss=0.69  (1847.5  examples/sec;  0.069  sec/batch)
step 49540, loss=0.69  (1795.2  examples/sec;  0.071  sec/batch)
step 49550, loss=0.85  (1936.5  examples/sec;  0.066  sec/batch)
step 49560, loss=0.70  (1834.9  examples/sec;  0.070  sec/batch)
step 49570, loss=0.64  (1988.4  examples/sec;  0.064  sec/batch)
step 49580, loss=0.63  (1881.1  examples/sec;  0.068  sec/batch)
step 49590, loss=0.68  (1903.1  examples/sec;  0.067  sec/batch)
step 49600, loss=0.72  (1944.8  examples/sec;  0.066  sec/batch)
step 49610, loss=0.59  (1998.4  examples/sec;  0.064  sec/batch)
step 49620, loss=0.59  (1941.2  examples/sec;  0.066  sec/batch)
step 49630, loss=0.65  (1670.1  examples/sec;  0.077  sec/batch)
step 49640, loss=0.65  (1905.6  examples/sec;  0.067  sec/batch)
step 49650, loss=0.58  (1893.0  examples/sec;  0.068  sec/batch)
step 49660, loss=0.64  (1891.2  examples/sec;  0.068  sec/batch)
step 49670, loss=0.58  (1940.9  examples/sec;  0.066  sec/batch)
step 49680, loss=0.66  (1816.0  examples/sec;  0.070  sec/batch)
step 49690, loss=0.83  (1883.9  examples/sec;  0.068  sec/batch)
step 49700, loss=0.72  (1816.4  examples/sec;  0.070  sec/batch)
step 49710, loss=0.46  (1857.3  examples/sec;  0.069  sec/batch)
step 49720, loss=0.63  (1911.6  examples/sec;  0.067  sec/batch)
step 49730, loss=0.68  (1919.4  examples/sec;  0.067  sec/batch)
step 49740, loss=0.77  (1945.2  examples/sec;  0.066  sec/batch)
step 49750, loss=0.49  (1892.5  examples/sec;  0.068  sec/batch)
step 49760, loss=0.74  (1970.4  examples/sec;  0.065  sec/batch)
step 49770, loss=0.65  (1892.0  examples/sec;  0.068  sec/batch)
step 49780, loss=0.71  (1884.4  examples/sec;  0.068  sec/batch)
step 49790, loss=0.66  (1910.1  examples/sec;  0.067  sec/batch)
step 49800, loss=0.62  (1938.3  examples/sec;  0.066  sec/batch)
step 49810, loss=0.59  (1942.7  examples/sec;  0.066  sec/batch)
step 49820, loss=0.63  (1931.6  examples/sec;  0.066  sec/batch)
step 49830, loss=0.67  (1867.9  examples/sec;  0.069  sec/batch)
step 49840, loss=0.50  (1866.0  examples/sec;  0.069  sec/batch)
step 49850, loss=0.66  (1926.0  examples/sec;  0.066  sec/batch)
step 49860, loss=0.60  (1839.8  examples/sec;  0.070  sec/batch)
step 49870, loss=0.67  (1974.2  examples/sec;  0.065  sec/batch)
step 49880, loss=0.61  (1951.7  examples/sec;  0.066  sec/batch)
step 49890, loss=0.49  (1912.6  examples/sec;  0.067  sec/batch)
step 49900, loss=0.82  (1909.2  examples/sec;  0.067  sec/batch)
step 49910, loss=0.60  (1762.6  examples/sec;  0.073  sec/batch)
step 49920, loss=0.74  (1904.4  examples/sec;  0.067  sec/batch)
step 49930, loss=0.59  (1905.0  examples/sec;  0.067  sec/batch)
step 49940, loss=0.64  (1863.0  examples/sec;  0.069  sec/batch)
step 49950, loss=0.50  (1825.6  examples/sec;  0.070  sec/batch)
step 49960, loss=0.77  (1938.7  examples/sec;  0.066  sec/batch)
step 49970, loss=0.61  (1834.8  examples/sec;  0.070  sec/batch)
step 49980, loss=0.68  (1954.9  examples/sec;  0.065  sec/batch)
step 49990, loss=0.75  (1914.2  examples/sec;  0.067  sec/batch)
step 50000, loss=0.48  (1641.0  examples/sec;  0.078  sec/batch)
step 50010, loss=0.51  (1975.1  examples/sec;  0.065  sec/batch)
step 50020, loss=0.56  (2041.4  examples/sec;  0.063  sec/batch)
step 50030, loss=0.72  (1843.9  examples/sec;  0.069  sec/batch)
step 50040, loss=0.57  (1891.0  examples/sec;  0.068  sec/batch)
step 50050, loss=0.54  (1840.0  examples/sec;  0.070  sec/batch)
step 50060, loss=0.65  (1968.3  examples/sec;  0.065  sec/batch)
step 50070, loss=0.70  (1885.8  examples/sec;  0.068  sec/batch)
step 50080, loss=0.53  (1801.0  examples/sec;  0.071  sec/batch)
step 50090, loss=0.64  (1881.0  examples/sec;  0.068  sec/batch)
step 50100, loss=0.61  (1962.4  examples/sec;  0.065  sec/batch)
step 50110, loss=0.71  (1817.2  examples/sec;  0.070  sec/batch)
step 50120, loss=0.57  (1924.9  examples/sec;  0.066  sec/batch)
step 50130, loss=0.73  (1886.5  examples/sec;  0.068  sec/batch)
step 50140, loss=0.67  (1881.5  examples/sec;  0.068  sec/batch)
step 50150, loss=0.67  (1886.1  examples/sec;  0.068  sec/batch)
step 50160, loss=0.68  (1916.0  examples/sec;  0.067  sec/batch)
step 50170, loss=0.74  (1870.2  examples/sec;  0.068  sec/batch)
step 50180, loss=0.72  (1907.5  examples/sec;  0.067  sec/batch)
step 50190, loss=0.62  (1870.7  examples/sec;  0.068  sec/batch)
step 50200, loss=0.73  (1991.7  examples/sec;  0.064  sec/batch)
step 50210, loss=0.57  (1947.3  examples/sec;  0.066  sec/batch)
step 50220, loss=0.76  (1880.8  examples/sec;  0.068  sec/batch)
step 50230, loss=0.67  (1922.6  examples/sec;  0.067  sec/batch)
step 50240, loss=0.71  (1923.2  examples/sec;  0.067  sec/batch)
step 50250, loss=0.65  (1909.3  examples/sec;  0.067  sec/batch)
step 50260, loss=0.60  (1962.9  examples/sec;  0.065  sec/batch)
step 50270, loss=0.62  (1912.7  examples/sec;  0.067  sec/batch)
step 50280, loss=0.64  (1883.7  examples/sec;  0.068  sec/batch)
step 50290, loss=0.64  (1907.9  examples/sec;  0.067  sec/batch)
step 50300, loss=0.87  (1947.6  examples/sec;  0.066  sec/batch)
step 50310, loss=0.60  (1944.3  examples/sec;  0.066  sec/batch)
step 50320, loss=0.65  (1933.9  examples/sec;  0.066  sec/batch)
step 50330, loss=0.64  (1935.4  examples/sec;  0.066  sec/batch)
step 50340, loss=0.66  (1940.3  examples/sec;  0.066  sec/batch)
step 50350, loss=0.50  (1897.3  examples/sec;  0.067  sec/batch)
step 50360, loss=0.64  (1968.0  examples/sec;  0.065  sec/batch)
step 50370, loss=0.67  (1892.3  examples/sec;  0.068  sec/batch)
step 50380, loss=0.49  (1849.2  examples/sec;  0.069  sec/batch)
step 50390, loss=0.68  (1991.3  examples/sec;  0.064  sec/batch)
step 50400, loss=0.91  (1914.6  examples/sec;  0.067  sec/batch)
step 50410, loss=0.47  (1925.7  examples/sec;  0.066  sec/batch)
step 50420, loss=0.79  (1934.4  examples/sec;  0.066  sec/batch)
step 50430, loss=0.68  (1849.3  examples/sec;  0.069  sec/batch)
step 50440, loss=0.67  (1888.1  examples/sec;  0.068  sec/batch)
step 50450, loss=0.77  (1881.0  examples/sec;  0.068  sec/batch)
step 50460, loss=0.61  (1905.2  examples/sec;  0.067  sec/batch)
step 50470, loss=0.76  (1943.0  examples/sec;  0.066  sec/batch)
step 50480, loss=0.71  (1892.2  examples/sec;  0.068  sec/batch)
step 50490, loss=0.63  (1913.3  examples/sec;  0.067  sec/batch)
step 50500, loss=0.67  (1924.0  examples/sec;  0.067  sec/batch)
step 50510, loss=0.59  (1946.9  examples/sec;  0.066  sec/batch)
step 50520, loss=0.64  (1977.8  examples/sec;  0.065  sec/batch)
step 50530, loss=0.58  (1897.7  examples/sec;  0.067  sec/batch)
step 50540, loss=0.64  (1942.7  examples/sec;  0.066  sec/batch)
step 50550, loss=0.62  (2026.4  examples/sec;  0.063  sec/batch)
step 50560, loss=0.67  (1817.8  examples/sec;  0.070  sec/batch)
step 50570, loss=0.71  (1842.4  examples/sec;  0.069  sec/batch)
step 50580, loss=0.68  (1906.6  examples/sec;  0.067  sec/batch)
step 50590, loss=0.75  (1928.5  examples/sec;  0.066  sec/batch)
step 50600, loss=0.53  (1891.6  examples/sec;  0.068  sec/batch)
step 50610, loss=0.51  (1898.9  examples/sec;  0.067  sec/batch)
step 50620, loss=0.58  (1805.3  examples/sec;  0.071  sec/batch)
step 50630, loss=0.67  (1927.9  examples/sec;  0.066  sec/batch)
step 50640, loss=0.64  (1812.1  examples/sec;  0.071  sec/batch)
step 50650, loss=0.87  (1906.0  examples/sec;  0.067  sec/batch)
step 50660, loss=0.52  (1899.8  examples/sec;  0.067  sec/batch)
step 50670, loss=0.52  (1980.3  examples/sec;  0.065  sec/batch)
step 50680, loss=0.57  (1888.4  examples/sec;  0.068  sec/batch)
step 50690, loss=0.54  (1900.5  examples/sec;  0.067  sec/batch)
step 50700, loss=0.70  (1938.1  examples/sec;  0.066  sec/batch)
step 50710, loss=0.80  (1923.8  examples/sec;  0.067  sec/batch)
step 50720, loss=0.57  (1966.0  examples/sec;  0.065  sec/batch)
step 50730, loss=0.73  (1984.5  examples/sec;  0.065  sec/batch)
step 50740, loss=0.53  (1901.6  examples/sec;  0.067  sec/batch)
step 50750, loss=0.62  (1823.2  examples/sec;  0.070  sec/batch)
step 50760, loss=0.75  (1878.6  examples/sec;  0.068  sec/batch)
step 50770, loss=0.68  (1962.2  examples/sec;  0.065  sec/batch)
step 50780, loss=0.70  (1973.4  examples/sec;  0.065  sec/batch)
step 50790, loss=0.77  (1892.3  examples/sec;  0.068  sec/batch)
step 50800, loss=0.72  (1957.1  examples/sec;  0.065  sec/batch)
step 50810, loss=0.76  (1889.6  examples/sec;  0.068  sec/batch)
step 50820, loss=0.68  (1909.9  examples/sec;  0.067  sec/batch)
step 50830, loss=0.70  (1944.8  examples/sec;  0.066  sec/batch)
step 50840, loss=0.52  (1704.1  examples/sec;  0.075  sec/batch)
step 50850, loss=0.60  (1919.5  examples/sec;  0.067  sec/batch)
step 50860, loss=0.64  (1853.3  examples/sec;  0.069  sec/batch)
step 50870, loss=0.75  (1864.5  examples/sec;  0.069  sec/batch)
step 50880, loss=0.62  (1920.4  examples/sec;  0.067  sec/batch)
step 50890, loss=0.63  (1868.1  examples/sec;  0.069  sec/batch)
step 50900, loss=0.71  (1851.8  examples/sec;  0.069  sec/batch)
step 50910, loss=0.68  (1795.0  examples/sec;  0.071  sec/batch)
step 50920, loss=0.77  (1972.6  examples/sec;  0.065  sec/batch)
step 50930, loss=0.53  (1878.2  examples/sec;  0.068  sec/batch)
step 50940, loss=0.62  (1890.6  examples/sec;  0.068  sec/batch)
step 50950, loss=0.76  (1898.7  examples/sec;  0.067  sec/batch)
step 50960, loss=0.64  (1954.6  examples/sec;  0.065  sec/batch)
step 50970, loss=0.57  (1861.8  examples/sec;  0.069  sec/batch)
step 50980, loss=0.71  (1926.3  examples/sec;  0.066  sec/batch)
step 50990, loss=0.62  (1804.9  examples/sec;  0.071  sec/batch)
step 51000, loss=0.60  (1821.5  examples/sec;  0.070  sec/batch)
step 51010, loss=0.69  (1947.6  examples/sec;  0.066  sec/batch)
step 51020, loss=0.70  (1864.2  examples/sec;  0.069  sec/batch)
step 51030, loss=0.80  (1961.8  examples/sec;  0.065  sec/batch)
step 51040, loss=0.74  (1899.2  examples/sec;  0.067  sec/batch)
step 51050, loss=0.85  (2043.4  examples/sec;  0.063  sec/batch)
step 51060, loss=0.82  (1811.4  examples/sec;  0.071  sec/batch)
step 51070, loss=0.61  (1941.7  examples/sec;  0.066  sec/batch)
step 51080, loss=0.55  (1969.0  examples/sec;  0.065  sec/batch)
step 51090, loss=0.54  (1928.0  examples/sec;  0.066  sec/batch)
step 51100, loss=0.66  (1911.7  examples/sec;  0.067  sec/batch)
step 51110, loss=0.55  (1916.9  examples/sec;  0.067  sec/batch)
step 51120, loss=0.61  (1921.4  examples/sec;  0.067  sec/batch)
step 51130, loss=0.57  (1830.1  examples/sec;  0.070  sec/batch)
step 51140, loss=0.55  (1880.8  examples/sec;  0.068  sec/batch)
step 51150, loss=0.54  (1812.9  examples/sec;  0.071  sec/batch)
step 51160, loss=0.68  (1907.5  examples/sec;  0.067  sec/batch)
step 51170, loss=0.61  (1971.1  examples/sec;  0.065  sec/batch)
step 51180, loss=0.72  (1933.9  examples/sec;  0.066  sec/batch)
step 51190, loss=0.41  (1924.7  examples/sec;  0.067  sec/batch)
step 51200, loss=0.55  (1827.4  examples/sec;  0.070  sec/batch)
step 51210, loss=0.72  (1910.0  examples/sec;  0.067  sec/batch)
step 51220, loss=0.56  (1863.4  examples/sec;  0.069  sec/batch)
step 51230, loss=0.53  (1906.8  examples/sec;  0.067  sec/batch)
step 51240, loss=0.54  (1906.3  examples/sec;  0.067  sec/batch)
step 51250, loss=0.52  (1896.4  examples/sec;  0.067  sec/batch)
step 51260, loss=0.58  (1813.8  examples/sec;  0.071  sec/batch)
step 51270, loss=0.58  (1831.4  examples/sec;  0.070  sec/batch)
step 51280, loss=0.65  (1990.9  examples/sec;  0.064  sec/batch)
step 51290, loss=0.67  (1899.1  examples/sec;  0.067  sec/batch)
step 51300, loss=0.56  (1992.9  examples/sec;  0.064  sec/batch)
step 51310, loss=0.59  (1913.3  examples/sec;  0.067  sec/batch)
step 51320, loss=0.67  (1935.2  examples/sec;  0.066  sec/batch)
step 51330, loss=0.58  (1934.0  examples/sec;  0.066  sec/batch)
step 51340, loss=0.62  (1985.1  examples/sec;  0.064  sec/batch)
step 51350, loss=0.70  (1907.7  examples/sec;  0.067  sec/batch)
step 51360, loss=0.67  (1904.5  examples/sec;  0.067  sec/batch)
step 51370, loss=0.62  (1926.5  examples/sec;  0.066  sec/batch)
step 51380, loss=0.77  (1803.0  examples/sec;  0.071  sec/batch)
step 51390, loss=0.54  (1865.4  examples/sec;  0.069  sec/batch)
step 51400, loss=0.67  (1645.1  examples/sec;  0.078  sec/batch)
step 51410, loss=0.75  (1884.4  examples/sec;  0.068  sec/batch)
step 51420, loss=0.62  (1908.5  examples/sec;  0.067  sec/batch)
step 51430, loss=0.42  (1909.3  examples/sec;  0.067  sec/batch)
step 51440, loss=0.74  (1955.4  examples/sec;  0.065  sec/batch)
step 51450, loss=0.60  (1936.1  examples/sec;  0.066  sec/batch)
step 51460, loss=0.65  (1890.0  examples/sec;  0.068  sec/batch)
step 51470, loss=0.62  (1943.6  examples/sec;  0.066  sec/batch)
step 51480, loss=0.65  (1886.2  examples/sec;  0.068  sec/batch)
step 51490, loss=0.66  (1868.2  examples/sec;  0.069  sec/batch)
step 51500, loss=0.77  (1834.9  examples/sec;  0.070  sec/batch)
step 51510, loss=0.74  (1897.0  examples/sec;  0.067  sec/batch)
step 51520, loss=0.71  (1826.6  examples/sec;  0.070  sec/batch)
step 51530, loss=0.52  (1876.9  examples/sec;  0.068  sec/batch)
step 51540, loss=0.60  (1840.0  examples/sec;  0.070  sec/batch)
step 51550, loss=0.52  (1864.0  examples/sec;  0.069  sec/batch)
step 51560, loss=0.57  (1931.4  examples/sec;  0.066  sec/batch)
step 51570, loss=0.66  (1852.0  examples/sec;  0.069  sec/batch)
step 51580, loss=0.63  (1863.2  examples/sec;  0.069  sec/batch)
step 51590, loss=0.54  (1937.1  examples/sec;  0.066  sec/batch)
step 51600, loss=0.62  (1853.0  examples/sec;  0.069  sec/batch)
step 51610, loss=0.67  (2021.8  examples/sec;  0.063  sec/batch)
step 51620, loss=0.69  (1788.7  examples/sec;  0.072  sec/batch)
step 51630, loss=0.72  (1852.8  examples/sec;  0.069  sec/batch)
step 51640, loss=0.62  (1942.7  examples/sec;  0.066  sec/batch)
step 51650, loss=0.55  (1888.6  examples/sec;  0.068  sec/batch)
step 51660, loss=0.65  (1915.8  examples/sec;  0.067  sec/batch)
step 51670, loss=0.82  (1922.2  examples/sec;  0.067  sec/batch)
step 51680, loss=0.76  (1996.8  examples/sec;  0.064  sec/batch)
step 51690, loss=0.58  (1782.7  examples/sec;  0.072  sec/batch)
step 51700, loss=0.60  (1854.3  examples/sec;  0.069  sec/batch)
step 51710, loss=0.58  (1874.9  examples/sec;  0.068  sec/batch)
step 51720, loss=0.49  (1864.5  examples/sec;  0.069  sec/batch)
step 51730, loss=0.71  (1969.9  examples/sec;  0.065  sec/batch)
step 51740, loss=0.70  (1904.4  examples/sec;  0.067  sec/batch)
step 51750, loss=0.54  (1884.0  examples/sec;  0.068  sec/batch)
step 51760, loss=0.63  (1945.8  examples/sec;  0.066  sec/batch)
step 51770, loss=0.63  (1899.6  examples/sec;  0.067  sec/batch)
step 51780, loss=0.68  (1994.5  examples/sec;  0.064  sec/batch)
step 51790, loss=0.56  (1876.6  examples/sec;  0.068  sec/batch)
step 51800, loss=0.55  (1891.6  examples/sec;  0.068  sec/batch)
step 51810, loss=0.62  (1987.2  examples/sec;  0.064  sec/batch)
step 51820, loss=0.76  (1904.9  examples/sec;  0.067  sec/batch)
step 51830, loss=0.71  (1965.7  examples/sec;  0.065  sec/batch)
step 51840, loss=0.69  (1862.9  examples/sec;  0.069  sec/batch)
step 51850, loss=0.65  (1873.6  examples/sec;  0.068  sec/batch)
step 51860, loss=0.56  (1894.6  examples/sec;  0.068  sec/batch)
step 51870, loss=0.63  (1833.3  examples/sec;  0.070  sec/batch)
step 51880, loss=0.52  (1903.8  examples/sec;  0.067  sec/batch)
step 51890, loss=0.56  (1917.5  examples/sec;  0.067  sec/batch)
step 51900, loss=0.67  (1903.1  examples/sec;  0.067  sec/batch)
step 51910, loss=0.62  (1971.3  examples/sec;  0.065  sec/batch)
step 51920, loss=0.82  (1933.7  examples/sec;  0.066  sec/batch)
step 51930, loss=0.77  (1939.5  examples/sec;  0.066  sec/batch)
step 51940, loss=0.79  (1905.9  examples/sec;  0.067  sec/batch)
step 51950, loss=0.76  (1893.7  examples/sec;  0.068  sec/batch)
step 51960, loss=0.69  (1787.8  examples/sec;  0.072  sec/batch)
step 51970, loss=0.58  (1848.1  examples/sec;  0.069  sec/batch)
step 51980, loss=0.70  (1986.1  examples/sec;  0.064  sec/batch)
step 51990, loss=0.60  (1925.5  examples/sec;  0.066  sec/batch)
step 52000, loss=0.60  (1886.9  examples/sec;  0.068  sec/batch)
step 52010, loss=0.67  (1887.6  examples/sec;  0.068  sec/batch)
step 52020, loss=0.60  (1814.2  examples/sec;  0.071  sec/batch)
step 52030, loss=0.53  (1856.3  examples/sec;  0.069  sec/batch)
step 52040, loss=0.69  (1924.2  examples/sec;  0.067  sec/batch)
step 52050, loss=0.57  (1976.8  examples/sec;  0.065  sec/batch)
step 52060, loss=0.79  (1890.2  examples/sec;  0.068  sec/batch)
step 52070, loss=0.71  (1894.0  examples/sec;  0.068  sec/batch)
step 52080, loss=0.79  (1870.2  examples/sec;  0.068  sec/batch)
step 52090, loss=0.52  (1877.8  examples/sec;  0.068  sec/batch)
step 52100, loss=0.55  (1798.9  examples/sec;  0.071  sec/batch)
step 52110, loss=0.68  (1916.3  examples/sec;  0.067  sec/batch)
step 52120, loss=0.65  (2045.8  examples/sec;  0.063  sec/batch)
step 52130, loss=0.68  (1891.2  examples/sec;  0.068  sec/batch)
step 52140, loss=0.63  (1975.8  examples/sec;  0.065  sec/batch)
step 52150, loss=0.53  (1879.7  examples/sec;  0.068  sec/batch)
step 52160, loss=0.56  (1930.4  examples/sec;  0.066  sec/batch)
step 52170, loss=0.76  (2007.2  examples/sec;  0.064  sec/batch)
step 52180, loss=0.67  (1937.5  examples/sec;  0.066  sec/batch)
step 52190, loss=0.74  (1870.8  examples/sec;  0.068  sec/batch)
step 52200, loss=0.69  (1970.7  examples/sec;  0.065  sec/batch)
step 52210, loss=0.61  (1889.5  examples/sec;  0.068  sec/batch)
step 52220, loss=0.70  (1833.9  examples/sec;  0.070  sec/batch)
step 52230, loss=0.64  (1896.5  examples/sec;  0.067  sec/batch)
step 52240, loss=0.61  (1915.1  examples/sec;  0.067  sec/batch)
step 52250, loss=0.56  (1960.0  examples/sec;  0.065  sec/batch)
step 52260, loss=0.65  (1990.5  examples/sec;  0.064  sec/batch)
step 52270, loss=0.65  (1795.5  examples/sec;  0.071  sec/batch)
step 52280, loss=0.67  (1858.5  examples/sec;  0.069  sec/batch)
step 52290, loss=0.65  (1899.3  examples/sec;  0.067  sec/batch)
step 52300, loss=0.55  (1982.7  examples/sec;  0.065  sec/batch)
step 52310, loss=0.65  (1885.6  examples/sec;  0.068  sec/batch)
step 52320, loss=0.65  (1956.5  examples/sec;  0.065  sec/batch)
step 52330, loss=0.74  (1900.4  examples/sec;  0.067  sec/batch)
step 52340, loss=0.53  (1910.5  examples/sec;  0.067  sec/batch)
step 52350, loss=0.65  (1741.0  examples/sec;  0.074  sec/batch)
step 52360, loss=0.55  (1914.0  examples/sec;  0.067  sec/batch)
step 52370, loss=0.65  (1939.5  examples/sec;  0.066  sec/batch)
step 52380, loss=0.65  (1843.6  examples/sec;  0.069  sec/batch)
step 52390, loss=0.75  (1944.8  examples/sec;  0.066  sec/batch)
step 52400, loss=0.57  (1825.8  examples/sec;  0.070  sec/batch)
step 52410, loss=0.70  (1905.1  examples/sec;  0.067  sec/batch)
step 52420, loss=0.69  (1894.1  examples/sec;  0.068  sec/batch)
step 52430, loss=0.69  (1848.3  examples/sec;  0.069  sec/batch)
step 52440, loss=0.69  (1923.1  examples/sec;  0.067  sec/batch)
step 52450, loss=0.80  (1926.4  examples/sec;  0.066  sec/batch)
step 52460, loss=0.70  (1842.2  examples/sec;  0.069  sec/batch)
step 52470, loss=0.56  (1785.0  examples/sec;  0.072  sec/batch)
step 52480, loss=0.58  (1856.2  examples/sec;  0.069  sec/batch)
step 52490, loss=0.62  (1634.3  examples/sec;  0.078  sec/batch)
step 52500, loss=0.51  (1898.2  examples/sec;  0.067  sec/batch)
step 52510, loss=0.70  (1959.4  examples/sec;  0.065  sec/batch)
step 52520, loss=0.76  (1926.1  examples/sec;  0.066  sec/batch)
step 52530, loss=0.78  (1917.5  examples/sec;  0.067  sec/batch)
step 52540, loss=0.64  (1904.2  examples/sec;  0.067  sec/batch)
step 52550, loss=0.59  (1873.0  examples/sec;  0.068  sec/batch)
step 52560, loss=0.66  (1866.3  examples/sec;  0.069  sec/batch)
step 52570, loss=0.57  (1886.0  examples/sec;  0.068  sec/batch)
step 52580, loss=0.71  (1836.7  examples/sec;  0.070  sec/batch)
step 52590, loss=0.65  (1829.9  examples/sec;  0.070  sec/batch)
step 52600, loss=0.61  (1855.4  examples/sec;  0.069  sec/batch)
step 52610, loss=0.62  (1991.6  examples/sec;  0.064  sec/batch)
step 52620, loss=0.72  (1889.0  examples/sec;  0.068  sec/batch)
step 52630, loss=0.62  (1928.1  examples/sec;  0.066  sec/batch)
step 52640, loss=0.57  (1856.6  examples/sec;  0.069  sec/batch)
step 52650, loss=0.74  (1915.7  examples/sec;  0.067  sec/batch)
step 52660, loss=0.57  (1949.1  examples/sec;  0.066  sec/batch)
step 52670, loss=0.74  (1882.7  examples/sec;  0.068  sec/batch)
step 52680, loss=0.64  (1889.9  examples/sec;  0.068  sec/batch)
step 52690, loss=0.72  (1804.7  examples/sec;  0.071  sec/batch)
step 52700, loss=0.82  (1888.6  examples/sec;  0.068  sec/batch)
step 52710, loss=0.63  (1943.0  examples/sec;  0.066  sec/batch)
step 52720, loss=0.52  (1883.8  examples/sec;  0.068  sec/batch)
step 52730, loss=0.75  (1979.6  examples/sec;  0.065  sec/batch)
step 52740, loss=0.59  (1903.1  examples/sec;  0.067  sec/batch)
step 52750, loss=0.48  (1872.6  examples/sec;  0.068  sec/batch)
step 52760, loss=0.85  (1887.3  examples/sec;  0.068  sec/batch)
step 52770, loss=0.63  (1882.3  examples/sec;  0.068  sec/batch)
step 52780, loss=0.63  (1963.2  examples/sec;  0.065  sec/batch)
step 52790, loss=0.86  (1914.4  examples/sec;  0.067  sec/batch)
step 52800, loss=0.66  (1853.5  examples/sec;  0.069  sec/batch)
step 52810, loss=0.64  (1938.4  examples/sec;  0.066  sec/batch)
step 52820, loss=0.57  (1925.6  examples/sec;  0.066  sec/batch)
step 52830, loss=0.57  (1903.8  examples/sec;  0.067  sec/batch)
step 52840, loss=0.58  (1859.2  examples/sec;  0.069  sec/batch)
step 52850, loss=0.66  (1839.2  examples/sec;  0.070  sec/batch)
step 52860, loss=0.54  (1950.2  examples/sec;  0.066  sec/batch)
step 52870, loss=0.65  (1892.3  examples/sec;  0.068  sec/batch)
step 52880, loss=0.74  (1825.3  examples/sec;  0.070  sec/batch)
step 52890, loss=0.69  (1918.8  examples/sec;  0.067  sec/batch)
step 52900, loss=0.57  (1840.5  examples/sec;  0.070  sec/batch)
step 52910, loss=0.63  (1960.9  examples/sec;  0.065  sec/batch)
step 52920, loss=0.49  (1920.7  examples/sec;  0.067  sec/batch)
step 52930, loss=0.63  (1934.7  examples/sec;  0.066  sec/batch)
step 52940, loss=0.74  (1926.9  examples/sec;  0.066  sec/batch)
step 52950, loss=0.55  (1874.2  examples/sec;  0.068  sec/batch)
step 52960, loss=0.71  (1722.2  examples/sec;  0.074  sec/batch)
step 52970, loss=0.81  (1907.6  examples/sec;  0.067  sec/batch)
step 52980, loss=0.62  (1883.1  examples/sec;  0.068  sec/batch)
step 52990, loss=0.61  (1967.0  examples/sec;  0.065  sec/batch)
step 53000, loss=0.50  (1876.9  examples/sec;  0.068  sec/batch)
step 53010, loss=0.58  (1872.4  examples/sec;  0.068  sec/batch)
step 53020, loss=0.74  (1897.3  examples/sec;  0.067  sec/batch)
step 53030, loss=0.56  (1884.1  examples/sec;  0.068  sec/batch)
step 53040, loss=0.59  (1886.8  examples/sec;  0.068  sec/batch)
step 53050, loss=0.51  (1877.0  examples/sec;  0.068  sec/batch)
step 53060, loss=0.69  (2061.9  examples/sec;  0.062  sec/batch)
step 53070, loss=0.56  (1874.6  examples/sec;  0.068  sec/batch)
step 53080, loss=0.49  (1482.4  examples/sec;  0.086  sec/batch)
step 53090, loss=0.68  (1854.9  examples/sec;  0.069  sec/batch)
step 53100, loss=0.72  (1962.4  examples/sec;  0.065  sec/batch)
step 53110, loss=0.60  (1948.6  examples/sec;  0.066  sec/batch)
step 53120, loss=0.69  (1939.9  examples/sec;  0.066  sec/batch)
step 53130, loss=0.53  (1922.3  examples/sec;  0.067  sec/batch)
step 53140, loss=0.59  (1932.0  examples/sec;  0.066  sec/batch)
step 53150, loss=0.59  (1878.5  examples/sec;  0.068  sec/batch)
step 53160, loss=0.61  (1844.3  examples/sec;  0.069  sec/batch)
step 53170, loss=0.48  (1938.4  examples/sec;  0.066  sec/batch)
step 53180, loss=0.55  (1869.7  examples/sec;  0.068  sec/batch)
step 53190, loss=0.68  (1714.1  examples/sec;  0.075  sec/batch)
step 53200, loss=0.53  (1893.3  examples/sec;  0.068  sec/batch)
step 53210, loss=0.72  (1962.7  examples/sec;  0.065  sec/batch)
step 53220, loss=0.76  (1925.4  examples/sec;  0.066  sec/batch)
step 53230, loss=0.78  (1894.9  examples/sec;  0.068  sec/batch)
step 53240, loss=0.56  (1846.5  examples/sec;  0.069  sec/batch)
step 53250, loss=0.74  (1966.8  examples/sec;  0.065  sec/batch)
step 53260, loss=0.80  (1902.2  examples/sec;  0.067  sec/batch)
step 53270, loss=0.83  (1890.6  examples/sec;  0.068  sec/batch)
step 53280, loss=0.70  (1975.2  examples/sec;  0.065  sec/batch)
step 53290, loss=0.69  (1921.1  examples/sec;  0.067  sec/batch)
step 53300, loss=0.62  (1860.9  examples/sec;  0.069  sec/batch)
step 53310, loss=0.62  (1952.4  examples/sec;  0.066  sec/batch)
step 53320, loss=0.78  (1905.5  examples/sec;  0.067  sec/batch)
step 53330, loss=0.53  (1988.6  examples/sec;  0.064  sec/batch)
step 53340, loss=0.42  (1893.7  examples/sec;  0.068  sec/batch)
step 53350, loss=0.54  (1818.3  examples/sec;  0.070  sec/batch)
step 53360, loss=0.54  (1673.4  examples/sec;  0.076  sec/batch)
step 53370, loss=0.85  (1950.3  examples/sec;  0.066  sec/batch)
step 53380, loss=0.55  (1778.7  examples/sec;  0.072  sec/batch)
step 53390, loss=0.58  (1643.0  examples/sec;  0.078  sec/batch)
step 53400, loss=0.69  (1922.0  examples/sec;  0.067  sec/batch)
step 53410, loss=0.57  (1909.8  examples/sec;  0.067  sec/batch)
step 53420, loss=0.57  (1895.4  examples/sec;  0.068  sec/batch)
step 53430, loss=0.62  (1870.4  examples/sec;  0.068  sec/batch)
step 53440, loss=0.51  (1936.1  examples/sec;  0.066  sec/batch)
step 53450, loss=0.66  (1886.0  examples/sec;  0.068  sec/batch)
step 53460, loss=0.55  (1694.1  examples/sec;  0.076  sec/batch)
step 53470, loss=0.70  (1807.7  examples/sec;  0.071  sec/batch)
step 53480, loss=0.60  (1885.0  examples/sec;  0.068  sec/batch)
step 53490, loss=0.58  (1857.8  examples/sec;  0.069  sec/batch)
step 53500, loss=0.66  (1845.7  examples/sec;  0.069  sec/batch)
step 53510, loss=0.68  (1937.2  examples/sec;  0.066  sec/batch)
step 53520, loss=0.53  (1715.9  examples/sec;  0.075  sec/batch)
step 53530, loss=0.59  (1908.2  examples/sec;  0.067  sec/batch)
step 53540, loss=0.65  (1864.0  examples/sec;  0.069  sec/batch)
step 53550, loss=0.64  (1894.4  examples/sec;  0.068  sec/batch)
step 53560, loss=0.65  (1971.6  examples/sec;  0.065  sec/batch)
step 53570, loss=0.60  (1922.9  examples/sec;  0.067  sec/batch)
step 53580, loss=0.57  (1990.7  examples/sec;  0.064  sec/batch)
step 53590, loss=0.54  (1869.6  examples/sec;  0.068  sec/batch)
step 53600, loss=0.57  (1960.1  examples/sec;  0.065  sec/batch)
step 53610, loss=0.60  (1899.7  examples/sec;  0.067  sec/batch)
step 53620, loss=0.75  (1816.3  examples/sec;  0.070  sec/batch)
step 53630, loss=0.80  (1820.1  examples/sec;  0.070  sec/batch)
step 53640, loss=0.59  (1830.3  examples/sec;  0.070  sec/batch)
step 53650, loss=0.63  (1885.5  examples/sec;  0.068  sec/batch)
step 53660, loss=0.53  (1902.1  examples/sec;  0.067  sec/batch)
step 53670, loss=0.49  (1812.5  examples/sec;  0.071  sec/batch)
step 53680, loss=0.65  (1914.9  examples/sec;  0.067  sec/batch)
step 53690, loss=0.76  (1790.0  examples/sec;  0.072  sec/batch)
step 53700, loss=0.59  (1886.1  examples/sec;  0.068  sec/batch)
step 53710, loss=0.52  (1929.6  examples/sec;  0.066  sec/batch)
step 53720, loss=0.73  (1972.7  examples/sec;  0.065  sec/batch)
step 53730, loss=0.73  (1849.9  examples/sec;  0.069  sec/batch)
step 53740, loss=0.84  (1867.8  examples/sec;  0.069  sec/batch)
step 53750, loss=0.61  (1641.1  examples/sec;  0.078  sec/batch)
step 53760, loss=0.67  (1835.8  examples/sec;  0.070  sec/batch)
step 53770, loss=0.59  (1845.1  examples/sec;  0.069  sec/batch)
step 53780, loss=0.55  (1990.4  examples/sec;  0.064  sec/batch)
step 53790, loss=0.59  (1876.4  examples/sec;  0.068  sec/batch)
step 53800, loss=0.65  (1980.0  examples/sec;  0.065  sec/batch)
step 53810, loss=0.76  (1928.6  examples/sec;  0.066  sec/batch)
step 53820, loss=0.62  (1892.9  examples/sec;  0.068  sec/batch)
step 53830, loss=0.49  (1904.3  examples/sec;  0.067  sec/batch)
step 53840, loss=0.71  (1829.8  examples/sec;  0.070  sec/batch)
step 53850, loss=0.68  (1550.8  examples/sec;  0.083  sec/batch)
step 53860, loss=0.73  (1903.9  examples/sec;  0.067  sec/batch)
step 53870, loss=0.56  (1879.1  examples/sec;  0.068  sec/batch)
step 53880, loss=0.68  (1982.4  examples/sec;  0.065  sec/batch)
step 53890, loss=0.56  (1875.3  examples/sec;  0.068  sec/batch)
step 53900, loss=0.59  (1900.2  examples/sec;  0.067  sec/batch)
step 53910, loss=0.76  (1848.8  examples/sec;  0.069  sec/batch)
step 53920, loss=0.71  (1939.1  examples/sec;  0.066  sec/batch)
step 53930, loss=0.74  (1846.7  examples/sec;  0.069  sec/batch)
step 53940, loss=0.61  (1923.6  examples/sec;  0.067  sec/batch)
step 53950, loss=0.74  (1969.3  examples/sec;  0.065  sec/batch)
step 53960, loss=0.80  (1968.7  examples/sec;  0.065  sec/batch)
step 53970, loss=0.55  (1884.2  examples/sec;  0.068  sec/batch)
step 53980, loss=0.63  (1929.8  examples/sec;  0.066  sec/batch)
step 53990, loss=0.62  (1979.3  examples/sec;  0.065  sec/batch)
step 54000, loss=0.59  (1914.2  examples/sec;  0.067  sec/batch)
step 54010, loss=0.54  (1944.3  examples/sec;  0.066  sec/batch)
step 54020, loss=0.67  (1998.4  examples/sec;  0.064  sec/batch)
step 54030, loss=0.55  (1925.3  examples/sec;  0.066  sec/batch)
step 54040, loss=0.67  (1929.0  examples/sec;  0.066  sec/batch)
step 54050, loss=0.69  (1934.1  examples/sec;  0.066  sec/batch)
step 54060, loss=0.56  (1879.4  examples/sec;  0.068  sec/batch)
step 54070, loss=0.73  (1976.8  examples/sec;  0.065  sec/batch)
step 54080, loss=0.65  (1821.0  examples/sec;  0.070  sec/batch)
step 54090, loss=0.73  (1854.2  examples/sec;  0.069  sec/batch)
step 54100, loss=0.64  (1948.6  examples/sec;  0.066  sec/batch)
step 54110, loss=0.65  (1822.1  examples/sec;  0.070  sec/batch)
step 54120, loss=0.64  (1801.0  examples/sec;  0.071  sec/batch)
step 54130, loss=0.82  (1869.6  examples/sec;  0.068  sec/batch)
step 54140, loss=0.75  (1420.9  examples/sec;  0.090  sec/batch)
step 54150, loss=0.77  (1891.7  examples/sec;  0.068  sec/batch)
step 54160, loss=0.66  (1926.2  examples/sec;  0.066  sec/batch)
step 54170, loss=0.68  (1913.5  examples/sec;  0.067  sec/batch)
step 54180, loss=0.69  (1891.3  examples/sec;  0.068  sec/batch)
step 54190, loss=0.86  (1916.8  examples/sec;  0.067  sec/batch)
step 54200, loss=0.63  (1920.3  examples/sec;  0.067  sec/batch)
step 54210, loss=0.57  (1897.3  examples/sec;  0.067  sec/batch)
step 54220, loss=0.71  (1931.8  examples/sec;  0.066  sec/batch)
step 54230, loss=0.65  (1507.6  examples/sec;  0.085  sec/batch)
step 54240, loss=0.57  (1991.8  examples/sec;  0.064  sec/batch)
step 54250, loss=0.67  (1982.8  examples/sec;  0.065  sec/batch)
step 54260, loss=0.64  (1570.6  examples/sec;  0.082  sec/batch)
step 54270, loss=0.55  (1902.3  examples/sec;  0.067  sec/batch)
step 54280, loss=0.63  (1877.8  examples/sec;  0.068  sec/batch)
step 54290, loss=0.59  (1916.6  examples/sec;  0.067  sec/batch)
step 54300, loss=0.59  (1903.1  examples/sec;  0.067  sec/batch)
step 54310, loss=0.54  (1849.8  examples/sec;  0.069  sec/batch)
step 54320, loss=0.66  (1929.1  examples/sec;  0.066  sec/batch)
step 54330, loss=0.54  (1954.0  examples/sec;  0.066  sec/batch)
step 54340, loss=0.67  (1944.1  examples/sec;  0.066  sec/batch)
step 54350, loss=0.67  (1849.6  examples/sec;  0.069  sec/batch)
step 54360, loss=0.70  (1988.7  examples/sec;  0.064  sec/batch)
step 54370, loss=0.69  (1852.5  examples/sec;  0.069  sec/batch)
step 54380, loss=0.58  (1885.3  examples/sec;  0.068  sec/batch)
step 54390, loss=0.52  (1907.4  examples/sec;  0.067  sec/batch)
step 54400, loss=0.59  (1853.7  examples/sec;  0.069  sec/batch)
step 54410, loss=0.71  (1864.7  examples/sec;  0.069  sec/batch)
step 54420, loss=0.65  (1871.5  examples/sec;  0.068  sec/batch)
step 54430, loss=0.66  (1825.8  examples/sec;  0.070  sec/batch)
step 54440, loss=0.69  (1901.5  examples/sec;  0.067  sec/batch)
step 54450, loss=0.76  (1942.9  examples/sec;  0.066  sec/batch)
step 54460, loss=0.66  (1825.2  examples/sec;  0.070  sec/batch)
step 54470, loss=0.69  (2023.5  examples/sec;  0.063  sec/batch)
step 54480, loss=0.60  (1932.3  examples/sec;  0.066  sec/batch)
step 54490, loss=0.55  (1951.3  examples/sec;  0.066  sec/batch)
step 54500, loss=0.52  (1867.4  examples/sec;  0.069  sec/batch)
step 54510, loss=0.74  (1856.9  examples/sec;  0.069  sec/batch)
step 54520, loss=0.71  (1930.3  examples/sec;  0.066  sec/batch)
step 54530, loss=0.62  (1936.2  examples/sec;  0.066  sec/batch)
step 54540, loss=0.68  (2020.1  examples/sec;  0.063  sec/batch)
step 54550, loss=0.54  (1950.8  examples/sec;  0.066  sec/batch)
step 54560, loss=0.58  (1601.7  examples/sec;  0.080  sec/batch)
step 54570, loss=0.56  (1891.8  examples/sec;  0.068  sec/batch)
step 54580, loss=0.79  (1820.2  examples/sec;  0.070  sec/batch)
step 54590, loss=0.66  (1942.0  examples/sec;  0.066  sec/batch)
step 54600, loss=0.66  (1874.6  examples/sec;  0.068  sec/batch)
step 54610, loss=0.52  (1894.0  examples/sec;  0.068  sec/batch)
step 54620, loss=0.80  (1869.2  examples/sec;  0.068  sec/batch)
step 54630, loss=0.82  (1856.7  examples/sec;  0.069  sec/batch)
step 54640, loss=0.54  (1897.3  examples/sec;  0.067  sec/batch)
step 54650, loss=0.70  (1833.4  examples/sec;  0.070  sec/batch)
step 54660, loss=0.74  (1915.1  examples/sec;  0.067  sec/batch)
step 54670, loss=0.73  (1908.9  examples/sec;  0.067  sec/batch)
step 54680, loss=0.58  (1921.4  examples/sec;  0.067  sec/batch)
step 54690, loss=0.71  (1849.0  examples/sec;  0.069  sec/batch)
step 54700, loss=0.63  (1649.9  examples/sec;  0.078  sec/batch)
step 54710, loss=0.53  (1923.5  examples/sec;  0.067  sec/batch)
step 54720, loss=0.73  (1897.7  examples/sec;  0.067  sec/batch)
step 54730, loss=0.57  (1905.7  examples/sec;  0.067  sec/batch)
step 54740, loss=0.84  (2008.3  examples/sec;  0.064  sec/batch)
step 54750, loss=0.48  (1889.5  examples/sec;  0.068  sec/batch)
step 54760, loss=0.57  (1948.8  examples/sec;  0.066  sec/batch)
step 54770, loss=0.67  (1820.7  examples/sec;  0.070  sec/batch)
step 54780, loss=0.63  (1822.5  examples/sec;  0.070  sec/batch)
step 54790, loss=0.55  (1889.3  examples/sec;  0.068  sec/batch)
step 54800, loss=0.67  (1884.5  examples/sec;  0.068  sec/batch)
step 54810, loss=0.67  (1988.0  examples/sec;  0.064  sec/batch)
step 54820, loss=0.50  (1931.1  examples/sec;  0.066  sec/batch)
step 54830, loss=0.46  (1868.0  examples/sec;  0.069  sec/batch)
step 54840, loss=0.83  (1948.5  examples/sec;  0.066  sec/batch)
step 54850, loss=0.74  (1975.2  examples/sec;  0.065  sec/batch)
step 54860, loss=0.68  (1867.8  examples/sec;  0.069  sec/batch)
step 54870, loss=0.69  (2063.8  examples/sec;  0.062  sec/batch)
step 54880, loss=0.64  (1857.4  examples/sec;  0.069  sec/batch)
step 54890, loss=0.61  (1873.3  examples/sec;  0.068  sec/batch)
step 54900, loss=0.60  (1926.4  examples/sec;  0.066  sec/batch)
step 54910, loss=0.63  (1826.3  examples/sec;  0.070  sec/batch)
step 54920, loss=0.59  (1760.5  examples/sec;  0.073  sec/batch)
step 54930, loss=0.75  (1863.8  examples/sec;  0.069  sec/batch)
step 54940, loss=0.69  (1890.8  examples/sec;  0.068  sec/batch)
step 54950, loss=0.91  (1833.0  examples/sec;  0.070  sec/batch)
step 54960, loss=0.67  (1928.6  examples/sec;  0.066  sec/batch)
step 54970, loss=0.62  (1919.7  examples/sec;  0.067  sec/batch)
step 54980, loss=0.64  (1893.1  examples/sec;  0.068  sec/batch)
step 54990, loss=0.59  (2090.2  examples/sec;  0.061  sec/batch)
step 55000, loss=0.69  (1867.8  examples/sec;  0.069  sec/batch)
step 55010, loss=0.56  (1904.8  examples/sec;  0.067  sec/batch)
step 55020, loss=0.65  (1864.0  examples/sec;  0.069  sec/batch)
step 55030, loss=0.76  (1886.7  examples/sec;  0.068  sec/batch)
step 55040, loss=0.55  (1806.5  examples/sec;  0.071  sec/batch)
step 55050, loss=0.76  (1971.4  examples/sec;  0.065  sec/batch)
step 55060, loss=0.72  (1859.6  examples/sec;  0.069  sec/batch)
step 55070, loss=0.52  (1902.2  examples/sec;  0.067  sec/batch)
step 55080, loss=0.64  (1885.8  examples/sec;  0.068  sec/batch)
step 55090, loss=0.75  (1929.3  examples/sec;  0.066  sec/batch)
step 55100, loss=0.49  (1876.7  examples/sec;  0.068  sec/batch)
step 55110, loss=0.67  (1939.2  examples/sec;  0.066  sec/batch)
step 55120, loss=0.58  (1956.2  examples/sec;  0.065  sec/batch)
step 55130, loss=0.69  (1852.2  examples/sec;  0.069  sec/batch)
step 55140, loss=0.65  (1728.6  examples/sec;  0.074  sec/batch)
step 55150, loss=0.47  (1880.4  examples/sec;  0.068  sec/batch)
step 55160, loss=0.73  (1929.4  examples/sec;  0.066  sec/batch)
step 55170, loss=0.62  (1933.3  examples/sec;  0.066  sec/batch)
step 55180, loss=0.58  (1958.7  examples/sec;  0.065  sec/batch)
step 55190, loss=0.59  (1919.4  examples/sec;  0.067  sec/batch)
step 55200, loss=0.59  (1890.4  examples/sec;  0.068  sec/batch)
step 55210, loss=0.75  (1902.6  examples/sec;  0.067  sec/batch)
step 55220, loss=0.77  (1857.7  examples/sec;  0.069  sec/batch)
step 55230, loss=0.64  (1988.9  examples/sec;  0.064  sec/batch)
step 55240, loss=0.74  (1876.1  examples/sec;  0.068  sec/batch)
step 55250, loss=0.54  (1765.0  examples/sec;  0.073  sec/batch)
step 55260, loss=0.67  (1908.8  examples/sec;  0.067  sec/batch)
step 55270, loss=0.56  (1912.2  examples/sec;  0.067  sec/batch)
step 55280, loss=0.70  (1903.1  examples/sec;  0.067  sec/batch)
step 55290, loss=0.68  (1929.6  examples/sec;  0.066  sec/batch)
step 55300, loss=0.60  (1935.3  examples/sec;  0.066  sec/batch)
step 55310, loss=0.69  (1917.0  examples/sec;  0.067  sec/batch)
step 55320, loss=0.65  (1933.7  examples/sec;  0.066  sec/batch)
step 55330, loss=0.58  (1852.7  examples/sec;  0.069  sec/batch)
step 55340, loss=0.83  (1868.7  examples/sec;  0.068  sec/batch)
step 55350, loss=0.55  (1867.8  examples/sec;  0.069  sec/batch)
step 55360, loss=0.63  (1814.4  examples/sec;  0.071  sec/batch)
step 55370, loss=0.63  (1821.6  examples/sec;  0.070  sec/batch)
step 55380, loss=0.56  (1873.4  examples/sec;  0.068  sec/batch)
step 55390, loss=0.68  (1864.1  examples/sec;  0.069  sec/batch)
step 55400, loss=0.63  (1973.1  examples/sec;  0.065  sec/batch)
step 55410, loss=0.61  (1947.7  examples/sec;  0.066  sec/batch)
step 55420, loss=0.69  (1856.6  examples/sec;  0.069  sec/batch)
step 55430, loss=0.63  (1908.3  examples/sec;  0.067  sec/batch)
step 55440, loss=0.67  (1639.6  examples/sec;  0.078  sec/batch)
step 55450, loss=0.78  (1901.6  examples/sec;  0.067  sec/batch)
step 55460, loss=0.62  (1884.4  examples/sec;  0.068  sec/batch)
step 55470, loss=0.94  (1870.6  examples/sec;  0.068  sec/batch)
step 55480, loss=0.70  (1825.5  examples/sec;  0.070  sec/batch)
step 55490, loss=0.60  (1985.8  examples/sec;  0.064  sec/batch)
step 55500, loss=0.69  (1906.0  examples/sec;  0.067  sec/batch)
step 55510, loss=0.67  (1853.4  examples/sec;  0.069  sec/batch)
step 55520, loss=0.67  (1931.0  examples/sec;  0.066  sec/batch)
step 55530, loss=0.69  (1867.8  examples/sec;  0.069  sec/batch)
step 55540, loss=0.64  (1886.3  examples/sec;  0.068  sec/batch)
step 55550, loss=0.61  (1774.6  examples/sec;  0.072  sec/batch)
step 55560, loss=0.74  (1798.2  examples/sec;  0.071  sec/batch)
step 55570, loss=0.58  (1912.4  examples/sec;  0.067  sec/batch)
step 55580, loss=0.59  (1897.2  examples/sec;  0.067  sec/batch)
step 55590, loss=0.55  (1866.1  examples/sec;  0.069  sec/batch)
step 55600, loss=0.69  (1916.2  examples/sec;  0.067  sec/batch)
step 55610, loss=0.67  (1905.3  examples/sec;  0.067  sec/batch)
step 55620, loss=0.64  (1898.1  examples/sec;  0.067  sec/batch)
step 55630, loss=0.57  (1932.6  examples/sec;  0.066  sec/batch)
step 55640, loss=0.63  (1903.5  examples/sec;  0.067  sec/batch)
step 55650, loss=0.57  (1745.5  examples/sec;  0.073  sec/batch)
step 55660, loss=0.64  (1955.0  examples/sec;  0.065  sec/batch)
step 55670, loss=0.64  (1913.8  examples/sec;  0.067  sec/batch)
step 55680, loss=0.57  (1948.8  examples/sec;  0.066  sec/batch)
step 55690, loss=0.49  (1862.3  examples/sec;  0.069  sec/batch)
step 55700, loss=0.72  (1906.4  examples/sec;  0.067  sec/batch)
step 55710, loss=0.64  (1911.8  examples/sec;  0.067  sec/batch)
step 55720, loss=0.80  (1877.0  examples/sec;  0.068  sec/batch)
step 55730, loss=0.58  (1899.1  examples/sec;  0.067  sec/batch)
step 55740, loss=0.55  (1935.8  examples/sec;  0.066  sec/batch)
step 55750, loss=0.59  (1918.6  examples/sec;  0.067  sec/batch)
step 55760, loss=0.64  (1956.6  examples/sec;  0.065  sec/batch)
step 55770, loss=0.61  (1781.9  examples/sec;  0.072  sec/batch)
step 55780, loss=0.68  (1884.5  examples/sec;  0.068  sec/batch)
step 55790, loss=0.71  (1966.6  examples/sec;  0.065  sec/batch)
step 55800, loss=0.59  (1902.4  examples/sec;  0.067  sec/batch)
step 55810, loss=0.56  (1944.0  examples/sec;  0.066  sec/batch)
step 55820, loss=0.69  (1920.3  examples/sec;  0.067  sec/batch)
step 55830, loss=0.61  (1873.5  examples/sec;  0.068  sec/batch)
step 55840, loss=0.59  (1962.8  examples/sec;  0.065  sec/batch)
step 55850, loss=0.75  (1826.0  examples/sec;  0.070  sec/batch)
step 55860, loss=0.67  (1941.8  examples/sec;  0.066  sec/batch)
step 55870, loss=0.69  (1821.9  examples/sec;  0.070  sec/batch)
step 55880, loss=0.68  (1493.3  examples/sec;  0.086  sec/batch)
step 55890, loss=0.71  (1872.5  examples/sec;  0.068  sec/batch)
step 55900, loss=0.70  (1858.1  examples/sec;  0.069  sec/batch)
step 55910, loss=0.54  (1912.0  examples/sec;  0.067  sec/batch)
step 55920, loss=0.81  (1875.4  examples/sec;  0.068  sec/batch)
step 55930, loss=0.60  (1948.0  examples/sec;  0.066  sec/batch)
step 55940, loss=0.57  (1936.5  examples/sec;  0.066  sec/batch)
step 55950, loss=0.73  (1867.9  examples/sec;  0.069  sec/batch)
step 55960, loss=0.68  (1907.2  examples/sec;  0.067  sec/batch)
step 55970, loss=0.56  (1884.3  examples/sec;  0.068  sec/batch)
step 55980, loss=0.60  (1880.3  examples/sec;  0.068  sec/batch)
step 55990, loss=0.53  (1948.1  examples/sec;  0.066  sec/batch)
step 56000, loss=0.80  (1683.4  examples/sec;  0.076  sec/batch)
step 56010, loss=0.72  (1836.0  examples/sec;  0.070  sec/batch)
step 56020, loss=0.50  (1895.8  examples/sec;  0.068  sec/batch)
step 56030, loss=0.70  (1926.7  examples/sec;  0.066  sec/batch)
step 56040, loss=0.62  (1837.6  examples/sec;  0.070  sec/batch)
step 56050, loss=0.98  (1791.4  examples/sec;  0.071  sec/batch)
step 56060, loss=0.62  (1927.5  examples/sec;  0.066  sec/batch)
step 56070, loss=0.74  (1930.1  examples/sec;  0.066  sec/batch)
step 56080, loss=0.81  (1955.6  examples/sec;  0.065  sec/batch)
step 56090, loss=0.63  (1959.6  examples/sec;  0.065  sec/batch)
step 56100, loss=0.77  (1844.6  examples/sec;  0.069  sec/batch)
step 56110, loss=0.62  (1914.3  examples/sec;  0.067  sec/batch)
step 56120, loss=0.72  (1919.0  examples/sec;  0.067  sec/batch)
step 56130, loss=0.68  (1999.4  examples/sec;  0.064  sec/batch)
step 56140, loss=0.76  (1912.0  examples/sec;  0.067  sec/batch)
step 56150, loss=0.63  (1907.8  examples/sec;  0.067  sec/batch)
step 56160, loss=0.77  (1876.4  examples/sec;  0.068  sec/batch)
step 56170, loss=0.52  (1921.3  examples/sec;  0.067  sec/batch)
step 56180, loss=0.63  (1925.9  examples/sec;  0.066  sec/batch)
step 56190, loss=0.58  (1855.9  examples/sec;  0.069  sec/batch)
step 56200, loss=0.63  (1891.9  examples/sec;  0.068  sec/batch)
step 56210, loss=0.55  (1903.6  examples/sec;  0.067  sec/batch)
step 56220, loss=0.70  (1851.4  examples/sec;  0.069  sec/batch)
step 56230, loss=0.65  (1961.9  examples/sec;  0.065  sec/batch)
step 56240, loss=0.62  (1835.1  examples/sec;  0.070  sec/batch)
step 56250, loss=0.67  (1699.9  examples/sec;  0.075  sec/batch)
step 56260, loss=0.67  (1853.5  examples/sec;  0.069  sec/batch)
step 56270, loss=0.72  (1969.9  examples/sec;  0.065  sec/batch)
step 56280, loss=0.70  (1936.2  examples/sec;  0.066  sec/batch)
step 56290, loss=0.72  (1872.9  examples/sec;  0.068  sec/batch)
step 56300, loss=0.64  (1904.2  examples/sec;  0.067  sec/batch)
step 56310, loss=0.75  (1887.5  examples/sec;  0.068  sec/batch)
step 56320, loss=0.63  (1870.0  examples/sec;  0.068  sec/batch)
step 56330, loss=0.68  (1947.6  examples/sec;  0.066  sec/batch)
step 56340, loss=0.47  (1970.0  examples/sec;  0.065  sec/batch)
step 56350, loss=0.63  (1884.6  examples/sec;  0.068  sec/batch)
step 56360, loss=0.73  (1777.3  examples/sec;  0.072  sec/batch)
step 56370, loss=0.57  (2082.0  examples/sec;  0.061  sec/batch)
step 56380, loss=0.62  (1878.1  examples/sec;  0.068  sec/batch)
step 56390, loss=0.47  (1882.1  examples/sec;  0.068  sec/batch)
step 56400, loss=0.55  (1933.8  examples/sec;  0.066  sec/batch)
step 56410, loss=0.66  (1910.1  examples/sec;  0.067  sec/batch)
step 56420, loss=0.48  (1900.2  examples/sec;  0.067  sec/batch)
step 56430, loss=0.51  (1932.4  examples/sec;  0.066  sec/batch)
step 56440, loss=0.58  (1890.3  examples/sec;  0.068  sec/batch)
step 56450, loss=0.48  (1965.3  examples/sec;  0.065  sec/batch)
step 56460, loss=0.54  (1927.2  examples/sec;  0.066  sec/batch)
step 56470, loss=0.60  (1903.2  examples/sec;  0.067  sec/batch)
step 56480, loss=0.69  (1681.9  examples/sec;  0.076  sec/batch)
step 56490, loss=0.69  (1942.5  examples/sec;  0.066  sec/batch)
step 56500, loss=0.75  (1976.7  examples/sec;  0.065  sec/batch)
step 56510, loss=0.64  (1999.8  examples/sec;  0.064  sec/batch)
step 56520, loss=0.65  (1885.0  examples/sec;  0.068  sec/batch)
step 56530, loss=0.47  (1918.1  examples/sec;  0.067  sec/batch)
step 56540, loss=0.70  (1897.4  examples/sec;  0.067  sec/batch)
step 56550, loss=0.66  (1626.7  examples/sec;  0.079  sec/batch)
step 56560, loss=0.75  (1881.5  examples/sec;  0.068  sec/batch)
step 56570, loss=0.53  (1859.4  examples/sec;  0.069  sec/batch)
step 56580, loss=0.50  (1906.8  examples/sec;  0.067  sec/batch)
step 56590, loss=0.67  (1842.3  examples/sec;  0.069  sec/batch)
step 56600, loss=0.81  (1861.0  examples/sec;  0.069  sec/batch)
step 56610, loss=0.79  (1864.7  examples/sec;  0.069  sec/batch)
step 56620, loss=0.58  (1860.1  examples/sec;  0.069  sec/batch)
step 56630, loss=0.70  (1887.8  examples/sec;  0.068  sec/batch)
step 56640, loss=0.71  (1836.0  examples/sec;  0.070  sec/batch)
step 56650, loss=0.48  (1925.8  examples/sec;  0.066  sec/batch)
step 56660, loss=0.57  (1602.1  examples/sec;  0.080  sec/batch)
step 56670, loss=0.53  (1890.2  examples/sec;  0.068  sec/batch)
step 56680, loss=0.76  (1918.6  examples/sec;  0.067  sec/batch)
step 56690, loss=0.77  (1949.4  examples/sec;  0.066  sec/batch)
step 56700, loss=0.64  (1887.6  examples/sec;  0.068  sec/batch)
step 56710, loss=0.57  (1850.7  examples/sec;  0.069  sec/batch)
step 56720, loss=0.67  (1938.4  examples/sec;  0.066  sec/batch)
step 56730, loss=0.70  (1847.4  examples/sec;  0.069  sec/batch)
step 56740, loss=0.68  (1879.1  examples/sec;  0.068  sec/batch)
step 56750, loss=0.52  (1871.9  examples/sec;  0.068  sec/batch)
step 56760, loss=0.55  (1847.6  examples/sec;  0.069  sec/batch)
step 56770, loss=0.73  (2018.5  examples/sec;  0.063  sec/batch)
step 56780, loss=0.64  (1937.0  examples/sec;  0.066  sec/batch)
step 56790, loss=0.62  (1875.4  examples/sec;  0.068  sec/batch)
step 56800, loss=0.63  (1914.8  examples/sec;  0.067  sec/batch)
step 56810, loss=0.74  (1796.8  examples/sec;  0.071  sec/batch)
step 56820, loss=0.63  (1986.8  examples/sec;  0.064  sec/batch)
step 56830, loss=0.62  (1897.5  examples/sec;  0.067  sec/batch)
step 56840, loss=0.69  (1891.1  examples/sec;  0.068  sec/batch)
step 56850, loss=0.76  (1878.3  examples/sec;  0.068  sec/batch)
step 56860, loss=0.70  (1885.1  examples/sec;  0.068  sec/batch)
step 56870, loss=0.64  (1929.9  examples/sec;  0.066  sec/batch)
step 56880, loss=0.58  (1945.3  examples/sec;  0.066  sec/batch)
step 56890, loss=0.52  (1939.4  examples/sec;  0.066  sec/batch)
step 56900, loss=0.67  (1919.4  examples/sec;  0.067  sec/batch)
step 56910, loss=0.73  (1860.9  examples/sec;  0.069  sec/batch)
step 56920, loss=0.76  (1932.3  examples/sec;  0.066  sec/batch)
step 56930, loss=0.52  (1852.2  examples/sec;  0.069  sec/batch)
step 56940, loss=0.61  (1909.2  examples/sec;  0.067  sec/batch)
step 56950, loss=0.80  (1813.1  examples/sec;  0.071  sec/batch)
step 56960, loss=0.73  (1917.6  examples/sec;  0.067  sec/batch)
step 56970, loss=0.54  (1886.1  examples/sec;  0.068  sec/batch)
step 56980, loss=0.59  (1865.8  examples/sec;  0.069  sec/batch)
step 56990, loss=0.50  (1888.0  examples/sec;  0.068  sec/batch)
step 57000, loss=0.56  (1920.1  examples/sec;  0.067  sec/batch)
step 57010, loss=0.61  (1799.8  examples/sec;  0.071  sec/batch)
step 57020, loss=0.66  (1826.8  examples/sec;  0.070  sec/batch)
step 57030, loss=0.51  (1956.8  examples/sec;  0.065  sec/batch)
step 57040, loss=0.73  (1758.2  examples/sec;  0.073  sec/batch)
step 57050, loss=0.62  (1926.3  examples/sec;  0.066  sec/batch)
step 57060, loss=0.79  (1912.2  examples/sec;  0.067  sec/batch)
step 57070, loss=0.56  (1993.0  examples/sec;  0.064  sec/batch)
step 57080, loss=0.76  (1884.8  examples/sec;  0.068  sec/batch)
step 57090, loss=0.69  (1834.3  examples/sec;  0.070  sec/batch)
step 57100, loss=0.64  (1918.9  examples/sec;  0.067  sec/batch)
step 57110, loss=0.66  (1935.6  examples/sec;  0.066  sec/batch)
step 57120, loss=0.72  (1872.4  examples/sec;  0.068  sec/batch)
step 57130, loss=0.69  (1928.0  examples/sec;  0.066  sec/batch)
step 57140, loss=0.44  (1848.8  examples/sec;  0.069  sec/batch)
step 57150, loss=0.65  (1923.5  examples/sec;  0.067  sec/batch)
step 57160, loss=0.54  (1999.1  examples/sec;  0.064  sec/batch)
step 57170, loss=0.74  (1748.0  examples/sec;  0.073  sec/batch)
step 57180, loss=0.60  (1900.5  examples/sec;  0.067  sec/batch)
step 57190, loss=0.62  (1908.2  examples/sec;  0.067  sec/batch)
step 57200, loss=0.67  (1931.9  examples/sec;  0.066  sec/batch)
step 57210, loss=0.66  (1961.8  examples/sec;  0.065  sec/batch)
step 57220, loss=0.76  (1933.1  examples/sec;  0.066  sec/batch)
step 57230, loss=0.58  (1912.0  examples/sec;  0.067  sec/batch)
step 57240, loss=0.87  (1841.3  examples/sec;  0.070  sec/batch)
step 57250, loss=0.70  (1845.5  examples/sec;  0.069  sec/batch)
step 57260, loss=0.64  (1930.6  examples/sec;  0.066  sec/batch)
step 57270, loss=0.58  (1853.7  examples/sec;  0.069  sec/batch)
step 57280, loss=0.67  (1931.7  examples/sec;  0.066  sec/batch)
step 57290, loss=0.69  (1944.0  examples/sec;  0.066  sec/batch)
step 57300, loss=0.65  (1893.9  examples/sec;  0.068  sec/batch)
step 57310, loss=0.55  (1885.0  examples/sec;  0.068  sec/batch)
step 57320, loss=0.71  (1891.5  examples/sec;  0.068  sec/batch)
step 57330, loss=0.52  (1843.7  examples/sec;  0.069  sec/batch)
step 57340, loss=0.60  (1879.1  examples/sec;  0.068  sec/batch)
step 57350, loss=0.76  (1910.2  examples/sec;  0.067  sec/batch)
step 57360, loss=0.55  (1983.2  examples/sec;  0.065  sec/batch)
step 57370, loss=0.61  (1821.6  examples/sec;  0.070  sec/batch)
step 57380, loss=0.74  (1902.1  examples/sec;  0.067  sec/batch)
step 57390, loss=0.62  (1915.0  examples/sec;  0.067  sec/batch)
step 57400, loss=0.67  (1930.1  examples/sec;  0.066  sec/batch)
step 57410, loss=0.70  (1962.8  examples/sec;  0.065  sec/batch)
step 57420, loss=0.69  (1882.7  examples/sec;  0.068  sec/batch)
step 57430, loss=0.67  (1858.7  examples/sec;  0.069  sec/batch)
step 57440, loss=0.63  (1851.0  examples/sec;  0.069  sec/batch)
step 57450, loss=0.53  (1823.0  examples/sec;  0.070  sec/batch)
step 57460, loss=0.65  (1915.2  examples/sec;  0.067  sec/batch)
step 57470, loss=0.69  (1960.9  examples/sec;  0.065  sec/batch)
step 57480, loss=0.54  (1997.0  examples/sec;  0.064  sec/batch)
step 57490, loss=0.67  (1836.0  examples/sec;  0.070  sec/batch)
step 57500, loss=0.56  (1872.4  examples/sec;  0.068  sec/batch)
step 57510, loss=0.54  (1845.3  examples/sec;  0.069  sec/batch)
step 57520, loss=0.51  (1920.6  examples/sec;  0.067  sec/batch)
step 57530, loss=0.57  (1904.3  examples/sec;  0.067  sec/batch)
step 57540, loss=0.65  (1874.4  examples/sec;  0.068  sec/batch)
step 57550, loss=0.59  (1919.5  examples/sec;  0.067  sec/batch)
step 57560, loss=0.64  (1878.5  examples/sec;  0.068  sec/batch)
step 57570, loss=0.74  (1896.5  examples/sec;  0.067  sec/batch)
step 57580, loss=0.59  (1922.6  examples/sec;  0.067  sec/batch)
step 57590, loss=0.72  (1939.6  examples/sec;  0.066  sec/batch)
step 57600, loss=0.67  (1887.0  examples/sec;  0.068  sec/batch)
step 57610, loss=0.55  (1986.9  examples/sec;  0.064  sec/batch)
step 57620, loss=0.63  (1632.6  examples/sec;  0.078  sec/batch)
step 57630, loss=0.71  (1927.0  examples/sec;  0.066  sec/batch)
step 57640, loss=0.64  (1993.4  examples/sec;  0.064  sec/batch)
step 57650, loss=0.69  (1862.7  examples/sec;  0.069  sec/batch)
step 57660, loss=0.70  (1876.6  examples/sec;  0.068  sec/batch)
step 57670, loss=0.59  (1695.9  examples/sec;  0.075  sec/batch)
step 57680, loss=0.64  (1900.8  examples/sec;  0.067  sec/batch)
step 57690, loss=0.47  (1967.9  examples/sec;  0.065  sec/batch)
step 57700, loss=0.68  (1813.8  examples/sec;  0.071  sec/batch)
step 57710, loss=0.69  (1925.7  examples/sec;  0.066  sec/batch)
step 57720, loss=0.58  (1954.4  examples/sec;  0.065  sec/batch)
step 57730, loss=0.65  (1954.1  examples/sec;  0.066  sec/batch)
step 57740, loss=0.62  (1890.6  examples/sec;  0.068  sec/batch)
step 57750, loss=0.60  (1853.9  examples/sec;  0.069  sec/batch)
step 57760, loss=0.50  (1908.2  examples/sec;  0.067  sec/batch)
step 57770, loss=0.62  (1880.4  examples/sec;  0.068  sec/batch)
step 57780, loss=0.52  (1778.2  examples/sec;  0.072  sec/batch)
step 57790, loss=0.57  (1900.2  examples/sec;  0.067  sec/batch)
step 57800, loss=0.62  (1925.0  examples/sec;  0.066  sec/batch)
step 57810, loss=0.62  (2012.6  examples/sec;  0.064  sec/batch)
step 57820, loss=0.53  (1839.5  examples/sec;  0.070  sec/batch)
step 57830, loss=0.57  (1908.3  examples/sec;  0.067  sec/batch)
step 57840, loss=0.68  (1884.0  examples/sec;  0.068  sec/batch)
step 57850, loss=0.55  (1919.8  examples/sec;  0.067  sec/batch)
step 57860, loss=0.62  (1891.0  examples/sec;  0.068  sec/batch)
step 57870, loss=0.63  (1812.1  examples/sec;  0.071  sec/batch)
step 57880, loss=0.79  (1919.7  examples/sec;  0.067  sec/batch)
step 57890, loss=0.52  (1867.6  examples/sec;  0.069  sec/batch)
step 57900, loss=0.54  (1873.6  examples/sec;  0.068  sec/batch)
step 57910, loss=0.56  (1938.0  examples/sec;  0.066  sec/batch)
step 57920, loss=0.54  (1889.8  examples/sec;  0.068  sec/batch)
step 57930, loss=0.73  (1959.0  examples/sec;  0.065  sec/batch)
step 57940, loss=0.63  (1928.7  examples/sec;  0.066  sec/batch)
step 57950, loss=0.69  (1862.8  examples/sec;  0.069  sec/batch)
step 57960, loss=0.66  (1928.1  examples/sec;  0.066  sec/batch)
step 57970, loss=0.78  (1858.8  examples/sec;  0.069  sec/batch)
step 57980, loss=0.63  (1729.7  examples/sec;  0.074  sec/batch)
step 57990, loss=0.65  (1926.0  examples/sec;  0.066  sec/batch)
step 58000, loss=0.66  (1986.2  examples/sec;  0.064  sec/batch)
step 58010, loss=0.65  (1923.4  examples/sec;  0.067  sec/batch)
step 58020, loss=0.51  (1887.0  examples/sec;  0.068  sec/batch)
step 58030, loss=0.58  (1920.6  examples/sec;  0.067  sec/batch)
step 58040, loss=0.74  (2013.0  examples/sec;  0.064  sec/batch)
step 58050, loss=0.66  (1913.8  examples/sec;  0.067  sec/batch)
step 58060, loss=0.60  (1922.9  examples/sec;  0.067  sec/batch)
step 58070, loss=0.58  (1883.7  examples/sec;  0.068  sec/batch)
step 58080, loss=0.47  (2029.9  examples/sec;  0.063  sec/batch)
step 58090, loss=0.63  (1903.1  examples/sec;  0.067  sec/batch)
step 58100, loss=0.72  (1950.0  examples/sec;  0.066  sec/batch)
step 58110, loss=0.69  (1899.6  examples/sec;  0.067  sec/batch)
step 58120, loss=0.58  (1607.6  examples/sec;  0.080  sec/batch)
step 58130, loss=0.60  (1851.3  examples/sec;  0.069  sec/batch)
step 58140, loss=0.75  (1871.3  examples/sec;  0.068  sec/batch)
step 58150, loss=0.84  (1919.2  examples/sec;  0.067  sec/batch)
step 58160, loss=0.67  (1622.7  examples/sec;  0.079  sec/batch)
step 58170, loss=0.58  (1915.7  examples/sec;  0.067  sec/batch)
step 58180, loss=0.52  (1908.0  examples/sec;  0.067  sec/batch)
step 58190, loss=0.68  (1892.1  examples/sec;  0.068  sec/batch)
step 58200, loss=0.84  (1916.7  examples/sec;  0.067  sec/batch)
step 58210, loss=0.50  (1957.0  examples/sec;  0.065  sec/batch)
step 58220, loss=0.75  (1754.4  examples/sec;  0.073  sec/batch)
step 58230, loss=0.56  (1877.0  examples/sec;  0.068  sec/batch)
step 58240, loss=0.70  (1816.9  examples/sec;  0.070  sec/batch)
step 58250, loss=0.84  (1930.4  examples/sec;  0.066  sec/batch)
step 58260, loss=0.65  (1899.3  examples/sec;  0.067  sec/batch)
step 58270, loss=0.62  (1902.9  examples/sec;  0.067  sec/batch)
step 58280, loss=0.68  (1960.8  examples/sec;  0.065  sec/batch)
step 58290, loss=0.50  (2037.6  examples/sec;  0.063  sec/batch)
step 58300, loss=0.66  (1910.3  examples/sec;  0.067  sec/batch)
step 58310, loss=0.65  (1921.4  examples/sec;  0.067  sec/batch)
step 58320, loss=0.61  (1848.3  examples/sec;  0.069  sec/batch)
step 58330, loss=0.62  (1871.0  examples/sec;  0.068  sec/batch)
step 58340, loss=0.49  (1869.4  examples/sec;  0.068  sec/batch)
step 58350, loss=0.57  (1932.1  examples/sec;  0.066  sec/batch)
step 58360, loss=0.78  (1945.7  examples/sec;  0.066  sec/batch)
step 58370, loss=0.65  (1895.4  examples/sec;  0.068  sec/batch)
step 58380, loss=0.76  (1950.7  examples/sec;  0.066  sec/batch)
step 58390, loss=0.70  (1928.5  examples/sec;  0.066  sec/batch)
step 58400, loss=0.53  (1890.1  examples/sec;  0.068  sec/batch)
step 58410, loss=0.67  (1889.7  examples/sec;  0.068  sec/batch)
step 58420, loss=0.48  (1866.2  examples/sec;  0.069  sec/batch)
step 58430, loss=0.73  (1989.2  examples/sec;  0.064  sec/batch)
step 58440, loss=0.70  (1979.4  examples/sec;  0.065  sec/batch)
step 58450, loss=0.69  (1907.4  examples/sec;  0.067  sec/batch)
step 58460, loss=0.51  (1888.9  examples/sec;  0.068  sec/batch)
step 58470, loss=0.71  (1976.9  examples/sec;  0.065  sec/batch)
step 58480, loss=0.56  (1957.2  examples/sec;  0.065  sec/batch)
step 58490, loss=0.65  (1903.9  examples/sec;  0.067  sec/batch)
step 58500, loss=0.66  (1823.1  examples/sec;  0.070  sec/batch)
step 58510, loss=0.72  (1860.6  examples/sec;  0.069  sec/batch)
step 58520, loss=0.59  (1826.7  examples/sec;  0.070  sec/batch)
step 58530, loss=0.58  (1844.6  examples/sec;  0.069  sec/batch)
step 58540, loss=0.43  (1842.0  examples/sec;  0.069  sec/batch)
step 58550, loss=0.70  (1823.6  examples/sec;  0.070  sec/batch)
step 58560, loss=0.64  (1937.2  examples/sec;  0.066  sec/batch)
step 58570, loss=0.59  (1871.8  examples/sec;  0.068  sec/batch)
step 58580, loss=0.45  (1941.9  examples/sec;  0.066  sec/batch)
step 58590, loss=0.75  (1902.4  examples/sec;  0.067  sec/batch)
step 58600, loss=0.64  (1967.1  examples/sec;  0.065  sec/batch)
step 58610, loss=0.56  (1845.6  examples/sec;  0.069  sec/batch)
step 58620, loss=0.66  (1888.9  examples/sec;  0.068  sec/batch)
step 58630, loss=0.58  (1929.6  examples/sec;  0.066  sec/batch)
step 58640, loss=0.71  (1861.7  examples/sec;  0.069  sec/batch)
step 58650, loss=0.76  (1935.9  examples/sec;  0.066  sec/batch)
step 58660, loss=0.64  (1916.1  examples/sec;  0.067  sec/batch)
step 58670, loss=0.61  (1881.2  examples/sec;  0.068  sec/batch)
step 58680, loss=0.69  (1916.1  examples/sec;  0.067  sec/batch)
step 58690, loss=0.66  (1830.0  examples/sec;  0.070  sec/batch)
step 58700, loss=0.55  (1867.0  examples/sec;  0.069  sec/batch)
step 58710, loss=0.63  (1932.5  examples/sec;  0.066  sec/batch)
step 58720, loss=0.66  (1957.7  examples/sec;  0.065  sec/batch)
step 58730, loss=0.45  (1633.9  examples/sec;  0.078  sec/batch)
step 58740, loss=0.55  (1920.8  examples/sec;  0.067  sec/batch)
step 58750, loss=0.70  (1936.5  examples/sec;  0.066  sec/batch)
step 58760, loss=0.49  (1684.1  examples/sec;  0.076  sec/batch)
step 58770, loss=0.72  (1807.9  examples/sec;  0.071  sec/batch)
step 58780, loss=0.78  (1859.9  examples/sec;  0.069  sec/batch)
step 58790, loss=0.52  (1805.7  examples/sec;  0.071  sec/batch)
step 58800, loss=0.71  (1807.8  examples/sec;  0.071  sec/batch)
step 58810, loss=0.58  (1891.6  examples/sec;  0.068  sec/batch)
step 58820, loss=0.66  (1858.1  examples/sec;  0.069  sec/batch)
step 58830, loss=0.50  (1708.1  examples/sec;  0.075  sec/batch)
step 58840, loss=0.60  (1896.1  examples/sec;  0.068  sec/batch)
step 58850, loss=0.56  (1893.3  examples/sec;  0.068  sec/batch)
step 58860, loss=0.87  (1867.0  examples/sec;  0.069  sec/batch)
step 58870, loss=0.55  (1982.8  examples/sec;  0.065  sec/batch)
step 58880, loss=0.65  (1894.1  examples/sec;  0.068  sec/batch)
step 58890, loss=0.57  (1752.4  examples/sec;  0.073  sec/batch)
step 58900, loss=0.68  (1910.9  examples/sec;  0.067  sec/batch)
step 58910, loss=0.73  (1851.0  examples/sec;  0.069  sec/batch)
step 58920, loss=0.68  (1922.2  examples/sec;  0.067  sec/batch)
step 58930, loss=0.68  (1958.5  examples/sec;  0.065  sec/batch)
step 58940, loss=0.57  (1891.0  examples/sec;  0.068  sec/batch)
step 58950, loss=0.61  (1913.0  examples/sec;  0.067  sec/batch)
step 58960, loss=0.51  (1935.9  examples/sec;  0.066  sec/batch)
step 58970, loss=0.69  (1931.2  examples/sec;  0.066  sec/batch)
step 58980, loss=0.66  (1989.4  examples/sec;  0.064  sec/batch)
step 58990, loss=0.63  (1874.5  examples/sec;  0.068  sec/batch)
step 59000, loss=0.59  (1827.0  examples/sec;  0.070  sec/batch)
step 59010, loss=0.76  (1929.5  examples/sec;  0.066  sec/batch)
step 59020, loss=0.47  (1908.8  examples/sec;  0.067  sec/batch)
step 59030, loss=0.47  (1903.2  examples/sec;  0.067  sec/batch)
step 59040, loss=0.59  (1848.3  examples/sec;  0.069  sec/batch)
step 59050, loss=0.61  (1836.7  examples/sec;  0.070  sec/batch)
step 59060, loss=0.65  (1890.7  examples/sec;  0.068  sec/batch)
step 59070, loss=0.73  (1726.9  examples/sec;  0.074  sec/batch)
step 59080, loss=0.65  (1884.3  examples/sec;  0.068  sec/batch)
step 59090, loss=0.55  (1832.9  examples/sec;  0.070  sec/batch)
step 59100, loss=0.65  (1947.0  examples/sec;  0.066  sec/batch)
step 59110, loss=0.78  (1771.2  examples/sec;  0.072  sec/batch)
step 59120, loss=0.59  (1906.9  examples/sec;  0.067  sec/batch)
step 59130, loss=0.71  (1940.1  examples/sec;  0.066  sec/batch)
step 59140, loss=0.63  (1883.8  examples/sec;  0.068  sec/batch)
step 59150, loss=0.64  (1890.9  examples/sec;  0.068  sec/batch)
step 59160, loss=0.58  (1957.9  examples/sec;  0.065  sec/batch)
step 59170, loss=0.68  (1912.0  examples/sec;  0.067  sec/batch)
step 59180, loss=0.56  (1885.2  examples/sec;  0.068  sec/batch)
step 59190, loss=0.71  (1858.1  examples/sec;  0.069  sec/batch)
step 59200, loss=0.63  (1930.2  examples/sec;  0.066  sec/batch)
step 59210, loss=0.69  (1900.4  examples/sec;  0.067  sec/batch)
step 59220, loss=0.73  (1888.8  examples/sec;  0.068  sec/batch)
step 59230, loss=0.75  (1977.2  examples/sec;  0.065  sec/batch)
step 59240, loss=0.62  (1884.7  examples/sec;  0.068  sec/batch)
step 59250, loss=0.64  (1901.3  examples/sec;  0.067  sec/batch)
step 59260, loss=0.76  (1936.7  examples/sec;  0.066  sec/batch)
step 59270, loss=0.76  (1895.5  examples/sec;  0.068  sec/batch)
step 59280, loss=0.53  (1876.2  examples/sec;  0.068  sec/batch)
step 59290, loss=0.66  (1868.0  examples/sec;  0.069  sec/batch)
step 59300, loss=0.63  (1847.4  examples/sec;  0.069  sec/batch)
step 59310, loss=0.57  (1903.1  examples/sec;  0.067  sec/batch)
step 59320, loss=0.69  (1915.1  examples/sec;  0.067  sec/batch)
step 59330, loss=0.56  (1866.0  examples/sec;  0.069  sec/batch)
step 59340, loss=0.57  (1928.8  examples/sec;  0.066  sec/batch)
step 59350, loss=0.80  (1965.0  examples/sec;  0.065  sec/batch)
step 59360, loss=0.87  (1656.3  examples/sec;  0.077  sec/batch)
step 59370, loss=0.60  (1895.0  examples/sec;  0.068  sec/batch)
step 59380, loss=0.64  (1868.1  examples/sec;  0.069  sec/batch)
step 59390, loss=0.52  (1813.2  examples/sec;  0.071  sec/batch)
step 59400, loss=0.81  (1811.7  examples/sec;  0.071  sec/batch)
step 59410, loss=0.64  (1899.0  examples/sec;  0.067  sec/batch)
step 59420, loss=0.70  (1802.3  examples/sec;  0.071  sec/batch)
step 59430, loss=0.51  (1947.3  examples/sec;  0.066  sec/batch)
step 59440, loss=0.67  (1911.5  examples/sec;  0.067  sec/batch)
step 59450, loss=0.75  (2014.4  examples/sec;  0.064  sec/batch)
step 59460, loss=0.73  (1942.4  examples/sec;  0.066  sec/batch)
step 59470, loss=0.57  (1891.9  examples/sec;  0.068  sec/batch)
step 59480, loss=0.55  (1937.2  examples/sec;  0.066  sec/batch)
step 59490, loss=0.71  (1935.5  examples/sec;  0.066  sec/batch)
step 59500, loss=0.62  (1933.6  examples/sec;  0.066  sec/batch)
step 59510, loss=0.52  (1807.0  examples/sec;  0.071  sec/batch)
step 59520, loss=0.66  (1822.8  examples/sec;  0.070  sec/batch)
step 59530, loss=0.52  (1892.5  examples/sec;  0.068  sec/batch)
step 59540, loss=0.61  (1939.3  examples/sec;  0.066  sec/batch)
step 59550, loss=0.69  (1959.6  examples/sec;  0.065  sec/batch)
step 59560, loss=0.71  (1949.2  examples/sec;  0.066  sec/batch)
step 59570, loss=0.60  (1795.7  examples/sec;  0.071  sec/batch)
step 59580, loss=0.69  (1882.3  examples/sec;  0.068  sec/batch)
step 59590, loss=0.57  (1954.5  examples/sec;  0.065  sec/batch)
step 59600, loss=0.59  (1967.3  examples/sec;  0.065  sec/batch)
step 59610, loss=0.77  (1942.1  examples/sec;  0.066  sec/batch)
step 59620, loss=0.59  (1796.8  examples/sec;  0.071  sec/batch)
step 59630, loss=0.74  (1929.4  examples/sec;  0.066  sec/batch)
step 59640, loss=0.56  (1978.7  examples/sec;  0.065  sec/batch)
step 59650, loss=0.75  (1975.7  examples/sec;  0.065  sec/batch)
step 59660, loss=0.57  (1834.0  examples/sec;  0.070  sec/batch)
step 59670, loss=0.57  (1946.4  examples/sec;  0.066  sec/batch)
step 59680, loss=0.66  (1947.5  examples/sec;  0.066  sec/batch)
step 59690, loss=0.63  (1873.1  examples/sec;  0.068  sec/batch)
step 59700, loss=0.65  (1895.0  examples/sec;  0.068  sec/batch)
step 59710, loss=0.72  (1917.5  examples/sec;  0.067  sec/batch)
step 59720, loss=0.57  (1840.4  examples/sec;  0.070  sec/batch)
step 59730, loss=0.63  (1826.9  examples/sec;  0.070  sec/batch)
step 59740, loss=0.55  (1889.5  examples/sec;  0.068  sec/batch)
step 59750, loss=0.84  (1893.4  examples/sec;  0.068  sec/batch)
step 59760, loss=0.86  (1910.3  examples/sec;  0.067  sec/batch)
step 59770, loss=0.48  (1943.9  examples/sec;  0.066  sec/batch)
step 59780, loss=0.84  (1879.2  examples/sec;  0.068  sec/batch)
step 59790, loss=0.54  (1906.7  examples/sec;  0.067  sec/batch)
step 59800, loss=0.66  (1885.6  examples/sec;  0.068  sec/batch)
step 59810, loss=0.60  (1902.5  examples/sec;  0.067  sec/batch)
step 59820, loss=0.56  (1879.9  examples/sec;  0.068  sec/batch)
step 59830, loss=0.68  (1793.5  examples/sec;  0.071  sec/batch)
step 59840, loss=0.66  (1876.3  examples/sec;  0.068  sec/batch)
step 59850, loss=0.69  (1994.0  examples/sec;  0.064  sec/batch)
step 59860, loss=0.60  (1854.5  examples/sec;  0.069  sec/batch)
step 59870, loss=0.71  (2033.4  examples/sec;  0.063  sec/batch)
step 59880, loss=0.70  (1890.3  examples/sec;  0.068  sec/batch)
step 59890, loss=0.65  (1916.1  examples/sec;  0.067  sec/batch)
step 59900, loss=0.66  (1853.9  examples/sec;  0.069  sec/batch)
step 59910, loss=0.63  (1855.5  examples/sec;  0.069  sec/batch)
step 59920, loss=0.63  (1970.9  examples/sec;  0.065  sec/batch)
step 59930, loss=0.58  (1926.9  examples/sec;  0.066  sec/batch)
step 59940, loss=0.66  (1884.1  examples/sec;  0.068  sec/batch)
step 59950, loss=0.54  (1938.3  examples/sec;  0.066  sec/batch)
step 59960, loss=0.75  (1908.7  examples/sec;  0.067  sec/batch)
step 59970, loss=0.72  (1682.4  examples/sec;  0.076  sec/batch)
step 59980, loss=0.66  (1895.0  examples/sec;  0.068  sec/batch)
step 59990, loss=0.54  (1885.6  examples/sec;  0.068  sec/batch)
step 60000, loss=0.53  (1915.2  examples/sec;  0.067  sec/batch)
step 60010, loss=0.65  (2000.7  examples/sec;  0.064  sec/batch)
step 60020, loss=0.57  (1958.1  examples/sec;  0.065  sec/batch)
step 60030, loss=0.67  (1909.7  examples/sec;  0.067  sec/batch)
step 60040, loss=0.65  (1994.5  examples/sec;  0.064  sec/batch)
step 60050, loss=0.61  (1885.2  examples/sec;  0.068  sec/batch)
step 60060, loss=0.73  (1822.9  examples/sec;  0.070  sec/batch)
step 60070, loss=0.58  (1905.4  examples/sec;  0.067  sec/batch)
step 60080, loss=0.61  (1876.3  examples/sec;  0.068  sec/batch)
step 60090, loss=0.77  (1881.0  examples/sec;  0.068  sec/batch)
step 60100, loss=0.71  (1955.2  examples/sec;  0.065  sec/batch)
step 60110, loss=0.58  (1882.4  examples/sec;  0.068  sec/batch)
step 60120, loss=0.58  (1941.1  examples/sec;  0.066  sec/batch)
step 60130, loss=0.65  (1895.3  examples/sec;  0.068  sec/batch)
step 60140, loss=0.59  (2025.3  examples/sec;  0.063  sec/batch)
step 60150, loss=0.58  (1864.8  examples/sec;  0.069  sec/batch)
step 60160, loss=0.75  (1952.0  examples/sec;  0.066  sec/batch)
step 60170, loss=0.80  (1628.5  examples/sec;  0.079  sec/batch)
step 60180, loss=0.57  (1925.2  examples/sec;  0.066  sec/batch)
step 60190, loss=0.64  (1895.4  examples/sec;  0.068  sec/batch)
step 60200, loss=0.61  (2001.3  examples/sec;  0.064  sec/batch)
step 60210, loss=0.55  (1981.1  examples/sec;  0.065  sec/batch)
step 60220, loss=0.60  (1874.1  examples/sec;  0.068  sec/batch)
step 60230, loss=0.71  (1863.3  examples/sec;  0.069  sec/batch)
step 60240, loss=0.66  (1861.9  examples/sec;  0.069  sec/batch)
step 60250, loss=0.45  (1950.2  examples/sec;  0.066  sec/batch)
step 60260, loss=0.69  (1873.3  examples/sec;  0.068  sec/batch)
step 60270, loss=0.54  (1907.1  examples/sec;  0.067  sec/batch)
step 60280, loss=0.57  (1830.9  examples/sec;  0.070  sec/batch)
step 60290, loss=0.71  (1894.9  examples/sec;  0.068  sec/batch)
step 60300, loss=0.71  (1921.3  examples/sec;  0.067  sec/batch)
step 60310, loss=0.57  (1933.4  examples/sec;  0.066  sec/batch)
step 60320, loss=0.65  (1970.9  examples/sec;  0.065  sec/batch)
step 60330, loss=0.53  (1859.6  examples/sec;  0.069  sec/batch)
step 60340, loss=0.60  (1787.5  examples/sec;  0.072  sec/batch)
step 60350, loss=0.68  (1907.9  examples/sec;  0.067  sec/batch)
step 60360, loss=0.61  (1908.7  examples/sec;  0.067  sec/batch)
step 60370, loss=0.66  (1936.7  examples/sec;  0.066  sec/batch)
step 60380, loss=0.65  (1944.0  examples/sec;  0.066  sec/batch)
step 60390, loss=0.74  (2001.9  examples/sec;  0.064  sec/batch)
step 60400, loss=0.74  (1934.9  examples/sec;  0.066  sec/batch)
step 60410, loss=0.59  (1891.3  examples/sec;  0.068  sec/batch)
step 60420, loss=0.75  (1893.0  examples/sec;  0.068  sec/batch)
step 60430, loss=0.62  (1930.0  examples/sec;  0.066  sec/batch)
step 60440, loss=0.73  (1905.0  examples/sec;  0.067  sec/batch)
step 60450, loss=0.56  (1872.8  examples/sec;  0.068  sec/batch)
step 60460, loss=0.59  (1953.5  examples/sec;  0.066  sec/batch)
step 60470, loss=0.66  (1932.8  examples/sec;  0.066  sec/batch)
step 60480, loss=0.56  (1898.6  examples/sec;  0.067  sec/batch)
step 60490, loss=0.67  (1941.3  examples/sec;  0.066  sec/batch)
step 60500, loss=0.71  (1943.8  examples/sec;  0.066  sec/batch)
step 60510, loss=0.68  (1934.6  examples/sec;  0.066  sec/batch)
step 60520, loss=0.71  (1920.5  examples/sec;  0.067  sec/batch)
step 60530, loss=0.68  (1928.4  examples/sec;  0.066  sec/batch)
step 60540, loss=0.60  (1883.4  examples/sec;  0.068  sec/batch)
step 60550, loss=0.74  (1951.3  examples/sec;  0.066  sec/batch)
step 60560, loss=0.69  (1972.7  examples/sec;  0.065  sec/batch)
step 60570, loss=0.57  (2004.1  examples/sec;  0.064  sec/batch)
step 60580, loss=0.53  (1949.9  examples/sec;  0.066  sec/batch)
step 60590, loss=0.75  (1854.6  examples/sec;  0.069  sec/batch)
step 60600, loss=0.59  (1887.5  examples/sec;  0.068  sec/batch)
step 60610, loss=0.69  (1878.3  examples/sec;  0.068  sec/batch)
step 60620, loss=0.62  (1931.4  examples/sec;  0.066  sec/batch)
step 60630, loss=1.01  (1963.8  examples/sec;  0.065  sec/batch)
step 60640, loss=0.50  (1875.8  examples/sec;  0.068  sec/batch)
step 60650, loss=0.51  (1929.0  examples/sec;  0.066  sec/batch)
step 60660, loss=0.55  (1913.1  examples/sec;  0.067  sec/batch)
step 60670, loss=0.62  (1881.1  examples/sec;  0.068  sec/batch)
step 60680, loss=0.77  (1935.6  examples/sec;  0.066  sec/batch)
step 60690, loss=0.71  (1844.0  examples/sec;  0.069  sec/batch)
step 60700, loss=0.44  (1725.8  examples/sec;  0.074  sec/batch)
step 60710, loss=0.68  (1952.3  examples/sec;  0.066  sec/batch)
step 60720, loss=0.46  (1917.2  examples/sec;  0.067  sec/batch)
step 60730, loss=0.75  (1925.4  examples/sec;  0.066  sec/batch)
step 60740, loss=0.46  (1871.2  examples/sec;  0.068  sec/batch)
step 60750, loss=0.61  (1894.8  examples/sec;  0.068  sec/batch)
step 60760, loss=0.54  (1843.0  examples/sec;  0.069  sec/batch)
step 60770, loss=0.59  (2006.2  examples/sec;  0.064  sec/batch)
step 60780, loss=0.56  (1839.1  examples/sec;  0.070  sec/batch)
step 60790, loss=0.52  (1873.3  examples/sec;  0.068  sec/batch)
step 60800, loss=0.63  (2019.6  examples/sec;  0.063  sec/batch)
step 60810, loss=0.65  (1875.2  examples/sec;  0.068  sec/batch)
step 60820, loss=0.59  (1885.2  examples/sec;  0.068  sec/batch)
step 60830, loss=0.58  (1826.2  examples/sec;  0.070  sec/batch)
step 60840, loss=0.75  (1949.9  examples/sec;  0.066  sec/batch)
step 60850, loss=0.61  (1951.4  examples/sec;  0.066  sec/batch)
step 60860, loss=0.62  (1923.8  examples/sec;  0.067  sec/batch)
step 60870, loss=0.73  (1824.1  examples/sec;  0.070  sec/batch)
step 60880, loss=0.64  (1581.0  examples/sec;  0.081  sec/batch)
step 60890, loss=0.60  (1902.8  examples/sec;  0.067  sec/batch)
step 60900, loss=0.57  (1827.5  examples/sec;  0.070  sec/batch)
step 60910, loss=0.53  (1898.7  examples/sec;  0.067  sec/batch)
step 60920, loss=0.65  (1947.1  examples/sec;  0.066  sec/batch)
step 60930, loss=0.50  (1887.1  examples/sec;  0.068  sec/batch)
step 60940, loss=0.63  (1940.0  examples/sec;  0.066  sec/batch)
step 60950, loss=0.70  (1860.1  examples/sec;  0.069  sec/batch)
step 60960, loss=0.52  (1968.0  examples/sec;  0.065  sec/batch)
step 60970, loss=0.56  (1866.2  examples/sec;  0.069  sec/batch)
step 60980, loss=0.57  (1942.2  examples/sec;  0.066  sec/batch)
step 60990, loss=0.70  (1776.5  examples/sec;  0.072  sec/batch)
step 61000, loss=0.76  (1906.0  examples/sec;  0.067  sec/batch)
step 61010, loss=0.64  (1885.6  examples/sec;  0.068  sec/batch)
step 61020, loss=0.63  (1937.8  examples/sec;  0.066  sec/batch)
step 61030, loss=0.51  (1879.2  examples/sec;  0.068  sec/batch)
step 61040, loss=0.60  (1977.0  examples/sec;  0.065  sec/batch)
step 61050, loss=0.80  (1880.6  examples/sec;  0.068  sec/batch)
step 61060, loss=0.74  (1955.2  examples/sec;  0.065  sec/batch)
step 61070, loss=0.62  (1927.5  examples/sec;  0.066  sec/batch)
step 61080, loss=0.62  (1944.9  examples/sec;  0.066  sec/batch)
step 61090, loss=0.68  (1926.3  examples/sec;  0.066  sec/batch)
step 61100, loss=0.77  (1870.7  examples/sec;  0.068  sec/batch)
step 61110, loss=0.55  (1931.6  examples/sec;  0.066  sec/batch)
step 61120, loss=0.50  (1766.9  examples/sec;  0.072  sec/batch)
step 61130, loss=0.70  (1952.9  examples/sec;  0.066  sec/batch)
step 61140, loss=0.63  (1943.6  examples/sec;  0.066  sec/batch)
step 61150, loss=0.61  (1672.9  examples/sec;  0.077  sec/batch)
step 61160, loss=0.47  (1894.3  examples/sec;  0.068  sec/batch)
step 61170, loss=0.72  (1866.5  examples/sec;  0.069  sec/batch)
step 61180, loss=0.66  (1909.1  examples/sec;  0.067  sec/batch)
step 61190, loss=0.64  (1877.9  examples/sec;  0.068  sec/batch)
step 61200, loss=0.56  (1786.1  examples/sec;  0.072  sec/batch)
step 61210, loss=0.79  (1883.2  examples/sec;  0.068  sec/batch)
step 61220, loss=0.47  (2000.8  examples/sec;  0.064  sec/batch)
step 61230, loss=0.73  (1850.2  examples/sec;  0.069  sec/batch)
step 61240, loss=0.59  (1841.6  examples/sec;  0.070  sec/batch)
step 61250, loss=0.60  (1887.0  examples/sec;  0.068  sec/batch)
step 61260, loss=0.59  (1885.4  examples/sec;  0.068  sec/batch)
step 61270, loss=0.46  (1946.1  examples/sec;  0.066  sec/batch)
step 61280, loss=0.48  (1841.8  examples/sec;  0.069  sec/batch)
step 61290, loss=0.62  (1853.9  examples/sec;  0.069  sec/batch)
step 61300, loss=0.70  (1870.7  examples/sec;  0.068  sec/batch)
step 61310, loss=0.55  (1936.7  examples/sec;  0.066  sec/batch)
step 61320, loss=0.53  (1935.2  examples/sec;  0.066  sec/batch)
step 61330, loss=0.75  (1933.5  examples/sec;  0.066  sec/batch)
step 61340, loss=0.55  (1943.7  examples/sec;  0.066  sec/batch)
step 61350, loss=0.65  (1915.9  examples/sec;  0.067  sec/batch)
step 61360, loss=0.69  (1853.6  examples/sec;  0.069  sec/batch)
step 61370, loss=0.69  (1878.3  examples/sec;  0.068  sec/batch)
step 61380, loss=0.64  (1896.3  examples/sec;  0.067  sec/batch)
step 61390, loss=0.54  (1939.7  examples/sec;  0.066  sec/batch)
step 61400, loss=0.60  (1895.7  examples/sec;  0.068  sec/batch)
step 61410, loss=0.59  (1996.9  examples/sec;  0.064  sec/batch)
step 61420, loss=0.74  (1931.3  examples/sec;  0.066  sec/batch)
step 61430, loss=0.60  (1860.8  examples/sec;  0.069  sec/batch)
step 61440, loss=0.55  (1912.4  examples/sec;  0.067  sec/batch)
step 61450, loss=0.61  (1829.7  examples/sec;  0.070  sec/batch)
step 61460, loss=0.50  (1908.2  examples/sec;  0.067  sec/batch)
step 61470, loss=0.63  (1914.0  examples/sec;  0.067  sec/batch)
step 61480, loss=0.60  (1894.0  examples/sec;  0.068  sec/batch)
step 61490, loss=0.60  (1847.1  examples/sec;  0.069  sec/batch)
step 61500, loss=0.47  (1914.6  examples/sec;  0.067  sec/batch)
step 61510, loss=0.66  (1898.6  examples/sec;  0.067  sec/batch)
step 61520, loss=0.58  (1935.3  examples/sec;  0.066  sec/batch)
step 61530, loss=0.63  (1902.5  examples/sec;  0.067  sec/batch)
step 61540, loss=0.63  (1590.4  examples/sec;  0.080  sec/batch)
step 61550, loss=0.67  (1895.7  examples/sec;  0.068  sec/batch)
step 61560, loss=0.49  (1641.7  examples/sec;  0.078  sec/batch)
step 61570, loss=0.60  (1885.4  examples/sec;  0.068  sec/batch)
step 61580, loss=0.75  (1848.3  examples/sec;  0.069  sec/batch)
step 61590, loss=0.64  (1853.9  examples/sec;  0.069  sec/batch)
step 61600, loss=0.75  (1787.1  examples/sec;  0.072  sec/batch)
step 61610, loss=0.58  (1896.5  examples/sec;  0.067  sec/batch)
step 61620, loss=0.62  (1970.0  examples/sec;  0.065  sec/batch)
step 61630, loss=0.71  (1950.3  examples/sec;  0.066  sec/batch)
step 61640, loss=0.77  (1951.3  examples/sec;  0.066  sec/batch)
step 61650, loss=0.58  (1945.7  examples/sec;  0.066  sec/batch)
step 61660, loss=0.48  (1741.1  examples/sec;  0.074  sec/batch)
step 61670, loss=0.78  (1949.1  examples/sec;  0.066  sec/batch)
step 61680, loss=0.60  (1930.6  examples/sec;  0.066  sec/batch)
step 61690, loss=0.68  (1966.0  examples/sec;  0.065  sec/batch)
step 61700, loss=0.65  (1923.3  examples/sec;  0.067  sec/batch)
step 61710, loss=0.44  (1905.5  examples/sec;  0.067  sec/batch)
step 61720, loss=0.78  (1914.2  examples/sec;  0.067  sec/batch)
step 61730, loss=0.60  (1980.2  examples/sec;  0.065  sec/batch)
step 61740, loss=0.60  (1945.1  examples/sec;  0.066  sec/batch)
step 61750, loss=0.76  (1911.0  examples/sec;  0.067  sec/batch)
step 61760, loss=0.65  (2035.7  examples/sec;  0.063  sec/batch)
step 61770, loss=0.64  (1973.7  examples/sec;  0.065  sec/batch)
step 61780, loss=0.58  (1854.0  examples/sec;  0.069  sec/batch)
step 61790, loss=0.60  (2005.2  examples/sec;  0.064  sec/batch)
step 61800, loss=0.53  (1882.3  examples/sec;  0.068  sec/batch)
step 61810, loss=0.65  (1909.3  examples/sec;  0.067  sec/batch)
step 61820, loss=0.54  (1879.8  examples/sec;  0.068  sec/batch)
step 61830, loss=0.59  (1864.8  examples/sec;  0.069  sec/batch)
step 61840, loss=0.55  (1974.7  examples/sec;  0.065  sec/batch)
step 61850, loss=0.66  (1850.7  examples/sec;  0.069  sec/batch)
step 61860, loss=0.52  (1944.4  examples/sec;  0.066  sec/batch)
step 61870, loss=0.77  (1926.6  examples/sec;  0.066  sec/batch)
step 61880, loss=0.67  (1919.3  examples/sec;  0.067  sec/batch)
step 61890, loss=0.56  (1856.6  examples/sec;  0.069  sec/batch)
step 61900, loss=0.56  (1954.7  examples/sec;  0.065  sec/batch)
step 61910, loss=0.83  (1912.4  examples/sec;  0.067  sec/batch)
step 61920, loss=0.55  (1879.2  examples/sec;  0.068  sec/batch)
step 61930, loss=0.62  (1887.3  examples/sec;  0.068  sec/batch)
step 61940, loss=0.61  (1847.7  examples/sec;  0.069  sec/batch)
step 61950, loss=0.64  (1985.2  examples/sec;  0.064  sec/batch)
step 61960, loss=0.59  (1920.9  examples/sec;  0.067  sec/batch)
step 61970, loss=0.66  (1954.1  examples/sec;  0.066  sec/batch)
step 61980, loss=0.52  (1920.6  examples/sec;  0.067  sec/batch)
step 61990, loss=0.60  (1896.5  examples/sec;  0.067  sec/batch)
step 62000, loss=0.66  (1885.2  examples/sec;  0.068  sec/batch)
step 62010, loss=0.53  (1857.9  examples/sec;  0.069  sec/batch)
step 62020, loss=0.51  (1900.1  examples/sec;  0.067  sec/batch)
step 62030, loss=0.61  (1923.3  examples/sec;  0.067  sec/batch)
step 62040, loss=0.49  (1781.5  examples/sec;  0.072  sec/batch)
step 62050, loss=0.72  (1838.5  examples/sec;  0.070  sec/batch)
step 62060, loss=0.51  (1931.0  examples/sec;  0.066  sec/batch)
step 62070, loss=0.60  (1901.7  examples/sec;  0.067  sec/batch)
step 62080, loss=0.65  (1941.6  examples/sec;  0.066  sec/batch)
step 62090, loss=0.66  (1940.3  examples/sec;  0.066  sec/batch)
step 62100, loss=0.70  (1854.6  examples/sec;  0.069  sec/batch)
step 62110, loss=0.56  (1899.5  examples/sec;  0.067  sec/batch)
step 62120, loss=0.58  (1848.2  examples/sec;  0.069  sec/batch)
step 62130, loss=0.50  (1915.4  examples/sec;  0.067  sec/batch)
step 62140, loss=0.52  (1977.9  examples/sec;  0.065  sec/batch)
step 62150, loss=0.54  (1930.4  examples/sec;  0.066  sec/batch)
step 62160, loss=0.55  (1796.9  examples/sec;  0.071  sec/batch)
step 62170, loss=0.63  (1909.0  examples/sec;  0.067  sec/batch)
step 62180, loss=0.49  (1943.0  examples/sec;  0.066  sec/batch)
step 62190, loss=0.56  (1870.2  examples/sec;  0.068  sec/batch)
step 62200, loss=0.61  (2043.2  examples/sec;  0.063  sec/batch)
step 62210, loss=0.50  (1898.7  examples/sec;  0.067  sec/batch)
step 62220, loss=0.45  (1916.0  examples/sec;  0.067  sec/batch)
step 62230, loss=0.57  (1884.2  examples/sec;  0.068  sec/batch)
step 62240, loss=0.65  (1932.3  examples/sec;  0.066  sec/batch)
step 62250, loss=0.64  (1887.6  examples/sec;  0.068  sec/batch)
step 62260, loss=0.77  (1937.8  examples/sec;  0.066  sec/batch)
step 62270, loss=0.61  (1936.0  examples/sec;  0.066  sec/batch)
step 62280, loss=0.52  (1846.6  examples/sec;  0.069  sec/batch)
step 62290, loss=0.52  (1629.5  examples/sec;  0.079  sec/batch)
step 62300, loss=0.61  (1885.5  examples/sec;  0.068  sec/batch)
step 62310, loss=0.50  (1943.6  examples/sec;  0.066  sec/batch)
step 62320, loss=0.55  (1950.0  examples/sec;  0.066  sec/batch)
step 62330, loss=0.71  (1799.8  examples/sec;  0.071  sec/batch)
step 62340, loss=0.70  (1937.3  examples/sec;  0.066  sec/batch)
step 62350, loss=0.66  (1892.2  examples/sec;  0.068  sec/batch)
step 62360, loss=0.62  (1883.9  examples/sec;  0.068  sec/batch)
step 62370, loss=0.74  (1877.2  examples/sec;  0.068  sec/batch)
step 62380, loss=0.57  (1884.9  examples/sec;  0.068  sec/batch)
step 62390, loss=0.54  (1897.1  examples/sec;  0.067  sec/batch)
step 62400, loss=0.58  (1944.1  examples/sec;  0.066  sec/batch)
step 62410, loss=0.51  (1958.4  examples/sec;  0.065  sec/batch)
step 62420, loss=0.66  (1950.7  examples/sec;  0.066  sec/batch)
step 62430, loss=0.52  (1864.2  examples/sec;  0.069  sec/batch)
step 62440, loss=0.47  (1939.7  examples/sec;  0.066  sec/batch)
step 62450, loss=0.66  (1951.1  examples/sec;  0.066  sec/batch)
step 62460, loss=0.60  (1894.3  examples/sec;  0.068  sec/batch)
step 62470, loss=0.59  (1835.9  examples/sec;  0.070  sec/batch)
step 62480, loss=0.87  (1868.1  examples/sec;  0.069  sec/batch)
step 62490, loss=0.60  (1875.3  examples/sec;  0.068  sec/batch)
step 62500, loss=0.54  (1863.7  examples/sec;  0.069  sec/batch)
step 62510, loss=0.74  (1880.0  examples/sec;  0.068  sec/batch)
step 62520, loss=0.62  (1694.0  examples/sec;  0.076  sec/batch)
step 62530, loss=0.74  (1921.3  examples/sec;  0.067  sec/batch)
step 62540, loss=0.48  (1910.6  examples/sec;  0.067  sec/batch)
step 62550, loss=0.60  (1956.9  examples/sec;  0.065  sec/batch)
step 62560, loss=0.59  (1929.0  examples/sec;  0.066  sec/batch)
step 62570, loss=0.68  (1905.5  examples/sec;  0.067  sec/batch)
step 62580, loss=0.81  (1900.3  examples/sec;  0.067  sec/batch)
step 62590, loss=0.74  (1932.7  examples/sec;  0.066  sec/batch)
step 62600, loss=0.66  (1845.9  examples/sec;  0.069  sec/batch)
step 62610, loss=0.63  (1887.3  examples/sec;  0.068  sec/batch)
step 62620, loss=0.65  (1938.1  examples/sec;  0.066  sec/batch)
step 62630, loss=0.58  (2015.5  examples/sec;  0.064  sec/batch)
step 62640, loss=0.65  (1887.5  examples/sec;  0.068  sec/batch)
step 62650, loss=0.55  (1955.7  examples/sec;  0.065  sec/batch)
step 62660, loss=0.63  (1841.3  examples/sec;  0.070  sec/batch)
step 62670, loss=0.67  (1917.7  examples/sec;  0.067  sec/batch)
step 62680, loss=0.51  (1878.9  examples/sec;  0.068  sec/batch)
step 62690, loss=0.53  (1974.8  examples/sec;  0.065  sec/batch)
step 62700, loss=0.50  (1991.9  examples/sec;  0.064  sec/batch)
step 62710, loss=0.62  (1917.1  examples/sec;  0.067  sec/batch)
step 62720, loss=0.69  (1888.0  examples/sec;  0.068  sec/batch)
step 62730, loss=0.71  (1865.8  examples/sec;  0.069  sec/batch)
step 62740, loss=0.67  (1846.1  examples/sec;  0.069  sec/batch)
step 62750, loss=0.72  (1864.3  examples/sec;  0.069  sec/batch)
step 62760, loss=0.56  (1920.6  examples/sec;  0.067  sec/batch)
step 62770, loss=0.60  (1983.3  examples/sec;  0.065  sec/batch)
step 62780, loss=0.64  (1866.2  examples/sec;  0.069  sec/batch)
step 62790, loss=0.65  (1805.1  examples/sec;  0.071  sec/batch)
step 62800, loss=0.68  (1864.0  examples/sec;  0.069  sec/batch)
step 62810, loss=0.60  (1967.7  examples/sec;  0.065  sec/batch)
step 62820, loss=0.58  (1905.5  examples/sec;  0.067  sec/batch)
step 62830, loss=0.79  (1913.3  examples/sec;  0.067  sec/batch)
step 62840, loss=0.71  (1905.5  examples/sec;  0.067  sec/batch)
step 62850, loss=0.75  (1810.0  examples/sec;  0.071  sec/batch)
step 62860, loss=0.63  (1727.5  examples/sec;  0.074  sec/batch)
step 62870, loss=0.61  (1905.7  examples/sec;  0.067  sec/batch)
step 62880, loss=0.59  (1968.7  examples/sec;  0.065  sec/batch)
step 62890, loss=0.54  (1926.6  examples/sec;  0.066  sec/batch)
step 62900, loss=0.65  (1948.9  examples/sec;  0.066  sec/batch)
step 62910, loss=0.56  (1865.8  examples/sec;  0.069  sec/batch)
step 62920, loss=0.58  (1865.5  examples/sec;  0.069  sec/batch)
step 62930, loss=0.48  (1965.7  examples/sec;  0.065  sec/batch)
step 62940, loss=0.51  (1902.9  examples/sec;  0.067  sec/batch)
step 62950, loss=0.67  (1934.4  examples/sec;  0.066  sec/batch)
step 62960, loss=0.57  (1942.9  examples/sec;  0.066  sec/batch)
step 62970, loss=0.67  (1944.1  examples/sec;  0.066  sec/batch)
step 62980, loss=0.66  (1886.0  examples/sec;  0.068  sec/batch)
step 62990, loss=0.73  (1880.3  examples/sec;  0.068  sec/batch)
step 63000, loss=0.67  (1937.5  examples/sec;  0.066  sec/batch)
step 63010, loss=0.62  (1883.2  examples/sec;  0.068  sec/batch)
step 63020, loss=0.61  (1943.8  examples/sec;  0.066  sec/batch)
step 63030, loss=0.68  (1869.0  examples/sec;  0.068  sec/batch)
step 63040, loss=0.60  (1910.9  examples/sec;  0.067  sec/batch)
step 63050, loss=0.53  (1884.0  examples/sec;  0.068  sec/batch)
step 63060, loss=0.54  (1907.7  examples/sec;  0.067  sec/batch)
step 63070, loss=0.63  (1889.7  examples/sec;  0.068  sec/batch)
step 63080, loss=0.55  (1850.0  examples/sec;  0.069  sec/batch)
step 63090, loss=0.66  (1833.4  examples/sec;  0.070  sec/batch)
step 63100, loss=0.54  (1916.3  examples/sec;  0.067  sec/batch)
step 63110, loss=0.66  (1873.3  examples/sec;  0.068  sec/batch)
step 63120, loss=0.51  (1900.8  examples/sec;  0.067  sec/batch)
step 63130, loss=0.41  (1931.1  examples/sec;  0.066  sec/batch)
step 63140, loss=0.66  (1966.6  examples/sec;  0.065  sec/batch)
step 63150, loss=0.60  (1853.2  examples/sec;  0.069  sec/batch)
step 63160, loss=0.50  (1852.9  examples/sec;  0.069  sec/batch)
step 63170, loss=0.71  (1936.6  examples/sec;  0.066  sec/batch)
step 63180, loss=0.82  (1893.7  examples/sec;  0.068  sec/batch)
step 63190, loss=0.73  (1891.6  examples/sec;  0.068  sec/batch)
step 63200, loss=0.69  (1932.1  examples/sec;  0.066  sec/batch)
step 63210, loss=0.51  (1888.4  examples/sec;  0.068  sec/batch)
step 63220, loss=0.59  (1887.7  examples/sec;  0.068  sec/batch)
step 63230, loss=0.77  (1773.5  examples/sec;  0.072  sec/batch)
step 63240, loss=0.50  (1869.2  examples/sec;  0.068  sec/batch)
step 63250, loss=0.68  (1903.8  examples/sec;  0.067  sec/batch)
step 63260, loss=0.61  (1872.9  examples/sec;  0.068  sec/batch)
step 63270, loss=0.69  (1956.9  examples/sec;  0.065  sec/batch)
step 63280, loss=0.50  (1893.2  examples/sec;  0.068  sec/batch)
step 63290, loss=0.58  (1950.9  examples/sec;  0.066  sec/batch)
step 63300, loss=0.68  (1852.9  examples/sec;  0.069  sec/batch)
step 63310, loss=0.67  (1860.5  examples/sec;  0.069  sec/batch)
step 63320, loss=0.69  (1863.3  examples/sec;  0.069  sec/batch)
step 63330, loss=0.47  (1928.8  examples/sec;  0.066  sec/batch)
step 63340, loss=0.63  (1789.9  examples/sec;  0.072  sec/batch)
step 63350, loss=0.68  (1857.5  examples/sec;  0.069  sec/batch)
step 63360, loss=0.72  (1964.8  examples/sec;  0.065  sec/batch)
step 63370, loss=0.57  (1829.8  examples/sec;  0.070  sec/batch)
step 63380, loss=0.52  (1923.8  examples/sec;  0.067  sec/batch)
step 63390, loss=0.53  (1882.2  examples/sec;  0.068  sec/batch)
step 63400, loss=0.68  (1904.5  examples/sec;  0.067  sec/batch)
step 63410, loss=0.65  (1923.3  examples/sec;  0.067  sec/batch)
step 63420, loss=0.63  (1979.9  examples/sec;  0.065  sec/batch)
step 63430, loss=0.71  (1886.9  examples/sec;  0.068  sec/batch)
step 63440, loss=0.88  (1959.1  examples/sec;  0.065  sec/batch)
step 63450, loss=0.81  (1978.1  examples/sec;  0.065  sec/batch)
step 63460, loss=0.65  (1839.4  examples/sec;  0.070  sec/batch)
step 63470, loss=0.55  (1923.0  examples/sec;  0.067  sec/batch)
step 63480, loss=0.53  (1695.9  examples/sec;  0.075  sec/batch)
step 63490, loss=0.67  (1887.8  examples/sec;  0.068  sec/batch)
step 63500, loss=0.67  (1497.0  examples/sec;  0.086  sec/batch)
step 63510, loss=0.52  (1913.7  examples/sec;  0.067  sec/batch)
step 63520, loss=0.64  (1970.3  examples/sec;  0.065  sec/batch)
step 63530, loss=0.57  (1974.9  examples/sec;  0.065  sec/batch)
step 63540, loss=0.56  (1732.8  examples/sec;  0.074  sec/batch)
step 63550, loss=0.55  (1839.0  examples/sec;  0.070  sec/batch)
step 63560, loss=0.54  (1869.5  examples/sec;  0.068  sec/batch)
step 63570, loss=0.74  (1841.5  examples/sec;  0.070  sec/batch)
step 63580, loss=0.61  (1886.2  examples/sec;  0.068  sec/batch)
step 63590, loss=0.58  (1893.0  examples/sec;  0.068  sec/batch)
step 63600, loss=0.58  (1929.0  examples/sec;  0.066  sec/batch)
step 63610, loss=0.65  (1890.1  examples/sec;  0.068  sec/batch)
step 63620, loss=0.68  (1879.6  examples/sec;  0.068  sec/batch)
step 63630, loss=0.52  (1926.3  examples/sec;  0.066  sec/batch)
step 63640, loss=0.62  (1874.6  examples/sec;  0.068  sec/batch)
step 63650, loss=0.64  (1929.7  examples/sec;  0.066  sec/batch)
step 63660, loss=0.53  (1869.5  examples/sec;  0.068  sec/batch)
step 63670, loss=0.62  (1918.3  examples/sec;  0.067  sec/batch)
step 63680, loss=0.59  (1934.0  examples/sec;  0.066  sec/batch)
step 63690, loss=0.68  (2046.5  examples/sec;  0.063  sec/batch)
step 63700, loss=0.52  (1960.9  examples/sec;  0.065  sec/batch)
step 63710, loss=0.66  (1911.8  examples/sec;  0.067  sec/batch)
step 63720, loss=0.61  (1901.1  examples/sec;  0.067  sec/batch)
step 63730, loss=0.47  (1866.0  examples/sec;  0.069  sec/batch)
step 63740, loss=0.54  (1872.6  examples/sec;  0.068  sec/batch)
step 63750, loss=0.55  (1937.5  examples/sec;  0.066  sec/batch)
step 63760, loss=0.68  (1974.4  examples/sec;  0.065  sec/batch)
step 63770, loss=0.65  (1894.9  examples/sec;  0.068  sec/batch)
step 63780, loss=0.66  (1843.6  examples/sec;  0.069  sec/batch)
step 63790, loss=0.59  (1611.4  examples/sec;  0.079  sec/batch)
step 63800, loss=0.69  (1890.5  examples/sec;  0.068  sec/batch)
step 63810, loss=0.54  (1918.1  examples/sec;  0.067  sec/batch)
step 63820, loss=0.70  (1896.1  examples/sec;  0.068  sec/batch)
step 63830, loss=0.64  (1869.8  examples/sec;  0.068  sec/batch)
step 63840, loss=0.67  (1861.9  examples/sec;  0.069  sec/batch)
step 63850, loss=0.58  (1967.5  examples/sec;  0.065  sec/batch)
step 63860, loss=0.61  (1867.4  examples/sec;  0.069  sec/batch)
step 63870, loss=0.78  (1962.1  examples/sec;  0.065  sec/batch)
step 63880, loss=0.56  (1801.3  examples/sec;  0.071  sec/batch)
step 63890, loss=0.56  (1924.0  examples/sec;  0.067  sec/batch)
step 63900, loss=0.62  (1843.7  examples/sec;  0.069  sec/batch)
step 63910, loss=0.56  (1944.3  examples/sec;  0.066  sec/batch)
step 63920, loss=0.51  (1961.5  examples/sec;  0.065  sec/batch)
step 63930, loss=0.84  (1903.9  examples/sec;  0.067  sec/batch)
step 63940, loss=0.62  (1867.1  examples/sec;  0.069  sec/batch)
step 63950, loss=0.52  (1861.5  examples/sec;  0.069  sec/batch)
step 63960, loss=0.56  (1962.9  examples/sec;  0.065  sec/batch)
step 63970, loss=0.57  (2016.0  examples/sec;  0.063  sec/batch)
step 63980, loss=0.70  (1961.1  examples/sec;  0.065  sec/batch)
step 63990, loss=0.74  (1926.3  examples/sec;  0.066  sec/batch)
step 64000, loss=0.63  (1908.3  examples/sec;  0.067  sec/batch)
step 64010, loss=0.75  (2014.0  examples/sec;  0.064  sec/batch)
step 64020, loss=0.71  (1878.5  examples/sec;  0.068  sec/batch)
step 64030, loss=0.63  (1862.8  examples/sec;  0.069  sec/batch)
step 64040, loss=0.53  (1926.2  examples/sec;  0.066  sec/batch)
step 64050, loss=0.79  (1941.9  examples/sec;  0.066  sec/batch)
step 64060, loss=0.76  (1839.4  examples/sec;  0.070  sec/batch)
step 64070, loss=0.74  (1824.7  examples/sec;  0.070  sec/batch)
step 64080, loss=0.61  (1664.1  examples/sec;  0.077  sec/batch)
step 64090, loss=0.65  (1889.2  examples/sec;  0.068  sec/batch)
step 64100, loss=0.65  (1929.6  examples/sec;  0.066  sec/batch)
step 64110, loss=0.72  (1927.7  examples/sec;  0.066  sec/batch)
step 64120, loss=0.64  (1815.6  examples/sec;  0.070  sec/batch)
step 64130, loss=0.64  (1752.0  examples/sec;  0.073  sec/batch)
step 64140, loss=0.64  (1943.7  examples/sec;  0.066  sec/batch)
step 64150, loss=0.65  (1918.3  examples/sec;  0.067  sec/batch)
step 64160, loss=0.57  (1925.4  examples/sec;  0.066  sec/batch)
step 64170, loss=0.59  (1893.5  examples/sec;  0.068  sec/batch)
step 64180, loss=0.68  (1863.1  examples/sec;  0.069  sec/batch)
step 64190, loss=0.64  (1875.6  examples/sec;  0.068  sec/batch)
step 64200, loss=0.53  (1929.5  examples/sec;  0.066  sec/batch)
step 64210, loss=0.65  (1972.6  examples/sec;  0.065  sec/batch)
step 64220, loss=0.65  (1816.5  examples/sec;  0.070  sec/batch)
step 64230, loss=0.58  (1892.9  examples/sec;  0.068  sec/batch)
step 64240, loss=0.68  (1940.8  examples/sec;  0.066  sec/batch)
step 64250, loss=0.60  (1819.9  examples/sec;  0.070  sec/batch)
step 64260, loss=0.67  (2019.1  examples/sec;  0.063  sec/batch)
step 64270, loss=0.62  (1640.2  examples/sec;  0.078  sec/batch)
step 64280, loss=0.68  (1943.9  examples/sec;  0.066  sec/batch)
step 64290, loss=0.60  (1796.1  examples/sec;  0.071  sec/batch)
step 64300, loss=0.52  (1889.3  examples/sec;  0.068  sec/batch)
step 64310, loss=0.66  (1891.2  examples/sec;  0.068  sec/batch)
step 64320, loss=0.62  (1863.1  examples/sec;  0.069  sec/batch)
step 64330, loss=0.72  (1929.6  examples/sec;  0.066  sec/batch)
step 64340, loss=0.50  (1928.0  examples/sec;  0.066  sec/batch)
step 64350, loss=0.71  (1856.1  examples/sec;  0.069  sec/batch)
step 64360, loss=0.77  (1893.7  examples/sec;  0.068  sec/batch)
step 64370, loss=0.68  (1869.9  examples/sec;  0.068  sec/batch)
step 64380, loss=0.54  (1838.9  examples/sec;  0.070  sec/batch)
step 64390, loss=0.68  (1879.5  examples/sec;  0.068  sec/batch)
step 64400, loss=0.60  (1824.1  examples/sec;  0.070  sec/batch)
step 64410, loss=0.65  (1917.3  examples/sec;  0.067  sec/batch)
step 64420, loss=0.66  (1936.3  examples/sec;  0.066  sec/batch)
step 64430, loss=0.50  (1887.2  examples/sec;  0.068  sec/batch)
step 64440, loss=0.69  (1846.4  examples/sec;  0.069  sec/batch)
step 64450, loss=0.58  (1938.5  examples/sec;  0.066  sec/batch)
step 64460, loss=0.69  (2024.2  examples/sec;  0.063  sec/batch)
step 64470, loss=0.83  (1830.9  examples/sec;  0.070  sec/batch)
step 64480, loss=0.56  (1933.7  examples/sec;  0.066  sec/batch)
step 64490, loss=0.57  (1864.1  examples/sec;  0.069  sec/batch)
step 64500, loss=0.58  (1887.4  examples/sec;  0.068  sec/batch)
step 64510, loss=0.54  (1932.7  examples/sec;  0.066  sec/batch)
step 64520, loss=0.63  (1957.0  examples/sec;  0.065  sec/batch)
step 64530, loss=0.67  (1700.0  examples/sec;  0.075  sec/batch)
step 64540, loss=0.64  (1883.3  examples/sec;  0.068  sec/batch)
step 64550, loss=0.58  (1971.0  examples/sec;  0.065  sec/batch)
step 64560, loss=0.59  (1930.0  examples/sec;  0.066  sec/batch)
step 64570, loss=0.54  (1978.1  examples/sec;  0.065  sec/batch)
step 64580, loss=0.61  (1926.3  examples/sec;  0.066  sec/batch)
step 64590, loss=0.61  (1919.9  examples/sec;  0.067  sec/batch)
step 64600, loss=0.68  (1916.0  examples/sec;  0.067  sec/batch)
step 64610, loss=0.75  (1906.2  examples/sec;  0.067  sec/batch)
step 64620, loss=0.60  (1869.0  examples/sec;  0.068  sec/batch)
step 64630, loss=0.64  (1926.3  examples/sec;  0.066  sec/batch)
step 64640, loss=0.68  (1954.1  examples/sec;  0.066  sec/batch)
step 64650, loss=0.71  (1875.1  examples/sec;  0.068  sec/batch)
step 64660, loss=0.73  (1941.2  examples/sec;  0.066  sec/batch)
step 64670, loss=0.68  (2018.0  examples/sec;  0.063  sec/batch)
step 64680, loss=0.76  (1904.3  examples/sec;  0.067  sec/batch)
step 64690, loss=0.68  (1967.2  examples/sec;  0.065  sec/batch)
step 64700, loss=0.71  (1885.4  examples/sec;  0.068  sec/batch)
step 64710, loss=0.52  (1831.5  examples/sec;  0.070  sec/batch)
step 64720, loss=0.48  (1933.5  examples/sec;  0.066  sec/batch)
step 64730, loss=0.62  (1933.9  examples/sec;  0.066  sec/batch)
step 64740, loss=0.49  (1838.9  examples/sec;  0.070  sec/batch)
step 64750, loss=0.59  (1932.8  examples/sec;  0.066  sec/batch)
step 64760, loss=0.83  (1910.5  examples/sec;  0.067  sec/batch)
step 64770, loss=0.69  (2032.1  examples/sec;  0.063  sec/batch)
step 64780, loss=0.75  (1744.5  examples/sec;  0.073  sec/batch)
step 64790, loss=0.74  (1878.9  examples/sec;  0.068  sec/batch)
step 64800, loss=0.66  (1940.5  examples/sec;  0.066  sec/batch)
step 64810, loss=0.63  (1960.7  examples/sec;  0.065  sec/batch)
step 64820, loss=0.59  (1932.4  examples/sec;  0.066  sec/batch)
step 64830, loss=0.58  (2065.2  examples/sec;  0.062  sec/batch)
step 64840, loss=0.65  (1898.2  examples/sec;  0.067  sec/batch)
step 64850, loss=0.53  (1902.9  examples/sec;  0.067  sec/batch)
step 64860, loss=0.49  (1845.4  examples/sec;  0.069  sec/batch)
step 64870, loss=0.48  (1898.6  examples/sec;  0.067  sec/batch)
step 64880, loss=0.64  (1926.3  examples/sec;  0.066  sec/batch)
step 64890, loss=0.61  (1858.8  examples/sec;  0.069  sec/batch)
step 64900, loss=0.53  (1881.2  examples/sec;  0.068  sec/batch)
step 64910, loss=0.49  (1912.1  examples/sec;  0.067  sec/batch)
step 64920, loss=0.54  (1939.1  examples/sec;  0.066  sec/batch)
step 64930, loss=0.65  (1926.5  examples/sec;  0.066  sec/batch)
step 64940, loss=0.72  (1971.6  examples/sec;  0.065  sec/batch)
step 64950, loss=0.66  (1812.5  examples/sec;  0.071  sec/batch)
step 64960, loss=0.61  (1848.2  examples/sec;  0.069  sec/batch)
step 64970, loss=0.64  (1930.0  examples/sec;  0.066  sec/batch)
step 64980, loss=0.57  (1804.3  examples/sec;  0.071  sec/batch)
step 64990, loss=0.66  (1874.4  examples/sec;  0.068  sec/batch)
step 65000, loss=0.49  (1716.3  examples/sec;  0.075  sec/batch)
step 65010, loss=0.75  (2026.4  examples/sec;  0.063  sec/batch)
step 65020, loss=0.66  (1954.8  examples/sec;  0.065  sec/batch)
step 65030, loss=0.70  (1855.1  examples/sec;  0.069  sec/batch)
step 65040, loss=0.79  (1962.3  examples/sec;  0.065  sec/batch)
step 65050, loss=0.62  (1963.9  examples/sec;  0.065  sec/batch)
step 65060, loss=0.54  (1972.8  examples/sec;  0.065  sec/batch)
step 65070, loss=0.62  (1904.5  examples/sec;  0.067  sec/batch)
step 65080, loss=0.57  (1995.6  examples/sec;  0.064  sec/batch)
step 65090, loss=0.63  (1886.3  examples/sec;  0.068  sec/batch)
step 65100, loss=0.59  (1814.5  examples/sec;  0.071  sec/batch)
step 65110, loss=0.54  (1821.1  examples/sec;  0.070  sec/batch)
step 65120, loss=0.54  (1954.7  examples/sec;  0.065  sec/batch)
step 65130, loss=0.75  (1958.5  examples/sec;  0.065  sec/batch)
step 65140, loss=0.73  (1897.1  examples/sec;  0.067  sec/batch)
step 65150, loss=0.68  (1888.2  examples/sec;  0.068  sec/batch)
step 65160, loss=0.67  (1961.7  examples/sec;  0.065  sec/batch)
step 65170, loss=0.60  (1978.5  examples/sec;  0.065  sec/batch)
step 65180, loss=0.58  (1974.7  examples/sec;  0.065  sec/batch)
step 65190, loss=0.72  (1850.0  examples/sec;  0.069  sec/batch)
step 65200, loss=0.59  (1896.3  examples/sec;  0.068  sec/batch)
step 65210, loss=0.58  (1831.2  examples/sec;  0.070  sec/batch)
step 65220, loss=0.60  (1800.2  examples/sec;  0.071  sec/batch)
step 65230, loss=0.39  (1860.9  examples/sec;  0.069  sec/batch)
step 65240, loss=0.69  (1960.3  examples/sec;  0.065  sec/batch)
step 65250, loss=0.60  (1890.9  examples/sec;  0.068  sec/batch)
step 65260, loss=0.54  (1792.1  examples/sec;  0.071  sec/batch)
step 65270, loss=0.61  (1646.9  examples/sec;  0.078  sec/batch)
step 65280, loss=0.55  (1883.0  examples/sec;  0.068  sec/batch)
step 65290, loss=0.69  (1824.0  examples/sec;  0.070  sec/batch)
step 65300, loss=0.57  (1930.3  examples/sec;  0.066  sec/batch)
step 65310, loss=0.68  (1959.9  examples/sec;  0.065  sec/batch)
step 65320, loss=0.67  (1885.0  examples/sec;  0.068  sec/batch)
step 65330, loss=0.61  (1891.2  examples/sec;  0.068  sec/batch)
step 65340, loss=0.69  (1839.3  examples/sec;  0.070  sec/batch)
step 65350, loss=0.60  (1931.1  examples/sec;  0.066  sec/batch)
step 65360, loss=0.59  (1983.7  examples/sec;  0.065  sec/batch)
step 65370, loss=0.76  (1825.9  examples/sec;  0.070  sec/batch)
step 65380, loss=0.75  (1918.4  examples/sec;  0.067  sec/batch)
step 65390, loss=0.56  (1902.5  examples/sec;  0.067  sec/batch)
step 65400, loss=0.59  (1659.8  examples/sec;  0.077  sec/batch)
step 65410, loss=0.73  (1934.7  examples/sec;  0.066  sec/batch)
step 65420, loss=0.75  (1921.2  examples/sec;  0.067  sec/batch)
step 65430, loss=0.69  (1883.3  examples/sec;  0.068  sec/batch)
step 65440, loss=0.63  (1891.8  examples/sec;  0.068  sec/batch)
step 65450, loss=0.77  (1861.7  examples/sec;  0.069  sec/batch)
step 65460, loss=0.65  (1842.7  examples/sec;  0.069  sec/batch)
step 65470, loss=0.64  (1904.6  examples/sec;  0.067  sec/batch)
step 65480, loss=0.78  (1892.6  examples/sec;  0.068  sec/batch)
step 65490, loss=0.68  (1871.9  examples/sec;  0.068  sec/batch)
step 65500, loss=0.60  (1947.8  examples/sec;  0.066  sec/batch)
step 65510, loss=0.56  (1930.4  examples/sec;  0.066  sec/batch)
step 65520, loss=0.65  (1944.6  examples/sec;  0.066  sec/batch)
step 65530, loss=0.64  (1841.8  examples/sec;  0.069  sec/batch)
step 65540, loss=0.72  (1884.1  examples/sec;  0.068  sec/batch)
step 65550, loss=0.49  (2025.4  examples/sec;  0.063  sec/batch)
step 65560, loss=0.56  (1895.7  examples/sec;  0.068  sec/batch)
step 65570, loss=0.65  (1904.2  examples/sec;  0.067  sec/batch)
step 65580, loss=0.72  (1866.9  examples/sec;  0.069  sec/batch)
step 65590, loss=0.44  (1758.7  examples/sec;  0.073  sec/batch)
step 65600, loss=0.52  (1833.1  examples/sec;  0.070  sec/batch)
step 65610, loss=0.58  (1876.1  examples/sec;  0.068  sec/batch)
step 65620, loss=0.69  (1986.7  examples/sec;  0.064  sec/batch)
step 65630, loss=0.54  (1972.1  examples/sec;  0.065  sec/batch)
step 65640, loss=0.91  (1937.2  examples/sec;  0.066  sec/batch)
step 65650, loss=0.46  (1941.9  examples/sec;  0.066  sec/batch)
step 65660, loss=0.54  (1856.4  examples/sec;  0.069  sec/batch)
step 65670, loss=0.72  (2046.6  examples/sec;  0.063  sec/batch)
step 65680, loss=0.57  (1861.2  examples/sec;  0.069  sec/batch)
step 65690, loss=0.47  (1859.2  examples/sec;  0.069  sec/batch)
step 65700, loss=0.54  (1824.5  examples/sec;  0.070  sec/batch)
step 65710, loss=0.51  (1866.4  examples/sec;  0.069  sec/batch)
step 65720, loss=0.65  (1949.8  examples/sec;  0.066  sec/batch)
step 65730, loss=0.58  (1862.0  examples/sec;  0.069  sec/batch)
step 65740, loss=0.70  (2004.7  examples/sec;  0.064  sec/batch)
step 65750, loss=0.70  (1906.3  examples/sec;  0.067  sec/batch)
step 65760, loss=0.60  (1640.0  examples/sec;  0.078  sec/batch)
step 65770, loss=0.60  (1870.4  examples/sec;  0.068  sec/batch)
step 65780, loss=0.71  (1891.4  examples/sec;  0.068  sec/batch)
step 65790, loss=0.56  (1973.8  examples/sec;  0.065  sec/batch)
step 65800, loss=0.55  (2037.8  examples/sec;  0.063  sec/batch)
step 65810, loss=0.57  (1902.4  examples/sec;  0.067  sec/batch)
step 65820, loss=0.51  (1934.9  examples/sec;  0.066  sec/batch)
step 65830, loss=0.68  (1951.3  examples/sec;  0.066  sec/batch)
step 65840, loss=0.66  (1907.9  examples/sec;  0.067  sec/batch)
step 65850, loss=0.64  (1834.1  examples/sec;  0.070  sec/batch)
step 65860, loss=0.68  (1925.8  examples/sec;  0.066  sec/batch)
step 65870, loss=0.64  (1868.5  examples/sec;  0.069  sec/batch)
step 65880, loss=0.53  (1905.4  examples/sec;  0.067  sec/batch)
step 65890, loss=0.67  (1871.2  examples/sec;  0.068  sec/batch)
step 65900, loss=0.74  (1980.4  examples/sec;  0.065  sec/batch)
step 65910, loss=0.71  (1789.3  examples/sec;  0.072  sec/batch)
step 65920, loss=0.80  (1959.3  examples/sec;  0.065  sec/batch)
step 65930, loss=0.65  (2011.8  examples/sec;  0.064  sec/batch)
step 65940, loss=0.58  (1911.9  examples/sec;  0.067  sec/batch)
step 65950, loss=0.58  (1932.4  examples/sec;  0.066  sec/batch)
step 65960, loss=0.47  (1938.1  examples/sec;  0.066  sec/batch)
step 65970, loss=0.60  (1954.1  examples/sec;  0.066  sec/batch)
step 65980, loss=0.54  (1848.6  examples/sec;  0.069  sec/batch)
step 65990, loss=0.62  (1889.8  examples/sec;  0.068  sec/batch)
step 66000, loss=0.69  (1908.8  examples/sec;  0.067  sec/batch)
step 66010, loss=0.66  (1878.3  examples/sec;  0.068  sec/batch)
step 66020, loss=0.67  (1980.9  examples/sec;  0.065  sec/batch)
step 66030, loss=0.52  (1927.2  examples/sec;  0.066  sec/batch)
step 66040, loss=0.62  (1915.9  examples/sec;  0.067  sec/batch)
step 66050, loss=0.65  (1934.5  examples/sec;  0.066  sec/batch)
step 66060, loss=0.65  (1940.2  examples/sec;  0.066  sec/batch)
step 66070, loss=0.72  (1838.5  examples/sec;  0.070  sec/batch)
step 66080, loss=0.62  (1890.9  examples/sec;  0.068  sec/batch)
step 66090, loss=0.63  (1822.0  examples/sec;  0.070  sec/batch)
step 66100, loss=0.55  (2076.4  examples/sec;  0.062  sec/batch)
step 66110, loss=0.90  (1913.9  examples/sec;  0.067  sec/batch)
step 66120, loss=0.65  (1808.9  examples/sec;  0.071  sec/batch)
step 66130, loss=0.64  (1976.7  examples/sec;  0.065  sec/batch)
step 66140, loss=0.56  (1896.9  examples/sec;  0.067  sec/batch)
step 66150, loss=0.57  (1786.6  examples/sec;  0.072  sec/batch)
step 66160, loss=0.57  (1923.2  examples/sec;  0.067  sec/batch)
step 66170, loss=0.63  (1916.2  examples/sec;  0.067  sec/batch)
step 66180, loss=0.59  (1858.3  examples/sec;  0.069  sec/batch)
step 66190, loss=0.52  (1834.2  examples/sec;  0.070  sec/batch)
step 66200, loss=0.70  (1946.9  examples/sec;  0.066  sec/batch)
step 66210, loss=0.65  (1927.1  examples/sec;  0.066  sec/batch)
step 66220, loss=0.65  (1896.7  examples/sec;  0.067  sec/batch)
step 66230, loss=0.68  (1889.8  examples/sec;  0.068  sec/batch)
step 66240, loss=0.66  (1952.1  examples/sec;  0.066  sec/batch)
step 66250, loss=0.70  (1883.1  examples/sec;  0.068  sec/batch)
step 66260, loss=0.63  (1659.9  examples/sec;  0.077  sec/batch)
step 66270, loss=0.52  (1915.1  examples/sec;  0.067  sec/batch)
step 66280, loss=0.66  (1937.3  examples/sec;  0.066  sec/batch)
step 66290, loss=0.67  (1884.8  examples/sec;  0.068  sec/batch)
step 66300, loss=0.64  (1929.4  examples/sec;  0.066  sec/batch)
step 66310, loss=0.69  (1884.8  examples/sec;  0.068  sec/batch)
step 66320, loss=0.66  (1910.5  examples/sec;  0.067  sec/batch)
step 66330, loss=0.66  (1940.1  examples/sec;  0.066  sec/batch)
step 66340, loss=0.60  (1871.7  examples/sec;  0.068  sec/batch)
step 66350, loss=0.58  (1902.9  examples/sec;  0.067  sec/batch)
step 66360, loss=0.66  (1562.5  examples/sec;  0.082  sec/batch)
step 66370, loss=0.82  (1995.8  examples/sec;  0.064  sec/batch)
step 66380, loss=0.56  (1879.1  examples/sec;  0.068  sec/batch)
step 66390, loss=0.62  (1931.7  examples/sec;  0.066  sec/batch)
step 66400, loss=0.42  (1848.6  examples/sec;  0.069  sec/batch)
step 66410, loss=0.52  (1791.0  examples/sec;  0.071  sec/batch)
step 66420, loss=0.72  (1900.6  examples/sec;  0.067  sec/batch)
step 66430, loss=0.47  (1927.2  examples/sec;  0.066  sec/batch)
step 66440, loss=0.57  (1886.7  examples/sec;  0.068  sec/batch)
step 66450, loss=0.59  (1952.6  examples/sec;  0.066  sec/batch)
step 66460, loss=0.74  (1856.4  examples/sec;  0.069  sec/batch)
step 66470, loss=0.56  (2001.1  examples/sec;  0.064  sec/batch)
step 66480, loss=0.62  (1846.6  examples/sec;  0.069  sec/batch)
step 66490, loss=0.61  (1974.7  examples/sec;  0.065  sec/batch)
step 66500, loss=0.53  (1871.0  examples/sec;  0.068  sec/batch)
step 66510, loss=0.45  (1877.8  examples/sec;  0.068  sec/batch)
step 66520, loss=0.66  (1910.9  examples/sec;  0.067  sec/batch)
step 66530, loss=0.63  (1841.3  examples/sec;  0.070  sec/batch)
step 66540, loss=0.57  (1861.7  examples/sec;  0.069  sec/batch)
step 66550, loss=0.56  (1913.0  examples/sec;  0.067  sec/batch)
step 66560, loss=0.54  (1917.9  examples/sec;  0.067  sec/batch)
step 66570, loss=0.88  (1920.5  examples/sec;  0.067  sec/batch)
step 66580, loss=0.63  (1798.1  examples/sec;  0.071  sec/batch)
step 66590, loss=0.72  (2054.9  examples/sec;  0.062  sec/batch)
step 66600, loss=0.52  (1942.3  examples/sec;  0.066  sec/batch)
step 66610, loss=0.75  (1677.2  examples/sec;  0.076  sec/batch)
step 66620, loss=0.66  (1922.2  examples/sec;  0.067  sec/batch)
step 66630, loss=0.58  (1860.9  examples/sec;  0.069  sec/batch)
step 66640, loss=0.66  (1899.7  examples/sec;  0.067  sec/batch)
step 66650, loss=0.73  (1965.2  examples/sec;  0.065  sec/batch)
step 66660, loss=0.63  (1913.2  examples/sec;  0.067  sec/batch)
step 66670, loss=0.72  (1891.4  examples/sec;  0.068  sec/batch)
step 66680, loss=0.62  (1918.0  examples/sec;  0.067  sec/batch)
step 66690, loss=0.60  (1968.8  examples/sec;  0.065  sec/batch)
step 66700, loss=0.64  (1925.2  examples/sec;  0.066  sec/batch)
step 66710, loss=0.59  (1895.9  examples/sec;  0.068  sec/batch)
step 66720, loss=0.56  (2011.5  examples/sec;  0.064  sec/batch)
step 66730, loss=0.69  (1868.4  examples/sec;  0.069  sec/batch)
step 66740, loss=0.66  (1947.0  examples/sec;  0.066  sec/batch)
step 66750, loss=0.63  (1890.6  examples/sec;  0.068  sec/batch)
step 66760, loss=0.64  (1910.0  examples/sec;  0.067  sec/batch)
step 66770, loss=0.67  (1857.7  examples/sec;  0.069  sec/batch)
step 66780, loss=0.61  (1897.4  examples/sec;  0.067  sec/batch)
step 66790, loss=0.65  (1893.9  examples/sec;  0.068  sec/batch)
step 66800, loss=0.67  (1915.6  examples/sec;  0.067  sec/batch)
step 66810, loss=0.67  (1928.5  examples/sec;  0.066  sec/batch)
step 66820, loss=0.63  (1880.7  examples/sec;  0.068  sec/batch)
step 66830, loss=0.79  (1924.1  examples/sec;  0.067  sec/batch)
step 66840, loss=0.64  (1855.2  examples/sec;  0.069  sec/batch)
step 66850, loss=0.50  (1926.8  examples/sec;  0.066  sec/batch)
step 66860, loss=0.67  (1875.7  examples/sec;  0.068  sec/batch)
step 66870, loss=0.68  (1959.1  examples/sec;  0.065  sec/batch)
step 66880, loss=0.53  (1873.1  examples/sec;  0.068  sec/batch)
step 66890, loss=0.60  (1892.3  examples/sec;  0.068  sec/batch)
step 66900, loss=0.65  (1939.1  examples/sec;  0.066  sec/batch)
step 66910, loss=0.65  (1876.1  examples/sec;  0.068  sec/batch)
step 66920, loss=0.66  (1895.7  examples/sec;  0.068  sec/batch)
step 66930, loss=0.61  (1943.7  examples/sec;  0.066  sec/batch)
step 66940, loss=0.66  (2025.9  examples/sec;  0.063  sec/batch)
step 66950, loss=0.77  (1907.3  examples/sec;  0.067  sec/batch)
step 66960, loss=0.58  (1899.1  examples/sec;  0.067  sec/batch)
step 66970, loss=0.49  (1885.4  examples/sec;  0.068  sec/batch)
step 66980, loss=0.71  (1836.3  examples/sec;  0.070  sec/batch)
step 66990, loss=0.61  (1825.2  examples/sec;  0.070  sec/batch)
step 67000, loss=0.70  (1958.6  examples/sec;  0.065  sec/batch)
step 67010, loss=0.63  (1913.8  examples/sec;  0.067  sec/batch)
step 67020, loss=0.48  (1885.9  examples/sec;  0.068  sec/batch)
step 67030, loss=0.65  (1888.2  examples/sec;  0.068  sec/batch)
step 67040, loss=0.60  (1827.2  examples/sec;  0.070  sec/batch)
step 67050, loss=0.66  (1865.7  examples/sec;  0.069  sec/batch)
step 67060, loss=0.50  (1948.5  examples/sec;  0.066  sec/batch)
step 67070, loss=0.60  (1903.5  examples/sec;  0.067  sec/batch)
step 67080, loss=0.57  (1905.0  examples/sec;  0.067  sec/batch)
step 67090, loss=0.68  (1968.4  examples/sec;  0.065  sec/batch)
step 67100, loss=0.60  (1927.8  examples/sec;  0.066  sec/batch)
step 67110, loss=0.61  (1734.2  examples/sec;  0.074  sec/batch)
step 67120, loss=0.69  (1945.6  examples/sec;  0.066  sec/batch)
step 67130, loss=0.51  (1799.0  examples/sec;  0.071  sec/batch)
step 67140, loss=0.54  (1842.5  examples/sec;  0.069  sec/batch)
step 67150, loss=0.84  (1852.1  examples/sec;  0.069  sec/batch)
step 67160, loss=0.72  (1921.3  examples/sec;  0.067  sec/batch)
step 67170, loss=0.59  (1978.1  examples/sec;  0.065  sec/batch)
step 67180, loss=0.60  (1926.7  examples/sec;  0.066  sec/batch)
step 67190, loss=0.60  (1940.9  examples/sec;  0.066  sec/batch)
step 67200, loss=0.71  (1866.7  examples/sec;  0.069  sec/batch)
step 67210, loss=0.59  (2000.4  examples/sec;  0.064  sec/batch)
step 67220, loss=0.74  (1901.4  examples/sec;  0.067  sec/batch)
step 67230, loss=0.63  (1876.0  examples/sec;  0.068  sec/batch)
step 67240, loss=0.70  (1964.6  examples/sec;  0.065  sec/batch)
step 67250, loss=0.65  (1825.7  examples/sec;  0.070  sec/batch)
step 67260, loss=0.77  (1900.9  examples/sec;  0.067  sec/batch)
step 67270, loss=0.61  (1968.3  examples/sec;  0.065  sec/batch)
step 67280, loss=0.56  (1883.5  examples/sec;  0.068  sec/batch)
step 67290, loss=0.46  (1930.3  examples/sec;  0.066  sec/batch)
step 67300, loss=0.50  (1926.8  examples/sec;  0.066  sec/batch)
step 67310, loss=0.68  (1977.0  examples/sec;  0.065  sec/batch)
step 67320, loss=0.59  (1854.9  examples/sec;  0.069  sec/batch)
step 67330, loss=0.62  (1963.3  examples/sec;  0.065  sec/batch)
step 67340, loss=0.59  (1907.9  examples/sec;  0.067  sec/batch)
step 67350, loss=0.67  (1983.4  examples/sec;  0.065  sec/batch)
step 67360, loss=0.65  (1846.2  examples/sec;  0.069  sec/batch)
step 67370, loss=0.66  (1881.7  examples/sec;  0.068  sec/batch)
step 67380, loss=0.61  (1800.4  examples/sec;  0.071  sec/batch)
step 67390, loss=0.68  (1866.4  examples/sec;  0.069  sec/batch)
step 67400, loss=0.56  (1943.6  examples/sec;  0.066  sec/batch)
step 67410, loss=0.59  (1901.2  examples/sec;  0.067  sec/batch)
step 67420, loss=0.66  (1936.4  examples/sec;  0.066  sec/batch)
step 67430, loss=0.63  (1802.3  examples/sec;  0.071  sec/batch)
step 67440, loss=0.55  (1893.0  examples/sec;  0.068  sec/batch)
step 67450, loss=0.83  (1960.6  examples/sec;  0.065  sec/batch)
step 67460, loss=0.71  (1925.9  examples/sec;  0.066  sec/batch)
step 67470, loss=0.68  (1975.9  examples/sec;  0.065  sec/batch)
step 67480, loss=0.55  (1843.0  examples/sec;  0.069  sec/batch)
step 67490, loss=0.60  (1919.5  examples/sec;  0.067  sec/batch)
step 67500, loss=0.57  (1838.7  examples/sec;  0.070  sec/batch)
step 67510, loss=0.69  (1976.6  examples/sec;  0.065  sec/batch)
step 67520, loss=0.58  (1839.3  examples/sec;  0.070  sec/batch)
step 67530, loss=0.50  (1961.4  examples/sec;  0.065  sec/batch)
step 67540, loss=0.50  (1832.7  examples/sec;  0.070  sec/batch)
step 67550, loss=0.60  (1855.7  examples/sec;  0.069  sec/batch)
step 67560, loss=0.62  (1885.4  examples/sec;  0.068  sec/batch)
step 67570, loss=0.58  (1899.6  examples/sec;  0.067  sec/batch)
step 67580, loss=0.51  (1956.7  examples/sec;  0.065  sec/batch)
step 67590, loss=0.70  (1937.9  examples/sec;  0.066  sec/batch)
step 67600, loss=0.56  (1879.6  examples/sec;  0.068  sec/batch)
step 67610, loss=0.50  (1899.7  examples/sec;  0.067  sec/batch)
step 67620, loss=0.58  (1956.5  examples/sec;  0.065  sec/batch)
step 67630, loss=0.66  (1933.9  examples/sec;  0.066  sec/batch)
step 67640, loss=0.72  (1929.1  examples/sec;  0.066  sec/batch)
step 67650, loss=0.72  (1849.8  examples/sec;  0.069  sec/batch)
step 67660, loss=0.64  (1875.4  examples/sec;  0.068  sec/batch)
step 67670, loss=0.69  (1957.7  examples/sec;  0.065  sec/batch)
step 67680, loss=0.89  (1958.8  examples/sec;  0.065  sec/batch)
step 67690, loss=0.59  (1938.8  examples/sec;  0.066  sec/batch)
step 67700, loss=0.50  (1899.0  examples/sec;  0.067  sec/batch)
step 67710, loss=0.63  (1852.1  examples/sec;  0.069  sec/batch)
step 67720, loss=0.76  (1871.1  examples/sec;  0.068  sec/batch)
step 67730, loss=0.63  (1946.8  examples/sec;  0.066  sec/batch)
step 67740, loss=0.67  (1846.2  examples/sec;  0.069  sec/batch)
step 67750, loss=0.45  (1832.0  examples/sec;  0.070  sec/batch)
step 67760, loss=0.60  (1800.3  examples/sec;  0.071  sec/batch)
step 67770, loss=0.59  (1908.1  examples/sec;  0.067  sec/batch)
step 67780, loss=0.61  (1874.4  examples/sec;  0.068  sec/batch)
step 67790, loss=0.58  (1860.9  examples/sec;  0.069  sec/batch)
step 67800, loss=0.68  (1864.3  examples/sec;  0.069  sec/batch)
step 67810, loss=0.67  (1887.9  examples/sec;  0.068  sec/batch)
step 67820, loss=0.56  (1895.1  examples/sec;  0.068  sec/batch)
step 67830, loss=0.52  (1961.3  examples/sec;  0.065  sec/batch)
step 67840, loss=0.71  (1815.1  examples/sec;  0.071  sec/batch)
step 67850, loss=0.68  (1914.9  examples/sec;  0.067  sec/batch)
step 67860, loss=0.66  (1935.7  examples/sec;  0.066  sec/batch)
step 67870, loss=0.45  (1914.6  examples/sec;  0.067  sec/batch)
step 67880, loss=0.63  (1646.4  examples/sec;  0.078  sec/batch)
step 67890, loss=0.48  (1924.0  examples/sec;  0.067  sec/batch)
step 67900, loss=0.56  (1875.0  examples/sec;  0.068  sec/batch)
step 67910, loss=0.75  (1967.9  examples/sec;  0.065  sec/batch)
step 67920, loss=0.57  (1869.6  examples/sec;  0.068  sec/batch)
step 67930, loss=0.78  (1880.1  examples/sec;  0.068  sec/batch)
step 67940, loss=0.56  (1983.1  examples/sec;  0.065  sec/batch)
step 67950, loss=0.63  (1959.8  examples/sec;  0.065  sec/batch)
step 67960, loss=0.66  (1885.5  examples/sec;  0.068  sec/batch)
step 67970, loss=0.64  (1926.5  examples/sec;  0.066  sec/batch)
step 67980, loss=0.70  (1899.6  examples/sec;  0.067  sec/batch)
step 67990, loss=0.58  (1870.2  examples/sec;  0.068  sec/batch)
step 68000, loss=0.61  (1945.7  examples/sec;  0.066  sec/batch)
step 68010, loss=0.55  (1883.0  examples/sec;  0.068  sec/batch)
step 68020, loss=0.49  (1972.1  examples/sec;  0.065  sec/batch)
step 68030, loss=0.63  (1941.0  examples/sec;  0.066  sec/batch)
step 68040, loss=0.68  (1766.2  examples/sec;  0.072  sec/batch)
step 68050, loss=0.75  (1830.1  examples/sec;  0.070  sec/batch)
step 68060, loss=0.60  (1885.0  examples/sec;  0.068  sec/batch)
step 68070, loss=0.69  (1783.6  examples/sec;  0.072  sec/batch)
step 68080, loss=0.64  (1980.7  examples/sec;  0.065  sec/batch)
step 68090, loss=0.62  (2013.2  examples/sec;  0.064  sec/batch)
step 68100, loss=0.58  (1875.0  examples/sec;  0.068  sec/batch)
step 68110, loss=0.47  (1981.9  examples/sec;  0.065  sec/batch)
step 68120, loss=0.56  (1831.0  examples/sec;  0.070  sec/batch)
step 68130, loss=0.63  (1775.2  examples/sec;  0.072  sec/batch)
step 68140, loss=0.60  (1692.8  examples/sec;  0.076  sec/batch)
step 68150, loss=0.56  (1905.1  examples/sec;  0.067  sec/batch)
step 68160, loss=0.65  (1905.4  examples/sec;  0.067  sec/batch)
step 68170, loss=0.66  (2010.9  examples/sec;  0.064  sec/batch)
step 68180, loss=0.83  (1942.5  examples/sec;  0.066  sec/batch)
step 68190, loss=0.55  (2005.4  examples/sec;  0.064  sec/batch)
step 68200, loss=0.64  (1916.7  examples/sec;  0.067  sec/batch)
step 68210, loss=0.54  (1876.1  examples/sec;  0.068  sec/batch)
step 68220, loss=0.65  (1948.2  examples/sec;  0.066  sec/batch)
step 68230, loss=0.82  (1859.7  examples/sec;  0.069  sec/batch)
step 68240, loss=0.59  (1887.1  examples/sec;  0.068  sec/batch)
step 68250, loss=0.58  (1955.7  examples/sec;  0.065  sec/batch)
step 68260, loss=0.57  (1898.8  examples/sec;  0.067  sec/batch)
step 68270, loss=0.63  (1850.2  examples/sec;  0.069  sec/batch)
step 68280, loss=0.66  (1920.6  examples/sec;  0.067  sec/batch)
step 68290, loss=0.79  (1875.9  examples/sec;  0.068  sec/batch)
step 68300, loss=0.59  (1897.7  examples/sec;  0.067  sec/batch)
step 68310, loss=0.71  (1863.0  examples/sec;  0.069  sec/batch)
step 68320, loss=0.63  (1898.3  examples/sec;  0.067  sec/batch)
step 68330, loss=0.57  (1933.7  examples/sec;  0.066  sec/batch)
step 68340, loss=0.58  (1937.7  examples/sec;  0.066  sec/batch)
step 68350, loss=0.83  (1888.2  examples/sec;  0.068  sec/batch)
step 68360, loss=0.55  (1765.9  examples/sec;  0.072  sec/batch)
step 68370, loss=0.73  (1884.2  examples/sec;  0.068  sec/batch)
step 68380, loss=0.71  (1882.7  examples/sec;  0.068  sec/batch)
step 68390, loss=0.47  (1955.9  examples/sec;  0.065  sec/batch)
step 68400, loss=0.77  (1927.0  examples/sec;  0.066  sec/batch)
step 68410, loss=0.74  (1914.7  examples/sec;  0.067  sec/batch)
step 68420, loss=0.70  (1827.6  examples/sec;  0.070  sec/batch)
step 68430, loss=0.85  (1840.3  examples/sec;  0.070  sec/batch)
step 68440, loss=0.66  (1900.0  examples/sec;  0.067  sec/batch)
step 68450, loss=0.58  (1859.5  examples/sec;  0.069  sec/batch)
step 68460, loss=0.57  (1869.6  examples/sec;  0.068  sec/batch)
step 68470, loss=0.79  (1948.8  examples/sec;  0.066  sec/batch)
step 68480, loss=0.61  (1973.3  examples/sec;  0.065  sec/batch)
step 68490, loss=0.61  (1926.0  examples/sec;  0.066  sec/batch)
step 68500, loss=0.59  (1951.0  examples/sec;  0.066  sec/batch)
step 68510, loss=0.66  (1841.7  examples/sec;  0.070  sec/batch)
step 68520, loss=0.81  (1909.6  examples/sec;  0.067  sec/batch)
step 68530, loss=0.72  (1959.8  examples/sec;  0.065  sec/batch)
step 68540, loss=0.76  (1841.6  examples/sec;  0.070  sec/batch)
step 68550, loss=0.49  (1980.7  examples/sec;  0.065  sec/batch)
step 68560, loss=0.55  (2026.3  examples/sec;  0.063  sec/batch)
step 68570, loss=0.70  (1970.8  examples/sec;  0.065  sec/batch)
step 68580, loss=0.77  (1853.2  examples/sec;  0.069  sec/batch)
step 68590, loss=0.58  (1811.3  examples/sec;  0.071  sec/batch)
step 68600, loss=0.55  (1854.5  examples/sec;  0.069  sec/batch)
step 68610, loss=0.70  (1952.1  examples/sec;  0.066  sec/batch)
step 68620, loss=0.65  (1922.1  examples/sec;  0.067  sec/batch)
step 68630, loss=0.68  (1908.3  examples/sec;  0.067  sec/batch)
step 68640, loss=0.47  (1910.4  examples/sec;  0.067  sec/batch)
step 68650, loss=0.55  (1892.8  examples/sec;  0.068  sec/batch)
step 68660, loss=0.61  (1940.7  examples/sec;  0.066  sec/batch)
step 68670, loss=0.66  (1891.1  examples/sec;  0.068  sec/batch)
step 68680, loss=0.50  (1922.4  examples/sec;  0.067  sec/batch)
step 68690, loss=0.68  (1882.3  examples/sec;  0.068  sec/batch)
step 68700, loss=0.68  (1943.0  examples/sec;  0.066  sec/batch)
step 68710, loss=0.70  (1890.1  examples/sec;  0.068  sec/batch)
step 68720, loss=0.58  (1800.2  examples/sec;  0.071  sec/batch)
step 68730, loss=0.70  (1867.4  examples/sec;  0.069  sec/batch)
step 68740, loss=0.69  (1571.5  examples/sec;  0.081  sec/batch)
step 68750, loss=0.53  (1890.4  examples/sec;  0.068  sec/batch)
step 68760, loss=0.70  (1972.3  examples/sec;  0.065  sec/batch)
step 68770, loss=0.70  (1868.3  examples/sec;  0.069  sec/batch)
step 68780, loss=0.56  (1898.3  examples/sec;  0.067  sec/batch)
step 68790, loss=0.63  (1929.8  examples/sec;  0.066  sec/batch)
step 68800, loss=0.63  (1779.1  examples/sec;  0.072  sec/batch)
step 68810, loss=0.63  (1959.9  examples/sec;  0.065  sec/batch)
step 68820, loss=0.71  (1974.6  examples/sec;  0.065  sec/batch)
step 68830, loss=0.50  (1876.4  examples/sec;  0.068  sec/batch)
step 68840, loss=0.61  (1882.3  examples/sec;  0.068  sec/batch)
step 68850, loss=0.44  (1946.6  examples/sec;  0.066  sec/batch)
step 68860, loss=0.58  (1741.0  examples/sec;  0.074  sec/batch)
step 68870, loss=0.65  (1884.0  examples/sec;  0.068  sec/batch)
step 68880, loss=0.59  (1843.5  examples/sec;  0.069  sec/batch)
step 68890, loss=0.60  (1863.0  examples/sec;  0.069  sec/batch)
step 68900, loss=0.75  (1912.0  examples/sec;  0.067  sec/batch)
step 68910, loss=0.73  (1884.3  examples/sec;  0.068  sec/batch)
step 68920, loss=0.53  (1947.4  examples/sec;  0.066  sec/batch)
step 68930, loss=0.60  (1833.3  examples/sec;  0.070  sec/batch)
step 68940, loss=0.57  (1895.0  examples/sec;  0.068  sec/batch)
step 68950, loss=0.68  (1925.6  examples/sec;  0.066  sec/batch)
step 68960, loss=0.74  (1962.4  examples/sec;  0.065  sec/batch)
step 68970, loss=0.69  (1785.6  examples/sec;  0.072  sec/batch)
step 68980, loss=0.59  (1907.8  examples/sec;  0.067  sec/batch)
step 68990, loss=0.67  (1964.6  examples/sec;  0.065  sec/batch)
step 69000, loss=0.61  (1644.3  examples/sec;  0.078  sec/batch)
step 69010, loss=0.70  (1918.7  examples/sec;  0.067  sec/batch)
step 69020, loss=0.66  (1829.4  examples/sec;  0.070  sec/batch)
step 69030, loss=0.54  (1931.8  examples/sec;  0.066  sec/batch)
step 69040, loss=0.71  (1461.3  examples/sec;  0.088  sec/batch)
step 69050, loss=0.63  (1916.7  examples/sec;  0.067  sec/batch)
step 69060, loss=0.56  (1935.5  examples/sec;  0.066  sec/batch)
step 69070, loss=0.67  (1936.1  examples/sec;  0.066  sec/batch)
step 69080, loss=0.45  (1951.0  examples/sec;  0.066  sec/batch)
step 69090, loss=0.62  (1880.5  examples/sec;  0.068  sec/batch)
step 69100, loss=0.78  (1940.5  examples/sec;  0.066  sec/batch)
step 69110, loss=0.55  (1850.4  examples/sec;  0.069  sec/batch)
step 69120, loss=0.59  (1829.1  examples/sec;  0.070  sec/batch)
step 69130, loss=0.50  (1952.1  examples/sec;  0.066  sec/batch)
step 69140, loss=0.53  (1847.8  examples/sec;  0.069  sec/batch)
step 69150, loss=0.66  (1958.3  examples/sec;  0.065  sec/batch)
step 69160, loss=0.65  (1946.5  examples/sec;  0.066  sec/batch)
step 69170, loss=0.79  (1804.9  examples/sec;  0.071  sec/batch)
step 69180, loss=0.66  (1857.7  examples/sec;  0.069  sec/batch)
step 69190, loss=0.58  (1875.4  examples/sec;  0.068  sec/batch)
step 69200, loss=0.65  (1914.6  examples/sec;  0.067  sec/batch)
step 69210, loss=0.68  (1917.8  examples/sec;  0.067  sec/batch)
step 69220, loss=0.61  (1839.4  examples/sec;  0.070  sec/batch)
step 69230, loss=0.81  (1880.6  examples/sec;  0.068  sec/batch)
step 69240, loss=0.65  (1917.6  examples/sec;  0.067  sec/batch)
step 69250, loss=0.52  (1886.0  examples/sec;  0.068  sec/batch)
step 69260, loss=0.53  (1879.5  examples/sec;  0.068  sec/batch)
step 69270, loss=0.55  (1942.3  examples/sec;  0.066  sec/batch)
step 69280, loss=0.48  (1913.2  examples/sec;  0.067  sec/batch)
step 69290, loss=0.43  (1850.1  examples/sec;  0.069  sec/batch)
step 69300, loss=0.55  (1894.7  examples/sec;  0.068  sec/batch)
step 69310, loss=0.56  (1968.1  examples/sec;  0.065  sec/batch)
step 69320, loss=0.69  (1885.7  examples/sec;  0.068  sec/batch)
step 69330, loss=0.67  (1813.7  examples/sec;  0.071  sec/batch)
step 69340, loss=0.53  (1881.4  examples/sec;  0.068  sec/batch)
step 69350, loss=0.59  (1817.5  examples/sec;  0.070  sec/batch)
step 69360, loss=0.77  (1964.0  examples/sec;  0.065  sec/batch)
step 69370, loss=0.67  (1479.5  examples/sec;  0.087  sec/batch)
step 69380, loss=0.66  (1905.0  examples/sec;  0.067  sec/batch)
step 69390, loss=0.70  (1843.2  examples/sec;  0.069  sec/batch)
step 69400, loss=0.66  (1965.9  examples/sec;  0.065  sec/batch)
step 69410, loss=0.55  (1858.0  examples/sec;  0.069  sec/batch)
step 69420, loss=0.52  (1966.5  examples/sec;  0.065  sec/batch)
step 69430, loss=0.55  (1819.0  examples/sec;  0.070  sec/batch)
step 69440, loss=0.63  (1942.9  examples/sec;  0.066  sec/batch)
step 69450, loss=0.65  (1812.0  examples/sec;  0.071  sec/batch)
step 69460, loss=0.43  (1959.0  examples/sec;  0.065  sec/batch)
step 69470, loss=0.56  (1916.1  examples/sec;  0.067  sec/batch)
step 69480, loss=0.72  (1947.0  examples/sec;  0.066  sec/batch)
step 69490, loss=0.58  (1857.1  examples/sec;  0.069  sec/batch)
step 69500, loss=0.58  (1960.2  examples/sec;  0.065  sec/batch)
step 69510, loss=0.72  (1971.1  examples/sec;  0.065  sec/batch)
step 69520, loss=0.60  (1967.7  examples/sec;  0.065  sec/batch)
step 69530, loss=0.60  (1957.7  examples/sec;  0.065  sec/batch)
step 69540, loss=0.66  (1883.2  examples/sec;  0.068  sec/batch)
step 69550, loss=0.55  (1920.8  examples/sec;  0.067  sec/batch)
step 69560, loss=0.53  (1898.6  examples/sec;  0.067  sec/batch)
step 69570, loss=0.63  (1772.6  examples/sec;  0.072  sec/batch)
step 69580, loss=0.64  (2012.1  examples/sec;  0.064  sec/batch)
step 69590, loss=0.60  (1888.3  examples/sec;  0.068  sec/batch)
step 69600, loss=0.68  (1890.0  examples/sec;  0.068  sec/batch)
step 69610, loss=0.51  (1939.5  examples/sec;  0.066  sec/batch)
step 69620, loss=0.53  (1925.7  examples/sec;  0.066  sec/batch)
step 69630, loss=0.68  (1915.6  examples/sec;  0.067  sec/batch)
step 69640, loss=0.67  (1936.0  examples/sec;  0.066  sec/batch)
step 69650, loss=0.64  (1840.6  examples/sec;  0.070  sec/batch)
step 69660, loss=0.58  (1845.6  examples/sec;  0.069  sec/batch)
step 69670, loss=0.52  (1941.8  examples/sec;  0.066  sec/batch)
step 69680, loss=0.59  (1934.8  examples/sec;  0.066  sec/batch)
step 69690, loss=0.45  (1809.4  examples/sec;  0.071  sec/batch)
step 69700, loss=0.56  (1918.3  examples/sec;  0.067  sec/batch)
step 69710, loss=0.63  (1884.6  examples/sec;  0.068  sec/batch)
step 69720, loss=0.64  (1896.5  examples/sec;  0.067  sec/batch)
step 69730, loss=0.79  (1929.3  examples/sec;  0.066  sec/batch)
step 69740, loss=0.57  (1876.6  examples/sec;  0.068  sec/batch)
step 69750, loss=0.56  (1942.1  examples/sec;  0.066  sec/batch)
step 69760, loss=0.64  (1966.9  examples/sec;  0.065  sec/batch)
step 69770, loss=0.65  (1912.5  examples/sec;  0.067  sec/batch)
step 69780, loss=0.65  (2012.6  examples/sec;  0.064  sec/batch)
step 69790, loss=0.68  (1995.7  examples/sec;  0.064  sec/batch)
step 69800, loss=0.61  (1854.3  examples/sec;  0.069  sec/batch)
step 69810, loss=0.64  (1889.1  examples/sec;  0.068  sec/batch)
step 69820, loss=0.48  (1961.2  examples/sec;  0.065  sec/batch)
step 69830, loss=0.67  (1739.5  examples/sec;  0.074  sec/batch)
step 69840, loss=0.53  (1929.4  examples/sec;  0.066  sec/batch)
step 69850, loss=0.71  (1956.9  examples/sec;  0.065  sec/batch)
step 69860, loss=0.59  (1989.1  examples/sec;  0.064  sec/batch)
step 69870, loss=0.65  (1882.9  examples/sec;  0.068  sec/batch)
step 69880, loss=0.73  (1899.1  examples/sec;  0.067  sec/batch)
step 69890, loss=0.51  (1951.8  examples/sec;  0.066  sec/batch)
step 69900, loss=0.72  (1945.9  examples/sec;  0.066  sec/batch)
step 69910, loss=0.57  (1839.8  examples/sec;  0.070  sec/batch)
step 69920, loss=0.52  (1918.9  examples/sec;  0.067  sec/batch)
step 69930, loss=0.65  (1944.3  examples/sec;  0.066  sec/batch)
step 69940, loss=0.65  (1891.7  examples/sec;  0.068  sec/batch)
step 69950, loss=0.54  (1886.3  examples/sec;  0.068  sec/batch)
step 69960, loss=0.58  (1867.1  examples/sec;  0.069  sec/batch)
step 69970, loss=0.61  (1820.4  examples/sec;  0.070  sec/batch)
step 69980, loss=0.64  (1926.3  examples/sec;  0.066  sec/batch)
step 69990, loss=0.63  (1917.4  examples/sec;  0.067  sec/batch)
step 70000, loss=0.69  (1963.2  examples/sec;  0.065  sec/batch)
step 70010, loss=0.54  (1948.5  examples/sec;  0.066  sec/batch)
step 70020, loss=0.62  (1822.9  examples/sec;  0.070  sec/batch)
step 70030, loss=0.50  (1877.4  examples/sec;  0.068  sec/batch)
step 70040, loss=0.58  (1920.6  examples/sec;  0.067  sec/batch)
step 70050, loss=0.82  (1971.3  examples/sec;  0.065  sec/batch)
step 70060, loss=0.65  (1908.5  examples/sec;  0.067  sec/batch)
step 70070, loss=0.63  (1939.2  examples/sec;  0.066  sec/batch)
step 70080, loss=0.64  (1851.5  examples/sec;  0.069  sec/batch)
step 70090, loss=0.60  (1888.3  examples/sec;  0.068  sec/batch)
step 70100, loss=0.54  (1861.9  examples/sec;  0.069  sec/batch)
step 70110, loss=0.65  (1752.3  examples/sec;  0.073  sec/batch)
step 70120, loss=0.68  (1857.0  examples/sec;  0.069  sec/batch)
step 70130, loss=0.64  (1872.8  examples/sec;  0.068  sec/batch)
step 70140, loss=0.63  (1927.1  examples/sec;  0.066  sec/batch)
step 70150, loss=0.63  (1902.4  examples/sec;  0.067  sec/batch)
step 70160, loss=0.82  (1858.0  examples/sec;  0.069  sec/batch)
step 70170, loss=0.51  (1903.9  examples/sec;  0.067  sec/batch)
step 70180, loss=0.72  (1924.3  examples/sec;  0.067  sec/batch)
step 70190, loss=0.61  (1919.2  examples/sec;  0.067  sec/batch)
step 70200, loss=0.60  (1935.7  examples/sec;  0.066  sec/batch)
step 70210, loss=0.69  (1981.9  examples/sec;  0.065  sec/batch)
step 70220, loss=0.66  (1862.6  examples/sec;  0.069  sec/batch)
step 70230, loss=0.73  (1845.6  examples/sec;  0.069  sec/batch)
step 70240, loss=0.57  (1859.3  examples/sec;  0.069  sec/batch)
step 70250, loss=0.82  (1717.0  examples/sec;  0.075  sec/batch)
step 70260, loss=0.64  (1900.4  examples/sec;  0.067  sec/batch)
step 70270, loss=0.71  (1888.2  examples/sec;  0.068  sec/batch)
step 70280, loss=0.50  (1896.7  examples/sec;  0.067  sec/batch)
step 70290, loss=0.58  (1877.5  examples/sec;  0.068  sec/batch)
step 70300, loss=0.75  (1940.2  examples/sec;  0.066  sec/batch)
step 70310, loss=0.72  (1989.8  examples/sec;  0.064  sec/batch)
step 70320, loss=0.63  (1952.3  examples/sec;  0.066  sec/batch)
step 70330, loss=0.55  (1893.8  examples/sec;  0.068  sec/batch)
step 70340, loss=0.50  (1938.8  examples/sec;  0.066  sec/batch)
step 70350, loss=0.52  (1850.9  examples/sec;  0.069  sec/batch)
step 70360, loss=0.66  (1914.4  examples/sec;  0.067  sec/batch)
step 70370, loss=0.70  (1886.2  examples/sec;  0.068  sec/batch)
step 70380, loss=0.72  (1944.8  examples/sec;  0.066  sec/batch)
step 70390, loss=0.62  (1931.5  examples/sec;  0.066  sec/batch)
step 70400, loss=0.73  (1857.9  examples/sec;  0.069  sec/batch)
step 70410, loss=0.83  (1902.6  examples/sec;  0.067  sec/batch)
step 70420, loss=0.74  (1879.6  examples/sec;  0.068  sec/batch)
step 70430, loss=0.65  (1837.8  examples/sec;  0.070  sec/batch)
step 70440, loss=0.65  (1872.3  examples/sec;  0.068  sec/batch)
step 70450, loss=0.61  (1891.5  examples/sec;  0.068  sec/batch)
step 70460, loss=0.67  (2027.2  examples/sec;  0.063  sec/batch)
step 70470, loss=0.53  (2013.1  examples/sec;  0.064  sec/batch)
step 70480, loss=0.66  (2042.5  examples/sec;  0.063  sec/batch)
step 70490, loss=0.54  (1832.9  examples/sec;  0.070  sec/batch)
step 70500, loss=0.55  (1806.5  examples/sec;  0.071  sec/batch)
step 70510, loss=0.71  (1993.5  examples/sec;  0.064  sec/batch)
step 70520, loss=0.61  (1916.7  examples/sec;  0.067  sec/batch)
step 70530, loss=0.66  (1902.6  examples/sec;  0.067  sec/batch)
step 70540, loss=0.57  (1916.3  examples/sec;  0.067  sec/batch)
step 70550, loss=0.70  (1929.5  examples/sec;  0.066  sec/batch)
step 70560, loss=0.68  (1915.9  examples/sec;  0.067  sec/batch)
step 70570, loss=0.55  (1900.0  examples/sec;  0.067  sec/batch)
step 70580, loss=0.55  (1979.6  examples/sec;  0.065  sec/batch)
step 70590, loss=0.45  (1912.6  examples/sec;  0.067  sec/batch)
step 70600, loss=0.60  (1855.7  examples/sec;  0.069  sec/batch)
step 70610, loss=0.67  (1915.7  examples/sec;  0.067  sec/batch)
step 70620, loss=0.55  (1945.8  examples/sec;  0.066  sec/batch)
step 70630, loss=0.59  (1887.0  examples/sec;  0.068  sec/batch)
step 70640, loss=0.86  (1882.6  examples/sec;  0.068  sec/batch)
step 70650, loss=0.58  (1967.8  examples/sec;  0.065  sec/batch)
step 70660, loss=0.64  (1918.3  examples/sec;  0.067  sec/batch)
step 70670, loss=0.77  (1860.7  examples/sec;  0.069  sec/batch)
step 70680, loss=0.73  (1872.7  examples/sec;  0.068  sec/batch)
step 70690, loss=0.66  (1765.0  examples/sec;  0.073  sec/batch)
step 70700, loss=0.56  (1880.4  examples/sec;  0.068  sec/batch)
step 70710, loss=0.58  (2031.7  examples/sec;  0.063  sec/batch)
step 70720, loss=0.62  (1923.8  examples/sec;  0.067  sec/batch)
step 70730, loss=0.75  (2043.5  examples/sec;  0.063  sec/batch)
step 70740, loss=0.62  (1847.0  examples/sec;  0.069  sec/batch)
step 70750, loss=0.64  (1955.4  examples/sec;  0.065  sec/batch)
step 70760, loss=0.69  (1839.8  examples/sec;  0.070  sec/batch)
step 70770, loss=0.54  (1795.4  examples/sec;  0.071  sec/batch)
step 70780, loss=0.65  (1941.6  examples/sec;  0.066  sec/batch)
step 70790, loss=0.53  (1665.1  examples/sec;  0.077  sec/batch)
step 70800, loss=0.65  (1886.9  examples/sec;  0.068  sec/batch)
step 70810, loss=0.57  (1817.5  examples/sec;  0.070  sec/batch)
step 70820, loss=0.64  (1871.7  examples/sec;  0.068  sec/batch)
step 70830, loss=0.82  (1815.5  examples/sec;  0.071  sec/batch)
step 70840, loss=0.56  (1710.6  examples/sec;  0.075  sec/batch)
step 70850, loss=0.56  (1875.6  examples/sec;  0.068  sec/batch)
step 70860, loss=0.75  (1902.4  examples/sec;  0.067  sec/batch)
step 70870, loss=0.54  (1952.2  examples/sec;  0.066  sec/batch)
step 70880, loss=0.70  (1830.2  examples/sec;  0.070  sec/batch)
step 70890, loss=0.59  (1937.5  examples/sec;  0.066  sec/batch)
step 70900, loss=0.71  (1905.6  examples/sec;  0.067  sec/batch)
step 70910, loss=0.72  (1929.3  examples/sec;  0.066  sec/batch)
step 70920, loss=0.51  (1901.5  examples/sec;  0.067  sec/batch)
step 70930, loss=0.63  (1786.7  examples/sec;  0.072  sec/batch)
step 70940, loss=0.71  (1955.8  examples/sec;  0.065  sec/batch)
step 70950, loss=0.52  (1857.8  examples/sec;  0.069  sec/batch)
step 70960, loss=0.69  (1802.1  examples/sec;  0.071  sec/batch)
step 70970, loss=0.53  (1940.6  examples/sec;  0.066  sec/batch)
step 70980, loss=0.81  (1830.7  examples/sec;  0.070  sec/batch)
step 70990, loss=0.49  (1921.1  examples/sec;  0.067  sec/batch)
step 71000, loss=0.74  (1928.0  examples/sec;  0.066  sec/batch)
step 71010, loss=0.65  (1860.7  examples/sec;  0.069  sec/batch)
step 71020, loss=0.56  (1938.8  examples/sec;  0.066  sec/batch)
step 71030, loss=0.57  (1876.6  examples/sec;  0.068  sec/batch)
step 71040, loss=0.84  (1878.1  examples/sec;  0.068  sec/batch)
step 71050, loss=0.63  (1904.9  examples/sec;  0.067  sec/batch)
step 71060, loss=0.54  (1964.2  examples/sec;  0.065  sec/batch)
step 71070, loss=0.53  (1859.0  examples/sec;  0.069  sec/batch)
step 71080, loss=0.69  (1819.8  examples/sec;  0.070  sec/batch)
step 71090, loss=0.64  (1894.9  examples/sec;  0.068  sec/batch)
step 71100, loss=0.79  (1930.9  examples/sec;  0.066  sec/batch)
step 71110, loss=0.53  (1904.2  examples/sec;  0.067  sec/batch)
step 71120, loss=0.61  (1932.4  examples/sec;  0.066  sec/batch)
step 71130, loss=0.66  (1876.8  examples/sec;  0.068  sec/batch)
step 71140, loss=0.52  (1933.7  examples/sec;  0.066  sec/batch)
step 71150, loss=0.71  (1743.5  examples/sec;  0.073  sec/batch)
step 71160, loss=0.71  (1891.1  examples/sec;  0.068  sec/batch)
step 71170, loss=0.76  (1931.2  examples/sec;  0.066  sec/batch)
step 71180, loss=0.53  (1856.0  examples/sec;  0.069  sec/batch)
step 71190, loss=0.53  (1882.6  examples/sec;  0.068  sec/batch)
step 71200, loss=0.60  (1881.1  examples/sec;  0.068  sec/batch)
step 71210, loss=0.53  (1930.3  examples/sec;  0.066  sec/batch)
step 71220, loss=0.57  (1910.8  examples/sec;  0.067  sec/batch)
step 71230, loss=0.56  (1809.8  examples/sec;  0.071  sec/batch)
step 71240, loss=0.67  (1925.8  examples/sec;  0.066  sec/batch)
step 71250, loss=0.56  (1924.2  examples/sec;  0.067  sec/batch)
step 71260, loss=0.63  (1966.0  examples/sec;  0.065  sec/batch)
step 71270, loss=0.58  (1715.5  examples/sec;  0.075  sec/batch)
step 71280, loss=0.64  (1916.8  examples/sec;  0.067  sec/batch)
step 71290, loss=0.60  (1877.1  examples/sec;  0.068  sec/batch)
step 71300, loss=0.72  (1928.7  examples/sec;  0.066  sec/batch)
step 71310, loss=0.60  (1839.1  examples/sec;  0.070  sec/batch)
step 71320, loss=0.68  (1834.2  examples/sec;  0.070  sec/batch)
step 71330, loss=0.57  (1962.0  examples/sec;  0.065  sec/batch)
step 71340, loss=0.61  (1960.8  examples/sec;  0.065  sec/batch)
step 71350, loss=0.69  (1860.8  examples/sec;  0.069  sec/batch)
step 71360, loss=0.57  (1898.5  examples/sec;  0.067  sec/batch)
step 71370, loss=0.66  (1796.0  examples/sec;  0.071  sec/batch)
step 71380, loss=0.58  (1936.8  examples/sec;  0.066  sec/batch)
step 71390, loss=0.64  (1890.2  examples/sec;  0.068  sec/batch)
step 71400, loss=0.72  (1906.9  examples/sec;  0.067  sec/batch)
step 71410, loss=0.79  (1806.3  examples/sec;  0.071  sec/batch)
step 71420, loss=0.70  (1836.4  examples/sec;  0.070  sec/batch)
step 71430, loss=0.58  (1849.3  examples/sec;  0.069  sec/batch)
step 71440, loss=0.77  (1861.7  examples/sec;  0.069  sec/batch)
step 71450, loss=0.66  (1852.7  examples/sec;  0.069  sec/batch)
step 71460, loss=0.51  (1942.4  examples/sec;  0.066  sec/batch)
step 71470, loss=0.61  (1841.8  examples/sec;  0.069  sec/batch)
step 71480, loss=0.74  (1847.4  examples/sec;  0.069  sec/batch)
step 71490, loss=0.68  (1946.5  examples/sec;  0.066  sec/batch)
step 71500, loss=0.61  (1928.2  examples/sec;  0.066  sec/batch)
step 71510, loss=0.73  (1938.8  examples/sec;  0.066  sec/batch)
step 71520, loss=0.54  (1944.1  examples/sec;  0.066  sec/batch)
step 71530, loss=0.61  (1929.5  examples/sec;  0.066  sec/batch)
step 71540, loss=0.51  (1882.1  examples/sec;  0.068  sec/batch)
step 71550, loss=0.65  (1847.5  examples/sec;  0.069  sec/batch)
step 71560, loss=0.65  (1872.0  examples/sec;  0.068  sec/batch)
step 71570, loss=0.64  (1892.9  examples/sec;  0.068  sec/batch)
step 71580, loss=0.62  (1949.9  examples/sec;  0.066  sec/batch)
step 71590, loss=0.62  (1895.7  examples/sec;  0.068  sec/batch)
step 71600, loss=0.68  (1983.5  examples/sec;  0.065  sec/batch)
step 71610, loss=0.56  (1861.7  examples/sec;  0.069  sec/batch)
step 71620, loss=0.55  (1932.9  examples/sec;  0.066  sec/batch)
step 71630, loss=0.70  (1908.7  examples/sec;  0.067  sec/batch)
step 71640, loss=0.66  (1958.7  examples/sec;  0.065  sec/batch)
step 71650, loss=0.64  (1891.4  examples/sec;  0.068  sec/batch)
step 71660, loss=0.70  (1937.1  examples/sec;  0.066  sec/batch)
step 71670, loss=0.58  (1961.0  examples/sec;  0.065  sec/batch)
step 71680, loss=0.76  (1869.6  examples/sec;  0.068  sec/batch)
step 71690, loss=0.69  (1910.6  examples/sec;  0.067  sec/batch)
step 71700, loss=0.87  (1735.5  examples/sec;  0.074  sec/batch)
step 71710, loss=0.61  (1827.1  examples/sec;  0.070  sec/batch)
step 71720, loss=0.66  (1869.7  examples/sec;  0.068  sec/batch)
step 71730, loss=0.73  (1886.5  examples/sec;  0.068  sec/batch)
step 71740, loss=0.72  (1893.7  examples/sec;  0.068  sec/batch)
step 71750, loss=0.50  (1918.5  examples/sec;  0.067  sec/batch)
step 71760, loss=0.65  (1828.6  examples/sec;  0.070  sec/batch)
step 71770, loss=0.54  (1861.0  examples/sec;  0.069  sec/batch)
step 71780, loss=0.63  (1960.5  examples/sec;  0.065  sec/batch)
step 71790, loss=0.72  (1992.4  examples/sec;  0.064  sec/batch)
step 71800, loss=0.60  (1907.4  examples/sec;  0.067  sec/batch)
step 71810, loss=0.55  (1991.4  examples/sec;  0.064  sec/batch)
step 71820, loss=0.59  (1909.6  examples/sec;  0.067  sec/batch)
step 71830, loss=0.61  (1908.8  examples/sec;  0.067  sec/batch)
step 71840, loss=0.52  (1937.6  examples/sec;  0.066  sec/batch)
step 71850, loss=0.81  (1877.9  examples/sec;  0.068  sec/batch)
step 71860, loss=0.80  (1845.9  examples/sec;  0.069  sec/batch)
step 71870, loss=0.61  (1940.5  examples/sec;  0.066  sec/batch)
step 71880, loss=0.71  (1960.5  examples/sec;  0.065  sec/batch)
step 71890, loss=0.64  (1923.7  examples/sec;  0.067  sec/batch)
step 71900, loss=0.57  (1953.3  examples/sec;  0.066  sec/batch)
step 71910, loss=0.63  (1905.6  examples/sec;  0.067  sec/batch)
step 71920, loss=0.58  (1794.9  examples/sec;  0.071  sec/batch)
step 71930, loss=0.53  (1836.5  examples/sec;  0.070  sec/batch)
step 71940, loss=0.73  (1934.3  examples/sec;  0.066  sec/batch)
step 71950, loss=0.52  (2046.1  examples/sec;  0.063  sec/batch)
step 71960, loss=0.71  (1858.5  examples/sec;  0.069  sec/batch)
step 71970, loss=0.59  (1971.0  examples/sec;  0.065  sec/batch)
step 71980, loss=0.82  (1876.2  examples/sec;  0.068  sec/batch)
step 71990, loss=0.64  (1927.8  examples/sec;  0.066  sec/batch)
step 72000, loss=0.52  (1967.3  examples/sec;  0.065  sec/batch)
step 72010, loss=0.51  (1988.6  examples/sec;  0.064  sec/batch)
step 72020, loss=0.64  (1875.4  examples/sec;  0.068  sec/batch)
step 72030, loss=0.52  (1918.1  examples/sec;  0.067  sec/batch)
step 72040, loss=0.53  (1878.6  examples/sec;  0.068  sec/batch)
step 72050, loss=0.44  (1932.9  examples/sec;  0.066  sec/batch)
step 72060, loss=0.77  (1880.1  examples/sec;  0.068  sec/batch)
step 72070, loss=0.52  (1919.5  examples/sec;  0.067  sec/batch)
step 72080, loss=0.60  (1908.5  examples/sec;  0.067  sec/batch)
step 72090, loss=0.68  (1965.0  examples/sec;  0.065  sec/batch)
step 72100, loss=0.71  (1991.2  examples/sec;  0.064  sec/batch)
step 72110, loss=0.73  (1837.5  examples/sec;  0.070  sec/batch)
step 72120, loss=0.83  (1849.8  examples/sec;  0.069  sec/batch)
step 72130, loss=0.68  (1831.3  examples/sec;  0.070  sec/batch)
step 72140, loss=0.62  (1872.2  examples/sec;  0.068  sec/batch)
step 72150, loss=0.50  (1896.2  examples/sec;  0.068  sec/batch)
step 72160, loss=0.51  (1845.6  examples/sec;  0.069  sec/batch)
step 72170, loss=0.54  (1911.5  examples/sec;  0.067  sec/batch)
step 72180, loss=0.53  (1782.4  examples/sec;  0.072  sec/batch)
step 72190, loss=0.52  (1943.0  examples/sec;  0.066  sec/batch)
step 72200, loss=0.51  (1883.6  examples/sec;  0.068  sec/batch)
step 72210, loss=0.59  (1830.7  examples/sec;  0.070  sec/batch)
step 72220, loss=0.64  (1975.2  examples/sec;  0.065  sec/batch)
step 72230, loss=0.52  (1856.0  examples/sec;  0.069  sec/batch)
step 72240, loss=0.57  (1626.4  examples/sec;  0.079  sec/batch)
step 72250, loss=0.63  (1826.0  examples/sec;  0.070  sec/batch)
step 72260, loss=0.56  (1909.3  examples/sec;  0.067  sec/batch)
step 72270, loss=0.57  (1890.7  examples/sec;  0.068  sec/batch)
step 72280, loss=0.55  (1891.4  examples/sec;  0.068  sec/batch)
step 72290, loss=0.57  (1885.1  examples/sec;  0.068  sec/batch)
step 72300, loss=0.75  (1899.7  examples/sec;  0.067  sec/batch)
step 72310, loss=0.63  (1787.5  examples/sec;  0.072  sec/batch)
step 72320, loss=0.64  (1850.0  examples/sec;  0.069  sec/batch)
step 72330, loss=0.52  (1877.0  examples/sec;  0.068  sec/batch)
step 72340, loss=0.57  (1926.6  examples/sec;  0.066  sec/batch)
step 72350, loss=0.61  (1898.6  examples/sec;  0.067  sec/batch)
step 72360, loss=0.54  (1892.9  examples/sec;  0.068  sec/batch)
step 72370, loss=0.71  (1832.0  examples/sec;  0.070  sec/batch)
step 72380, loss=0.64  (1866.5  examples/sec;  0.069  sec/batch)
step 72390, loss=0.62  (1845.3  examples/sec;  0.069  sec/batch)
step 72400, loss=0.59  (1879.7  examples/sec;  0.068  sec/batch)
step 72410, loss=0.54  (1927.8  examples/sec;  0.066  sec/batch)
step 72420, loss=0.57  (1961.2  examples/sec;  0.065  sec/batch)
step 72430, loss=0.60  (1882.6  examples/sec;  0.068  sec/batch)
step 72440, loss=0.68  (1905.3  examples/sec;  0.067  sec/batch)
step 72450, loss=0.64  (1916.0  examples/sec;  0.067  sec/batch)
step 72460, loss=0.53  (1842.2  examples/sec;  0.069  sec/batch)
step 72470, loss=0.77  (1860.0  examples/sec;  0.069  sec/batch)
step 72480, loss=0.66  (1910.5  examples/sec;  0.067  sec/batch)
step 72490, loss=0.51  (1900.6  examples/sec;  0.067  sec/batch)
step 72500, loss=0.61  (1930.2  examples/sec;  0.066  sec/batch)
step 72510, loss=0.49  (1886.9  examples/sec;  0.068  sec/batch)
step 72520, loss=0.63  (1892.5  examples/sec;  0.068  sec/batch)
step 72530, loss=0.59  (1880.7  examples/sec;  0.068  sec/batch)
step 72540, loss=0.61  (1949.6  examples/sec;  0.066  sec/batch)
step 72550, loss=0.54  (1926.1  examples/sec;  0.066  sec/batch)
step 72560, loss=0.51  (1855.5  examples/sec;  0.069  sec/batch)
step 72570, loss=0.54  (2059.8  examples/sec;  0.062  sec/batch)
step 72580, loss=0.58  (1846.1  examples/sec;  0.069  sec/batch)
step 72590, loss=0.47  (1818.6  examples/sec;  0.070  sec/batch)
step 72600, loss=0.69  (1893.7  examples/sec;  0.068  sec/batch)
step 72610, loss=0.69  (1830.4  examples/sec;  0.070  sec/batch)
step 72620, loss=0.65  (1915.6  examples/sec;  0.067  sec/batch)
step 72630, loss=0.59  (1836.4  examples/sec;  0.070  sec/batch)
step 72640, loss=0.63  (1991.6  examples/sec;  0.064  sec/batch)
step 72650, loss=0.62  (1947.4  examples/sec;  0.066  sec/batch)
step 72660, loss=0.82  (1701.2  examples/sec;  0.075  sec/batch)
step 72670, loss=0.64  (1912.6  examples/sec;  0.067  sec/batch)
step 72680, loss=0.62  (1897.3  examples/sec;  0.067  sec/batch)
step 72690, loss=0.73  (1876.7  examples/sec;  0.068  sec/batch)
step 72700, loss=0.78  (1923.2  examples/sec;  0.067  sec/batch)
step 72710, loss=0.63  (1993.9  examples/sec;  0.064  sec/batch)
step 72720, loss=0.52  (1828.8  examples/sec;  0.070  sec/batch)
step 72730, loss=0.53  (1918.5  examples/sec;  0.067  sec/batch)
step 72740, loss=0.75  (1927.0  examples/sec;  0.066  sec/batch)
step 72750, loss=0.62  (1942.4  examples/sec;  0.066  sec/batch)
step 72760, loss=0.63  (1914.3  examples/sec;  0.067  sec/batch)
step 72770, loss=0.64  (2008.4  examples/sec;  0.064  sec/batch)
step 72780, loss=0.62  (1648.0  examples/sec;  0.078  sec/batch)
step 72790, loss=0.53  (1894.6  examples/sec;  0.068  sec/batch)
step 72800, loss=0.62  (1860.8  examples/sec;  0.069  sec/batch)
step 72810, loss=0.71  (1827.8  examples/sec;  0.070  sec/batch)
step 72820, loss=0.83  (1870.0  examples/sec;  0.068  sec/batch)
step 72830, loss=0.61  (1892.7  examples/sec;  0.068  sec/batch)
step 72840, loss=0.68  (1990.9  examples/sec;  0.064  sec/batch)
step 72850, loss=0.64  (1836.4  examples/sec;  0.070  sec/batch)
step 72860, loss=0.81  (1937.9  examples/sec;  0.066  sec/batch)
step 72870, loss=0.49  (1884.8  examples/sec;  0.068  sec/batch)
step 72880, loss=0.63  (1951.5  examples/sec;  0.066  sec/batch)
step 72890, loss=0.56  (1965.7  examples/sec;  0.065  sec/batch)
step 72900, loss=0.59  (1904.0  examples/sec;  0.067  sec/batch)
step 72910, loss=0.73  (1926.1  examples/sec;  0.066  sec/batch)
step 72920, loss=0.58  (1953.9  examples/sec;  0.066  sec/batch)
step 72930, loss=0.64  (1885.3  examples/sec;  0.068  sec/batch)
step 72940, loss=0.53  (1822.5  examples/sec;  0.070  sec/batch)
step 72950, loss=0.48  (1811.3  examples/sec;  0.071  sec/batch)
step 72960, loss=0.62  (1822.1  examples/sec;  0.070  sec/batch)
step 72970, loss=0.62  (1944.6  examples/sec;  0.066  sec/batch)
step 72980, loss=0.66  (1944.8  examples/sec;  0.066  sec/batch)
step 72990, loss=0.54  (1944.9  examples/sec;  0.066  sec/batch)
step 73000, loss=0.44  (1914.0  examples/sec;  0.067  sec/batch)
step 73010, loss=0.57  (1870.0  examples/sec;  0.068  sec/batch)
step 73020, loss=0.73  (1819.3  examples/sec;  0.070  sec/batch)
step 73030, loss=0.73  (1848.5  examples/sec;  0.069  sec/batch)
step 73040, loss=0.63  (2012.9  examples/sec;  0.064  sec/batch)
step 73050, loss=0.56  (1951.7  examples/sec;  0.066  sec/batch)
step 73060, loss=0.52  (1860.9  examples/sec;  0.069  sec/batch)
step 73070, loss=0.59  (1875.4  examples/sec;  0.068  sec/batch)
step 73080, loss=0.61  (1877.8  examples/sec;  0.068  sec/batch)
step 73090, loss=0.65  (1871.8  examples/sec;  0.068  sec/batch)
step 73100, loss=0.65  (1897.8  examples/sec;  0.067  sec/batch)
step 73110, loss=0.67  (1908.1  examples/sec;  0.067  sec/batch)
step 73120, loss=0.64  (1907.2  examples/sec;  0.067  sec/batch)
step 73130, loss=0.57  (1993.2  examples/sec;  0.064  sec/batch)
step 73140, loss=0.47  (1847.1  examples/sec;  0.069  sec/batch)
step 73150, loss=0.54  (1891.9  examples/sec;  0.068  sec/batch)
step 73160, loss=0.62  (1835.3  examples/sec;  0.070  sec/batch)
step 73170, loss=0.78  (1939.2  examples/sec;  0.066  sec/batch)
step 73180, loss=0.51  (1852.9  examples/sec;  0.069  sec/batch)
step 73190, loss=0.65  (2033.5  examples/sec;  0.063  sec/batch)
step 73200, loss=0.73  (1875.8  examples/sec;  0.068  sec/batch)
step 73210, loss=0.59  (1900.2  examples/sec;  0.067  sec/batch)
step 73220, loss=0.66  (1873.6  examples/sec;  0.068  sec/batch)
step 73230, loss=0.66  (1877.4  examples/sec;  0.068  sec/batch)
step 73240, loss=0.62  (1690.7  examples/sec;  0.076  sec/batch)
step 73250, loss=0.70  (1897.6  examples/sec;  0.067  sec/batch)
step 73260, loss=0.67  (1900.2  examples/sec;  0.067  sec/batch)
step 73270, loss=0.53  (1900.3  examples/sec;  0.067  sec/batch)
step 73280, loss=0.60  (1894.7  examples/sec;  0.068  sec/batch)
step 73290, loss=0.70  (1951.8  examples/sec;  0.066  sec/batch)
step 73300, loss=0.68  (1904.6  examples/sec;  0.067  sec/batch)
step 73310, loss=0.68  (1893.3  examples/sec;  0.068  sec/batch)
step 73320, loss=0.78  (1861.3  examples/sec;  0.069  sec/batch)
step 73330, loss=0.51  (1917.0  examples/sec;  0.067  sec/batch)
step 73340, loss=0.56  (1904.3  examples/sec;  0.067  sec/batch)
step 73350, loss=0.54  (1874.1  examples/sec;  0.068  sec/batch)
step 73360, loss=0.57  (1950.1  examples/sec;  0.066  sec/batch)
step 73370, loss=0.62  (1485.6  examples/sec;  0.086  sec/batch)
step 73380, loss=0.62  (1944.4  examples/sec;  0.066  sec/batch)
step 73390, loss=0.57  (1953.1  examples/sec;  0.066  sec/batch)
step 73400, loss=0.88  (1910.1  examples/sec;  0.067  sec/batch)
step 73410, loss=0.77  (1968.7  examples/sec;  0.065  sec/batch)
step 73420, loss=0.59  (1859.9  examples/sec;  0.069  sec/batch)
step 73430, loss=0.64  (1910.1  examples/sec;  0.067  sec/batch)
step 73440, loss=0.52  (1850.9  examples/sec;  0.069  sec/batch)
step 73450, loss=0.57  (1990.2  examples/sec;  0.064  sec/batch)
step 73460, loss=0.54  (1713.3  examples/sec;  0.075  sec/batch)
step 73470, loss=0.55  (1930.6  examples/sec;  0.066  sec/batch)
step 73480, loss=0.52  (1845.4  examples/sec;  0.069  sec/batch)
step 73490, loss=0.59  (1871.0  examples/sec;  0.068  sec/batch)
step 73500, loss=0.48  (1700.7  examples/sec;  0.075  sec/batch)
step 73510, loss=0.56  (1957.6  examples/sec;  0.065  sec/batch)
step 73520, loss=0.59  (1904.2  examples/sec;  0.067  sec/batch)
step 73530, loss=0.72  (1868.5  examples/sec;  0.069  sec/batch)
step 73540, loss=0.57  (1924.7  examples/sec;  0.067  sec/batch)
step 73550, loss=0.55  (1985.9  examples/sec;  0.064  sec/batch)
step 73560, loss=0.45  (1867.5  examples/sec;  0.069  sec/batch)
step 73570, loss=0.53  (1878.2  examples/sec;  0.068  sec/batch)
step 73580, loss=0.66  (1929.9  examples/sec;  0.066  sec/batch)
step 73590, loss=0.72  (1957.2  examples/sec;  0.065  sec/batch)
step 73600, loss=0.64  (1902.6  examples/sec;  0.067  sec/batch)
step 73610, loss=0.63  (1855.4  examples/sec;  0.069  sec/batch)
step 73620, loss=0.52  (1927.7  examples/sec;  0.066  sec/batch)
step 73630, loss=0.68  (1862.0  examples/sec;  0.069  sec/batch)
step 73640, loss=0.59  (1880.2  examples/sec;  0.068  sec/batch)
step 73650, loss=0.66  (1852.8  examples/sec;  0.069  sec/batch)
step 73660, loss=0.57  (1961.0  examples/sec;  0.065  sec/batch)
step 73670, loss=0.59  (1891.3  examples/sec;  0.068  sec/batch)
step 73680, loss=0.68  (1938.1  examples/sec;  0.066  sec/batch)
step 73690, loss=0.49  (1773.4  examples/sec;  0.072  sec/batch)
step 73700, loss=0.74  (2056.7  examples/sec;  0.062  sec/batch)
step 73710, loss=0.46  (1945.7  examples/sec;  0.066  sec/batch)
step 73720, loss=0.64  (1629.1  examples/sec;  0.079  sec/batch)
step 73730, loss=0.74  (1990.1  examples/sec;  0.064  sec/batch)
step 73740, loss=0.58  (1994.2  examples/sec;  0.064  sec/batch)
step 73750, loss=0.59  (1902.9  examples/sec;  0.067  sec/batch)
step 73760, loss=0.66  (1913.5  examples/sec;  0.067  sec/batch)
step 73770, loss=0.54  (1897.0  examples/sec;  0.067  sec/batch)
step 73780, loss=0.57  (1929.3  examples/sec;  0.066  sec/batch)
step 73790, loss=0.62  (1882.6  examples/sec;  0.068  sec/batch)
step 73800, loss=0.56  (1978.7  examples/sec;  0.065  sec/batch)
step 73810, loss=0.64  (1898.6  examples/sec;  0.067  sec/batch)
step 73820, loss=0.65  (1844.9  examples/sec;  0.069  sec/batch)
step 73830, loss=0.68  (1935.2  examples/sec;  0.066  sec/batch)
step 73840, loss=0.52  (2014.4  examples/sec;  0.064  sec/batch)
step 73850, loss=0.70  (1932.2  examples/sec;  0.066  sec/batch)
step 73860, loss=0.69  (1879.3  examples/sec;  0.068  sec/batch)
step 73870, loss=0.52  (1847.2  examples/sec;  0.069  sec/batch)
step 73880, loss=0.73  (1916.3  examples/sec;  0.067  sec/batch)
step 73890, loss=0.72  (1939.0  examples/sec;  0.066  sec/batch)
step 73900, loss=0.62  (1907.6  examples/sec;  0.067  sec/batch)
step 73910, loss=0.58  (1844.4  examples/sec;  0.069  sec/batch)
step 73920, loss=0.65  (1696.8  examples/sec;  0.075  sec/batch)
step 73930, loss=0.55  (1899.2  examples/sec;  0.067  sec/batch)
step 73940, loss=0.59  (1853.3  examples/sec;  0.069  sec/batch)
step 73950, loss=0.68  (1957.9  examples/sec;  0.065  sec/batch)
step 73960, loss=0.72  (1858.7  examples/sec;  0.069  sec/batch)
step 73970, loss=0.69  (1852.7  examples/sec;  0.069  sec/batch)
step 73980, loss=0.54  (1805.6  examples/sec;  0.071  sec/batch)
step 73990, loss=0.65  (1905.9  examples/sec;  0.067  sec/batch)
step 74000, loss=0.59  (1899.9  examples/sec;  0.067  sec/batch)
step 74010, loss=0.44  (1822.5  examples/sec;  0.070  sec/batch)
step 74020, loss=0.42  (1904.7  examples/sec;  0.067  sec/batch)
step 74030, loss=0.65  (1948.1  examples/sec;  0.066  sec/batch)
step 74040, loss=0.42  (1854.7  examples/sec;  0.069  sec/batch)
step 74050, loss=0.49  (1820.4  examples/sec;  0.070  sec/batch)
step 74060, loss=0.52  (1862.8  examples/sec;  0.069  sec/batch)
step 74070, loss=0.64  (1880.4  examples/sec;  0.068  sec/batch)
step 74080, loss=0.63  (1981.2  examples/sec;  0.065  sec/batch)
step 74090, loss=0.55  (1767.7  examples/sec;  0.072  sec/batch)
step 74100, loss=0.61  (1927.0  examples/sec;  0.066  sec/batch)
step 74110, loss=0.73  (1923.4  examples/sec;  0.067  sec/batch)
step 74120, loss=0.68  (1970.3  examples/sec;  0.065  sec/batch)
step 74130, loss=0.67  (1860.9  examples/sec;  0.069  sec/batch)
step 74140, loss=0.54  (1930.9  examples/sec;  0.066  sec/batch)
step 74150, loss=0.66  (1867.8  examples/sec;  0.069  sec/batch)
step 74160, loss=0.61  (1843.1  examples/sec;  0.069  sec/batch)
step 74170, loss=0.58  (1843.5  examples/sec;  0.069  sec/batch)
step 74180, loss=0.83  (1865.2  examples/sec;  0.069  sec/batch)
step 74190, loss=0.57  (1877.8  examples/sec;  0.068  sec/batch)
step 74200, loss=0.57  (1879.9  examples/sec;  0.068  sec/batch)
step 74210, loss=0.68  (1926.7  examples/sec;  0.066  sec/batch)
step 74220, loss=0.55  (1838.7  examples/sec;  0.070  sec/batch)
step 74230, loss=0.60  (1876.4  examples/sec;  0.068  sec/batch)
step 74240, loss=0.65  (1964.4  examples/sec;  0.065  sec/batch)
step 74250, loss=0.65  (1920.9  examples/sec;  0.067  sec/batch)
step 74260, loss=0.63  (1867.1  examples/sec;  0.069  sec/batch)
step 74270, loss=0.71  (1936.7  examples/sec;  0.066  sec/batch)
step 74280, loss=0.60  (1903.2  examples/sec;  0.067  sec/batch)
step 74290, loss=0.57  (1906.5  examples/sec;  0.067  sec/batch)
step 74300, loss=0.55  (1963.2  examples/sec;  0.065  sec/batch)
step 74310, loss=0.49  (1832.9  examples/sec;  0.070  sec/batch)
step 74320, loss=0.52  (1893.7  examples/sec;  0.068  sec/batch)
step 74330, loss=0.67  (1904.1  examples/sec;  0.067  sec/batch)
step 74340, loss=0.70  (1911.1  examples/sec;  0.067  sec/batch)
step 74350, loss=0.58  (1936.5  examples/sec;  0.066  sec/batch)
step 74360, loss=0.53  (1865.3  examples/sec;  0.069  sec/batch)
step 74370, loss=0.68  (1925.5  examples/sec;  0.066  sec/batch)
step 74380, loss=0.54  (1864.0  examples/sec;  0.069  sec/batch)
step 74390, loss=0.76  (1917.6  examples/sec;  0.067  sec/batch)
step 74400, loss=0.58  (1927.2  examples/sec;  0.066  sec/batch)
step 74410, loss=0.53  (1887.3  examples/sec;  0.068  sec/batch)
step 74420, loss=0.74  (1818.6  examples/sec;  0.070  sec/batch)
step 74430, loss=0.62  (1838.0  examples/sec;  0.070  sec/batch)
step 74440, loss=0.66  (1917.8  examples/sec;  0.067  sec/batch)
step 74450, loss=0.76  (1922.0  examples/sec;  0.067  sec/batch)
step 74460, loss=0.54  (1908.4  examples/sec;  0.067  sec/batch)
step 74470, loss=0.57  (1850.3  examples/sec;  0.069  sec/batch)
step 74480, loss=0.86  (1853.3  examples/sec;  0.069  sec/batch)
step 74490, loss=0.56  (1718.6  examples/sec;  0.074  sec/batch)
step 74500, loss=0.64  (1973.2  examples/sec;  0.065  sec/batch)
step 74510, loss=0.70  (1906.7  examples/sec;  0.067  sec/batch)
step 74520, loss=0.60  (1888.1  examples/sec;  0.068  sec/batch)
step 74530, loss=0.78  (1933.5  examples/sec;  0.066  sec/batch)
step 74540, loss=0.68  (1906.1  examples/sec;  0.067  sec/batch)
step 74550, loss=0.55  (1949.7  examples/sec;  0.066  sec/batch)
step 74560, loss=0.64  (1886.9  examples/sec;  0.068  sec/batch)
step 74570, loss=0.57  (1914.4  examples/sec;  0.067  sec/batch)
step 74580, loss=0.54  (1953.3  examples/sec;  0.066  sec/batch)
step 74590, loss=0.63  (1928.8  examples/sec;  0.066  sec/batch)
step 74600, loss=0.71  (1733.4  examples/sec;  0.074  sec/batch)
step 74610, loss=0.63  (1786.6  examples/sec;  0.072  sec/batch)
step 74620, loss=0.48  (1818.8  examples/sec;  0.070  sec/batch)
step 74630, loss=0.61  (1911.1  examples/sec;  0.067  sec/batch)
step 74640, loss=0.54  (1757.0  examples/sec;  0.073  sec/batch)
step 74650, loss=0.70  (1888.4  examples/sec;  0.068  sec/batch)
step 74660, loss=0.59  (1753.0  examples/sec;  0.073  sec/batch)
step 74670, loss=0.61  (1808.6  examples/sec;  0.071  sec/batch)
step 74680, loss=0.47  (1834.2  examples/sec;  0.070  sec/batch)
step 74690, loss=0.69  (1909.7  examples/sec;  0.067  sec/batch)
step 74700, loss=0.50  (1915.8  examples/sec;  0.067  sec/batch)
step 74710, loss=0.62  (1929.0  examples/sec;  0.066  sec/batch)
step 74720, loss=0.87  (1860.4  examples/sec;  0.069  sec/batch)
step 74730, loss=0.63  (1893.2  examples/sec;  0.068  sec/batch)
step 74740, loss=0.75  (1846.6  examples/sec;  0.069  sec/batch)
step 74750, loss=0.51  (1818.4  examples/sec;  0.070  sec/batch)
step 74760, loss=0.62  (1920.2  examples/sec;  0.067  sec/batch)
step 74770, loss=0.64  (1884.7  examples/sec;  0.068  sec/batch)
step 74780, loss=0.63  (1788.8  examples/sec;  0.072  sec/batch)
step 74790, loss=0.72  (1908.3  examples/sec;  0.067  sec/batch)
step 74800, loss=0.64  (1639.0  examples/sec;  0.078  sec/batch)
step 74810, loss=0.73  (1898.2  examples/sec;  0.067  sec/batch)
step 74820, loss=0.58  (1927.1  examples/sec;  0.066  sec/batch)
step 74830, loss=0.60  (1852.1  examples/sec;  0.069  sec/batch)
step 74840, loss=0.67  (1897.5  examples/sec;  0.067  sec/batch)
step 74850, loss=0.74  (1836.7  examples/sec;  0.070  sec/batch)
step 74860, loss=0.66  (1927.2  examples/sec;  0.066  sec/batch)
step 74870, loss=0.70  (1919.7  examples/sec;  0.067  sec/batch)
step 74880, loss=0.52  (1891.0  examples/sec;  0.068  sec/batch)
step 74890, loss=0.58  (1841.1  examples/sec;  0.070  sec/batch)
step 74900, loss=0.69  (2026.9  examples/sec;  0.063  sec/batch)
step 74910, loss=0.70  (1851.1  examples/sec;  0.069  sec/batch)
step 74920, loss=0.68  (1875.2  examples/sec;  0.068  sec/batch)
step 74930, loss=0.64  (1886.0  examples/sec;  0.068  sec/batch)
step 74940, loss=0.78  (1968.4  examples/sec;  0.065  sec/batch)
step 74950, loss=0.65  (1925.3  examples/sec;  0.066  sec/batch)
step 74960, loss=0.55  (1657.3  examples/sec;  0.077  sec/batch)
step 74970, loss=0.58  (1858.4  examples/sec;  0.069  sec/batch)
step 74980, loss=0.61  (1857.7  examples/sec;  0.069  sec/batch)
step 74990, loss=0.63  (1879.7  examples/sec;  0.068  sec/batch)
step 75000, loss=0.79  (1908.3  examples/sec;  0.067  sec/batch)
step 75010, loss=0.69  (1902.7  examples/sec;  0.067  sec/batch)
step 75020, loss=0.81  (1741.1  examples/sec;  0.074  sec/batch)
step 75030, loss=0.56  (1848.3  examples/sec;  0.069  sec/batch)
step 75040, loss=0.63  (1853.8  examples/sec;  0.069  sec/batch)
step 75050, loss=0.56  (1925.0  examples/sec;  0.066  sec/batch)
step 75060, loss=0.54  (1762.8  examples/sec;  0.073  sec/batch)
step 75070, loss=0.57  (1948.1  examples/sec;  0.066  sec/batch)
step 75080, loss=0.75  (1824.8  examples/sec;  0.070  sec/batch)
step 75090, loss=0.51  (1739.9  examples/sec;  0.074  sec/batch)
step 75100, loss=0.64  (1876.6  examples/sec;  0.068  sec/batch)
step 75110, loss=0.51  (1856.8  examples/sec;  0.069  sec/batch)
step 75120, loss=0.70  (1778.3  examples/sec;  0.072  sec/batch)
step 75130, loss=0.68  (1884.0  examples/sec;  0.068  sec/batch)
step 75140, loss=0.58  (1802.3  examples/sec;  0.071  sec/batch)
step 75150, loss=0.68  (1873.1  examples/sec;  0.068  sec/batch)
step 75160, loss=0.51  (1825.1  examples/sec;  0.070  sec/batch)
step 75170, loss=0.61  (1788.2  examples/sec;  0.072  sec/batch)
step 75180, loss=0.57  (1857.3  examples/sec;  0.069  sec/batch)
step 75190, loss=0.60  (1938.3  examples/sec;  0.066  sec/batch)
step 75200, loss=0.60  (1895.1  examples/sec;  0.068  sec/batch)
step 75210, loss=0.57  (1635.2  examples/sec;  0.078  sec/batch)
step 75220, loss=0.63  (1880.1  examples/sec;  0.068  sec/batch)
step 75230, loss=0.60  (1887.9  examples/sec;  0.068  sec/batch)
step 75240, loss=0.64  (1869.0  examples/sec;  0.068  sec/batch)
step 75250, loss=0.50  (1918.1  examples/sec;  0.067  sec/batch)
step 75260, loss=0.69  (1686.3  examples/sec;  0.076  sec/batch)
step 75270, loss=0.58  (1763.3  examples/sec;  0.073  sec/batch)
step 75280, loss=0.74  (1917.9  examples/sec;  0.067  sec/batch)
step 75290, loss=0.72  (1900.6  examples/sec;  0.067  sec/batch)
step 75300, loss=0.66  (1849.7  examples/sec;  0.069  sec/batch)
step 75310, loss=0.51  (1848.8  examples/sec;  0.069  sec/batch)
step 75320, loss=0.50  (1980.1  examples/sec;  0.065  sec/batch)
step 75330, loss=0.58  (1871.5  examples/sec;  0.068  sec/batch)
step 75340, loss=0.65  (1899.3  examples/sec;  0.067  sec/batch)
step 75350, loss=0.56  (1941.0  examples/sec;  0.066  sec/batch)
step 75360, loss=0.57  (1930.1  examples/sec;  0.066  sec/batch)
step 75370, loss=0.59  (1914.1  examples/sec;  0.067  sec/batch)
step 75380, loss=0.61  (1831.5  examples/sec;  0.070  sec/batch)
step 75390, loss=0.59  (1966.7  examples/sec;  0.065  sec/batch)
step 75400, loss=0.64  (1643.1  examples/sec;  0.078  sec/batch)
step 75410, loss=0.75  (1957.6  examples/sec;  0.065  sec/batch)
step 75420, loss=0.55  (1961.1  examples/sec;  0.065  sec/batch)
step 75430, loss=0.59  (1884.3  examples/sec;  0.068  sec/batch)
step 75440, loss=0.62  (1685.5  examples/sec;  0.076  sec/batch)
step 75450, loss=0.47  (1946.3  examples/sec;  0.066  sec/batch)
step 75460, loss=0.70  (1889.2  examples/sec;  0.068  sec/batch)
step 75470, loss=0.50  (1765.0  examples/sec;  0.073  sec/batch)
step 75480, loss=0.68  (1915.5  examples/sec;  0.067  sec/batch)
step 75490, loss=0.64  (1796.1  examples/sec;  0.071  sec/batch)
step 75500, loss=0.42  (1788.9  examples/sec;  0.072  sec/batch)
step 75510, loss=0.60  (1888.7  examples/sec;  0.068  sec/batch)
step 75520, loss=0.49  (1908.8  examples/sec;  0.067  sec/batch)
step 75530, loss=0.72  (1839.9  examples/sec;  0.070  sec/batch)
step 75540, loss=0.71  (1869.1  examples/sec;  0.068  sec/batch)
step 75550, loss=0.58  (1938.7  examples/sec;  0.066  sec/batch)
step 75560, loss=0.65  (1965.9  examples/sec;  0.065  sec/batch)
step 75570, loss=0.75  (1968.6  examples/sec;  0.065  sec/batch)
step 75580, loss=0.63  (1905.5  examples/sec;  0.067  sec/batch)
step 75590, loss=0.62  (1849.6  examples/sec;  0.069  sec/batch)
step 75600, loss=0.78  (1860.5  examples/sec;  0.069  sec/batch)
step 75610, loss=0.65  (1904.6  examples/sec;  0.067  sec/batch)
step 75620, loss=0.76  (1985.4  examples/sec;  0.064  sec/batch)
step 75630, loss=0.60  (1931.6  examples/sec;  0.066  sec/batch)
step 75640, loss=0.71  (1856.6  examples/sec;  0.069  sec/batch)
step 75650, loss=0.58  (1921.0  examples/sec;  0.067  sec/batch)
step 75660, loss=0.68  (1847.7  examples/sec;  0.069  sec/batch)
step 75670, loss=0.61  (1945.1  examples/sec;  0.066  sec/batch)
step 75680, loss=0.62  (1878.4  examples/sec;  0.068  sec/batch)
step 75690, loss=0.54  (1841.3  examples/sec;  0.070  sec/batch)
step 75700, loss=0.66  (1906.3  examples/sec;  0.067  sec/batch)
step 75710, loss=0.56  (1921.1  examples/sec;  0.067  sec/batch)
step 75720, loss=0.64  (1903.4  examples/sec;  0.067  sec/batch)
step 75730, loss=0.41  (1930.0  examples/sec;  0.066  sec/batch)
step 75740, loss=0.47  (1883.7  examples/sec;  0.068  sec/batch)
step 75750, loss=0.62  (1913.2  examples/sec;  0.067  sec/batch)
step 75760, loss=0.66  (1853.4  examples/sec;  0.069  sec/batch)
step 75770, loss=0.57  (1899.1  examples/sec;  0.067  sec/batch)
step 75780, loss=0.54  (1909.9  examples/sec;  0.067  sec/batch)
step 75790, loss=0.57  (1906.8  examples/sec;  0.067  sec/batch)
step 75800, loss=0.44  (1818.2  examples/sec;  0.070  sec/batch)
step 75810, loss=0.65  (1898.7  examples/sec;  0.067  sec/batch)
step 75820, loss=0.69  (1946.6  examples/sec;  0.066  sec/batch)
step 75830, loss=0.58  (1944.9  examples/sec;  0.066  sec/batch)
step 75840, loss=0.49  (1868.6  examples/sec;  0.069  sec/batch)
step 75850, loss=0.46  (1954.1  examples/sec;  0.066  sec/batch)
step 75860, loss=0.50  (1939.1  examples/sec;  0.066  sec/batch)
step 75870, loss=0.61  (1927.0  examples/sec;  0.066  sec/batch)
step 75880, loss=0.75  (1922.2  examples/sec;  0.067  sec/batch)
step 75890, loss=0.56  (2015.6  examples/sec;  0.064  sec/batch)
step 75900, loss=0.54  (1892.9  examples/sec;  0.068  sec/batch)
step 75910, loss=0.64  (1982.4  examples/sec;  0.065  sec/batch)
step 75920, loss=0.55  (1911.9  examples/sec;  0.067  sec/batch)
step 75930, loss=0.77  (1922.1  examples/sec;  0.067  sec/batch)
step 75940, loss=0.53  (1845.7  examples/sec;  0.069  sec/batch)
step 75950, loss=0.65  (1838.9  examples/sec;  0.070  sec/batch)
step 75960, loss=0.54  (1885.8  examples/sec;  0.068  sec/batch)
step 75970, loss=0.61  (1976.2  examples/sec;  0.065  sec/batch)
step 75980, loss=0.66  (1816.1  examples/sec;  0.070  sec/batch)
step 75990, loss=0.60  (1980.4  examples/sec;  0.065  sec/batch)
step 76000, loss=0.47  (1824.7  examples/sec;  0.070  sec/batch)
step 76010, loss=0.65  (1862.1  examples/sec;  0.069  sec/batch)
step 76020, loss=0.63  (1768.2  examples/sec;  0.072  sec/batch)
step 76030, loss=0.59  (1946.9  examples/sec;  0.066  sec/batch)
step 76040, loss=0.52  (1847.5  examples/sec;  0.069  sec/batch)
step 76050, loss=0.47  (2011.2  examples/sec;  0.064  sec/batch)
step 76060, loss=0.52  (1894.1  examples/sec;  0.068  sec/batch)
step 76070, loss=0.66  (1898.3  examples/sec;  0.067  sec/batch)
step 76080, loss=0.67  (1874.2  examples/sec;  0.068  sec/batch)
step 76090, loss=0.75  (1937.7  examples/sec;  0.066  sec/batch)
step 76100, loss=0.52  (1895.5  examples/sec;  0.068  sec/batch)
step 76110, loss=0.57  (1874.7  examples/sec;  0.068  sec/batch)
step 76120, loss=0.60  (1952.4  examples/sec;  0.066  sec/batch)
step 76130, loss=0.59  (1855.0  examples/sec;  0.069  sec/batch)
step 76140, loss=0.60  (1922.1  examples/sec;  0.067  sec/batch)
step 76150, loss=0.67  (1928.1  examples/sec;  0.066  sec/batch)
step 76160, loss=0.45  (1698.2  examples/sec;  0.075  sec/batch)
step 76170, loss=0.53  (1979.7  examples/sec;  0.065  sec/batch)
step 76180, loss=0.75  (1813.9  examples/sec;  0.071  sec/batch)
step 76190, loss=0.74  (1922.8  examples/sec;  0.067  sec/batch)
step 76200, loss=0.69  (1955.4  examples/sec;  0.065  sec/batch)
step 76210, loss=0.57  (1947.7  examples/sec;  0.066  sec/batch)
step 76220, loss=0.61  (1937.3  examples/sec;  0.066  sec/batch)
step 76230, loss=0.60  (1926.4  examples/sec;  0.066  sec/batch)
step 76240, loss=0.68  (1770.0  examples/sec;  0.072  sec/batch)
step 76250, loss=0.73  (1939.3  examples/sec;  0.066  sec/batch)
step 76260, loss=0.43  (1872.2  examples/sec;  0.068  sec/batch)
step 76270, loss=0.62  (1878.0  examples/sec;  0.068  sec/batch)
step 76280, loss=0.84  (1961.5  examples/sec;  0.065  sec/batch)
step 76290, loss=0.75  (1871.3  examples/sec;  0.068  sec/batch)
step 76300, loss=0.58  (1940.1  examples/sec;  0.066  sec/batch)
step 76310, loss=0.68  (1931.5  examples/sec;  0.066  sec/batch)
step 76320, loss=0.53  (2004.4  examples/sec;  0.064  sec/batch)
step 76330, loss=0.74  (1879.1  examples/sec;  0.068  sec/batch)
step 76340, loss=0.42  (1865.9  examples/sec;  0.069  sec/batch)
step 76350, loss=0.63  (1950.3  examples/sec;  0.066  sec/batch)
step 76360, loss=0.49  (1943.1  examples/sec;  0.066  sec/batch)
step 76370, loss=0.61  (1914.0  examples/sec;  0.067  sec/batch)
step 76380, loss=0.42  (1867.0  examples/sec;  0.069  sec/batch)
step 76390, loss=0.59  (1854.0  examples/sec;  0.069  sec/batch)
step 76400, loss=0.58  (1808.4  examples/sec;  0.071  sec/batch)
step 76410, loss=0.47  (1900.8  examples/sec;  0.067  sec/batch)
step 76420, loss=0.60  (1862.3  examples/sec;  0.069  sec/batch)
step 76430, loss=0.69  (1909.6  examples/sec;  0.067  sec/batch)
step 76440, loss=0.59  (1882.2  examples/sec;  0.068  sec/batch)
step 76450, loss=0.47  (1831.9  examples/sec;  0.070  sec/batch)
step 76460, loss=0.52  (1889.7  examples/sec;  0.068  sec/batch)
step 76470, loss=0.82  (1921.1  examples/sec;  0.067  sec/batch)
step 76480, loss=0.60  (1901.8  examples/sec;  0.067  sec/batch)
step 76490, loss=0.68  (1936.5  examples/sec;  0.066  sec/batch)
step 76500, loss=0.57  (1822.0  examples/sec;  0.070  sec/batch)
step 76510, loss=0.56  (1886.8  examples/sec;  0.068  sec/batch)
step 76520, loss=0.60  (1937.3  examples/sec;  0.066  sec/batch)
step 76530, loss=0.64  (1952.1  examples/sec;  0.066  sec/batch)
step 76540, loss=0.65  (1946.9  examples/sec;  0.066  sec/batch)
step 76550, loss=0.53  (1841.1  examples/sec;  0.070  sec/batch)
step 76560, loss=0.55  (1903.1  examples/sec;  0.067  sec/batch)
step 76570, loss=0.63  (1735.6  examples/sec;  0.074  sec/batch)
step 76580, loss=0.51  (1835.2  examples/sec;  0.070  sec/batch)
step 76590, loss=0.80  (1911.0  examples/sec;  0.067  sec/batch)
step 76600, loss=0.57  (1826.4  examples/sec;  0.070  sec/batch)
step 76610, loss=0.55  (1887.9  examples/sec;  0.068  sec/batch)
step 76620, loss=0.50  (2005.5  examples/sec;  0.064  sec/batch)
step 76630, loss=0.64  (1880.7  examples/sec;  0.068  sec/batch)
step 76640, loss=0.70  (1853.2  examples/sec;  0.069  sec/batch)
step 76650, loss=0.51  (1920.4  examples/sec;  0.067  sec/batch)
step 76660, loss=0.51  (1867.4  examples/sec;  0.069  sec/batch)
step 76670, loss=0.71  (1988.3  examples/sec;  0.064  sec/batch)
step 76680, loss=0.57  (1950.7  examples/sec;  0.066  sec/batch)
step 76690, loss=0.56  (1950.7  examples/sec;  0.066  sec/batch)
step 76700, loss=0.65  (1670.9  examples/sec;  0.077  sec/batch)
step 76710, loss=0.45  (1764.4  examples/sec;  0.073  sec/batch)
step 76720, loss=0.61  (1866.8  examples/sec;  0.069  sec/batch)
step 76730, loss=0.65  (1863.9  examples/sec;  0.069  sec/batch)
step 76740, loss=0.62  (1937.4  examples/sec;  0.066  sec/batch)
step 76750, loss=0.61  (1816.9  examples/sec;  0.070  sec/batch)
step 76760, loss=0.59  (1937.7  examples/sec;  0.066  sec/batch)
step 76770, loss=0.64  (1844.8  examples/sec;  0.069  sec/batch)
step 76780, loss=0.61  (1828.5  examples/sec;  0.070  sec/batch)
step 76790, loss=0.51  (1896.7  examples/sec;  0.067  sec/batch)
step 76800, loss=0.64  (1869.1  examples/sec;  0.068  sec/batch)
step 76810, loss=0.53  (1998.3  examples/sec;  0.064  sec/batch)
step 76820, loss=0.64  (1804.8  examples/sec;  0.071  sec/batch)
step 76830, loss=0.69  (1845.9  examples/sec;  0.069  sec/batch)
step 76840, loss=0.70  (1858.9  examples/sec;  0.069  sec/batch)
step 76850, loss=0.50  (1871.4  examples/sec;  0.068  sec/batch)
step 76860, loss=0.45  (1793.2  examples/sec;  0.071  sec/batch)
step 76870, loss=0.57  (1964.3  examples/sec;  0.065  sec/batch)
step 76880, loss=0.65  (1821.6  examples/sec;  0.070  sec/batch)
step 76890, loss=0.52  (1830.1  examples/sec;  0.070  sec/batch)
step 76900, loss=0.53  (1876.7  examples/sec;  0.068  sec/batch)
step 76910, loss=0.48  (1881.0  examples/sec;  0.068  sec/batch)
step 76920, loss=0.47  (1804.2  examples/sec;  0.071  sec/batch)
step 76930, loss=0.59  (1966.0  examples/sec;  0.065  sec/batch)
step 76940, loss=0.57  (1787.2  examples/sec;  0.072  sec/batch)
step 76950, loss=0.68  (1934.6  examples/sec;  0.066  sec/batch)
step 76960, loss=0.60  (1913.0  examples/sec;  0.067  sec/batch)
step 76970, loss=0.54  (1955.4  examples/sec;  0.065  sec/batch)
step 76980, loss=0.54  (1888.6  examples/sec;  0.068  sec/batch)
step 76990, loss=0.65  (1893.2  examples/sec;  0.068  sec/batch)
step 77000, loss=0.77  (1846.1  examples/sec;  0.069  sec/batch)
step 77010, loss=0.56  (1817.6  examples/sec;  0.070  sec/batch)
step 77020, loss=0.73  (1896.0  examples/sec;  0.068  sec/batch)
step 77030, loss=0.60  (1946.0  examples/sec;  0.066  sec/batch)
step 77040, loss=0.57  (1939.7  examples/sec;  0.066  sec/batch)
step 77050, loss=0.50  (1906.1  examples/sec;  0.067  sec/batch)
step 77060, loss=0.37  (1933.7  examples/sec;  0.066  sec/batch)
step 77070, loss=0.51  (1859.1  examples/sec;  0.069  sec/batch)
step 77080, loss=0.64  (1883.9  examples/sec;  0.068  sec/batch)
step 77090, loss=0.67  (1931.9  examples/sec;  0.066  sec/batch)
step 77100, loss=0.63  (1918.8  examples/sec;  0.067  sec/batch)
step 77110, loss=0.44  (1996.3  examples/sec;  0.064  sec/batch)
step 77120, loss=0.57  (1883.1  examples/sec;  0.068  sec/batch)
step 77130, loss=0.56  (1898.1  examples/sec;  0.067  sec/batch)
step 77140, loss=0.77  (1849.4  examples/sec;  0.069  sec/batch)
step 77150, loss=0.62  (1854.2  examples/sec;  0.069  sec/batch)
step 77160, loss=0.50  (1858.7  examples/sec;  0.069  sec/batch)
step 77170, loss=0.69  (1901.7  examples/sec;  0.067  sec/batch)
step 77180, loss=0.56  (1908.0  examples/sec;  0.067  sec/batch)
step 77190, loss=0.64  (1774.9  examples/sec;  0.072  sec/batch)
step 77200, loss=0.68  (1927.1  examples/sec;  0.066  sec/batch)
step 77210, loss=0.72  (1864.6  examples/sec;  0.069  sec/batch)
step 77220, loss=0.61  (1897.4  examples/sec;  0.067  sec/batch)
step 77230, loss=0.52  (1913.0  examples/sec;  0.067  sec/batch)
step 77240, loss=0.57  (1876.0  examples/sec;  0.068  sec/batch)
step 77250, loss=0.56  (1946.4  examples/sec;  0.066  sec/batch)
step 77260, loss=0.56  (1828.6  examples/sec;  0.070  sec/batch)
step 77270, loss=0.64  (1846.1  examples/sec;  0.069  sec/batch)
step 77280, loss=0.53  (1879.6  examples/sec;  0.068  sec/batch)
step 77290, loss=0.42  (1854.8  examples/sec;  0.069  sec/batch)
step 77300, loss=0.78  (1892.9  examples/sec;  0.068  sec/batch)
step 77310, loss=0.53  (1825.2  examples/sec;  0.070  sec/batch)
step 77320, loss=0.70  (1890.2  examples/sec;  0.068  sec/batch)
step 77330, loss=0.57  (1918.1  examples/sec;  0.067  sec/batch)
step 77340, loss=0.52  (2001.7  examples/sec;  0.064  sec/batch)
step 77350, loss=0.52  (1792.0  examples/sec;  0.071  sec/batch)
step 77360, loss=0.59  (1841.0  examples/sec;  0.070  sec/batch)
step 77370, loss=0.84  (1937.7  examples/sec;  0.066  sec/batch)
step 77380, loss=0.70  (1883.2  examples/sec;  0.068  sec/batch)
step 77390, loss=0.70  (1938.7  examples/sec;  0.066  sec/batch)
step 77400, loss=0.58  (1950.2  examples/sec;  0.066  sec/batch)
step 77410, loss=0.63  (1891.2  examples/sec;  0.068  sec/batch)
step 77420, loss=0.57  (1831.7  examples/sec;  0.070  sec/batch)
step 77430, loss=0.57  (1551.6  examples/sec;  0.082  sec/batch)
step 77440, loss=0.61  (1902.9  examples/sec;  0.067  sec/batch)
step 77450, loss=0.55  (1835.0  examples/sec;  0.070  sec/batch)
step 77460, loss=0.67  (1921.5  examples/sec;  0.067  sec/batch)
step 77470, loss=0.58  (1883.3  examples/sec;  0.068  sec/batch)
step 77480, loss=0.73  (1872.4  examples/sec;  0.068  sec/batch)
step 77490, loss=0.63  (1977.6  examples/sec;  0.065  sec/batch)
step 77500, loss=0.54  (1786.0  examples/sec;  0.072  sec/batch)
step 77510, loss=0.62  (1889.5  examples/sec;  0.068  sec/batch)
step 77520, loss=0.55  (1875.6  examples/sec;  0.068  sec/batch)
step 77530, loss=0.68  (1785.0  examples/sec;  0.072  sec/batch)
step 77540, loss=0.59  (1929.3  examples/sec;  0.066  sec/batch)
step 77550, loss=0.61  (1942.7  examples/sec;  0.066  sec/batch)
step 77560, loss=0.63  (1900.3  examples/sec;  0.067  sec/batch)
step 77570, loss=0.69  (1887.1  examples/sec;  0.068  sec/batch)
step 77580, loss=0.57  (1919.6  examples/sec;  0.067  sec/batch)
step 77590, loss=0.58  (1894.7  examples/sec;  0.068  sec/batch)
step 77600, loss=0.54  (1854.5  examples/sec;  0.069  sec/batch)
step 77610, loss=0.47  (1958.4  examples/sec;  0.065  sec/batch)
step 77620, loss=0.60  (1908.8  examples/sec;  0.067  sec/batch)
step 77630, loss=0.57  (1877.0  examples/sec;  0.068  sec/batch)
step 77640, loss=0.48  (1959.2  examples/sec;  0.065  sec/batch)
step 77650, loss=0.55  (1877.7  examples/sec;  0.068  sec/batch)
step 77660, loss=0.55  (1842.1  examples/sec;  0.069  sec/batch)
step 77670, loss=0.69  (1963.8  examples/sec;  0.065  sec/batch)
step 77680, loss=0.69  (1990.9  examples/sec;  0.064  sec/batch)
step 77690, loss=0.60  (1895.3  examples/sec;  0.068  sec/batch)
step 77700, loss=0.60  (1864.7  examples/sec;  0.069  sec/batch)
step 77710, loss=0.57  (1886.4  examples/sec;  0.068  sec/batch)
step 77720, loss=0.71  (1967.3  examples/sec;  0.065  sec/batch)
step 77730, loss=0.60  (1899.3  examples/sec;  0.067  sec/batch)
step 77740, loss=0.76  (1942.0  examples/sec;  0.066  sec/batch)
step 77750, loss=0.62  (1850.0  examples/sec;  0.069  sec/batch)
step 77760, loss=0.79  (1858.0  examples/sec;  0.069  sec/batch)
step 77770, loss=0.71  (1892.1  examples/sec;  0.068  sec/batch)
step 77780, loss=0.57  (1824.7  examples/sec;  0.070  sec/batch)
step 77790, loss=0.70  (1917.3  examples/sec;  0.067  sec/batch)
step 77800, loss=0.64  (1962.0  examples/sec;  0.065  sec/batch)
step 77810, loss=0.69  (1859.4  examples/sec;  0.069  sec/batch)
step 77820, loss=0.54  (1781.7  examples/sec;  0.072  sec/batch)
step 77830, loss=0.51  (1761.0  examples/sec;  0.073  sec/batch)
step 77840, loss=0.61  (1857.5  examples/sec;  0.069  sec/batch)
step 77850, loss=0.67  (1909.7  examples/sec;  0.067  sec/batch)
step 77860, loss=0.58  (1886.1  examples/sec;  0.068  sec/batch)
step 77870, loss=0.60  (1864.9  examples/sec;  0.069  sec/batch)
step 77880, loss=0.72  (1901.2  examples/sec;  0.067  sec/batch)
step 77890, loss=0.51  (1873.6  examples/sec;  0.068  sec/batch)
step 77900, loss=0.57  (1915.4  examples/sec;  0.067  sec/batch)
step 77910, loss=0.53  (1910.4  examples/sec;  0.067  sec/batch)
step 77920, loss=0.71  (1956.6  examples/sec;  0.065  sec/batch)
step 77930, loss=0.72  (1903.0  examples/sec;  0.067  sec/batch)
step 77940, loss=0.50  (1879.6  examples/sec;  0.068  sec/batch)
step 77950, loss=0.49  (1913.3  examples/sec;  0.067  sec/batch)
step 77960, loss=0.71  (1930.1  examples/sec;  0.066  sec/batch)
step 77970, loss=0.69  (1949.6  examples/sec;  0.066  sec/batch)
step 77980, loss=0.67  (1895.4  examples/sec;  0.068  sec/batch)
step 77990, loss=0.70  (1870.6  examples/sec;  0.068  sec/batch)
step 78000, loss=0.77  (1868.7  examples/sec;  0.068  sec/batch)
step 78010, loss=0.54  (1805.0  examples/sec;  0.071  sec/batch)
step 78020, loss=0.60  (1909.3  examples/sec;  0.067  sec/batch)
step 78030, loss=0.67  (1956.9  examples/sec;  0.065  sec/batch)
step 78040, loss=0.62  (1976.8  examples/sec;  0.065  sec/batch)
step 78050, loss=0.50  (1897.5  examples/sec;  0.067  sec/batch)
step 78060, loss=0.59  (1861.2  examples/sec;  0.069  sec/batch)
step 78070, loss=0.61  (1887.0  examples/sec;  0.068  sec/batch)
step 78080, loss=0.61  (1901.2  examples/sec;  0.067  sec/batch)
step 78090, loss=0.54  (1918.0  examples/sec;  0.067  sec/batch)
step 78100, loss=0.63  (1870.2  examples/sec;  0.068  sec/batch)
step 78110, loss=0.60  (1904.3  examples/sec;  0.067  sec/batch)
step 78120, loss=0.53  (1923.3  examples/sec;  0.067  sec/batch)
step 78130, loss=0.72  (1847.4  examples/sec;  0.069  sec/batch)
step 78140, loss=0.80  (1894.1  examples/sec;  0.068  sec/batch)
step 78150, loss=0.67  (1855.4  examples/sec;  0.069  sec/batch)
step 78160, loss=0.65  (1891.1  examples/sec;  0.068  sec/batch)
step 78170, loss=0.56  (1848.5  examples/sec;  0.069  sec/batch)
step 78180, loss=0.54  (1797.9  examples/sec;  0.071  sec/batch)
step 78190, loss=0.60  (1930.5  examples/sec;  0.066  sec/batch)
step 78200, loss=0.70  (1880.2  examples/sec;  0.068  sec/batch)
step 78210, loss=0.57  (1852.6  examples/sec;  0.069  sec/batch)
step 78220, loss=0.54  (1896.6  examples/sec;  0.067  sec/batch)
step 78230, loss=0.65  (1924.8  examples/sec;  0.067  sec/batch)
step 78240, loss=0.70  (1861.0  examples/sec;  0.069  sec/batch)
step 78250, loss=0.65  (1947.5  examples/sec;  0.066  sec/batch)
step 78260, loss=0.65  (2011.6  examples/sec;  0.064  sec/batch)
step 78270, loss=0.59  (1838.5  examples/sec;  0.070  sec/batch)
step 78280, loss=0.61  (2019.6  examples/sec;  0.063  sec/batch)
step 78290, loss=0.48  (1907.2  examples/sec;  0.067  sec/batch)
step 78300, loss=0.60  (1880.7  examples/sec;  0.068  sec/batch)
step 78310, loss=0.58  (1936.9  examples/sec;  0.066  sec/batch)
step 78320, loss=0.63  (1886.5  examples/sec;  0.068  sec/batch)
step 78330, loss=0.61  (1866.5  examples/sec;  0.069  sec/batch)
step 78340, loss=0.57  (1645.2  examples/sec;  0.078  sec/batch)
step 78350, loss=0.53  (1868.6  examples/sec;  0.069  sec/batch)
step 78360, loss=0.75  (1912.2  examples/sec;  0.067  sec/batch)
step 78370, loss=0.54  (1918.3  examples/sec;  0.067  sec/batch)
step 78380, loss=0.66  (1803.3  examples/sec;  0.071  sec/batch)
step 78390, loss=0.51  (1884.6  examples/sec;  0.068  sec/batch)
step 78400, loss=0.63  (1922.7  examples/sec;  0.067  sec/batch)
step 78410, loss=0.64  (2014.6  examples/sec;  0.064  sec/batch)
step 78420, loss=0.45  (1893.8  examples/sec;  0.068  sec/batch)
step 78430, loss=0.48  (1933.4  examples/sec;  0.066  sec/batch)
step 78440, loss=0.67  (2030.2  examples/sec;  0.063  sec/batch)
step 78450, loss=0.73  (1871.5  examples/sec;  0.068  sec/batch)
step 78460, loss=0.70  (1902.5  examples/sec;  0.067  sec/batch)
step 78470, loss=0.58  (1958.0  examples/sec;  0.065  sec/batch)
step 78480, loss=0.68  (1894.1  examples/sec;  0.068  sec/batch)
step 78490, loss=0.73  (1909.4  examples/sec;  0.067  sec/batch)
step 78500, loss=0.72  (1908.5  examples/sec;  0.067  sec/batch)
step 78510, loss=0.58  (1899.4  examples/sec;  0.067  sec/batch)
step 78520, loss=0.50  (1918.6  examples/sec;  0.067  sec/batch)
step 78530, loss=0.54  (1887.6  examples/sec;  0.068  sec/batch)
step 78540, loss=0.46  (1928.9  examples/sec;  0.066  sec/batch)
step 78550, loss=0.58  (1874.2  examples/sec;  0.068  sec/batch)
step 78560, loss=0.59  (1965.0  examples/sec;  0.065  sec/batch)
step 78570, loss=0.42  (1985.0  examples/sec;  0.064  sec/batch)
step 78580, loss=0.61  (1932.1  examples/sec;  0.066  sec/batch)
step 78590, loss=0.44  (2035.8  examples/sec;  0.063  sec/batch)
step 78600, loss=0.62  (1827.2  examples/sec;  0.070  sec/batch)
step 78610, loss=0.69  (1867.1  examples/sec;  0.069  sec/batch)
step 78620, loss=0.59  (1914.5  examples/sec;  0.067  sec/batch)
step 78630, loss=0.52  (1821.2  examples/sec;  0.070  sec/batch)
step 78640, loss=0.72  (1916.1  examples/sec;  0.067  sec/batch)
step 78650, loss=0.66  (1987.4  examples/sec;  0.064  sec/batch)
step 78660, loss=0.46  (1953.2  examples/sec;  0.066  sec/batch)
step 78670, loss=0.43  (1821.5  examples/sec;  0.070  sec/batch)
step 78680, loss=0.66  (1967.9  examples/sec;  0.065  sec/batch)
step 78690, loss=0.87  (1863.4  examples/sec;  0.069  sec/batch)
step 78700, loss=0.71  (1920.7  examples/sec;  0.067  sec/batch)
step 78710, loss=0.71  (1893.0  examples/sec;  0.068  sec/batch)
step 78720, loss=0.58  (1950.1  examples/sec;  0.066  sec/batch)
step 78730, loss=0.51  (1914.8  examples/sec;  0.067  sec/batch)
step 78740, loss=0.54  (1857.2  examples/sec;  0.069  sec/batch)
step 78750, loss=0.61  (1914.4  examples/sec;  0.067  sec/batch)
step 78760, loss=0.75  (1903.1  examples/sec;  0.067  sec/batch)
step 78770, loss=0.48  (1960.8  examples/sec;  0.065  sec/batch)
step 78780, loss=0.59  (1908.9  examples/sec;  0.067  sec/batch)
step 78790, loss=0.60  (1867.1  examples/sec;  0.069  sec/batch)
step 78800, loss=0.42  (1865.2  examples/sec;  0.069  sec/batch)
step 78810, loss=0.41  (1771.2  examples/sec;  0.072  sec/batch)
step 78820, loss=0.67  (1927.4  examples/sec;  0.066  sec/batch)
step 78830, loss=0.52  (1921.7  examples/sec;  0.067  sec/batch)
step 78840, loss=0.67  (1818.8  examples/sec;  0.070  sec/batch)
step 78850, loss=0.46  (1863.1  examples/sec;  0.069  sec/batch)
step 78860, loss=0.62  (1830.5  examples/sec;  0.070  sec/batch)
step 78870, loss=0.58  (1784.7  examples/sec;  0.072  sec/batch)
step 78880, loss=0.55  (1770.5  examples/sec;  0.072  sec/batch)
step 78890, loss=0.72  (1891.9  examples/sec;  0.068  sec/batch)
step 78900, loss=0.62  (1612.7  examples/sec;  0.079  sec/batch)
step 78910, loss=0.56  (1621.5  examples/sec;  0.079  sec/batch)
step 78920, loss=0.61  (1893.6  examples/sec;  0.068  sec/batch)
step 78930, loss=0.56  (1933.4  examples/sec;  0.066  sec/batch)
step 78940, loss=0.59  (1957.1  examples/sec;  0.065  sec/batch)
step 78950, loss=0.57  (2008.5  examples/sec;  0.064  sec/batch)
step 78960, loss=0.58  (1942.5  examples/sec;  0.066  sec/batch)
step 78970, loss=0.56  (1902.3  examples/sec;  0.067  sec/batch)
step 78980, loss=0.65  (1664.9  examples/sec;  0.077  sec/batch)
step 78990, loss=0.51  (1911.9  examples/sec;  0.067  sec/batch)
step 79000, loss=0.59  (1935.9  examples/sec;  0.066  sec/batch)
step 79010, loss=0.62  (1919.7  examples/sec;  0.067  sec/batch)
step 79020, loss=0.74  (1948.8  examples/sec;  0.066  sec/batch)
step 79030, loss=0.50  (1851.2  examples/sec;  0.069  sec/batch)
step 79040, loss=0.58  (1906.9  examples/sec;  0.067  sec/batch)
step 79050, loss=0.69  (1929.0  examples/sec;  0.066  sec/batch)
step 79060, loss=0.60  (1919.8  examples/sec;  0.067  sec/batch)
step 79070, loss=0.72  (1950.6  examples/sec;  0.066  sec/batch)
step 79080, loss=0.51  (1726.7  examples/sec;  0.074  sec/batch)
step 79090, loss=0.59  (1856.7  examples/sec;  0.069  sec/batch)
step 79100, loss=0.62  (1907.9  examples/sec;  0.067  sec/batch)
step 79110, loss=0.64  (1965.5  examples/sec;  0.065  sec/batch)
step 79120, loss=0.68  (1842.0  examples/sec;  0.069  sec/batch)
step 79130, loss=0.73  (1925.8  examples/sec;  0.066  sec/batch)
step 79140, loss=0.64  (1873.4  examples/sec;  0.068  sec/batch)
step 79150, loss=0.52  (1897.3  examples/sec;  0.067  sec/batch)
step 79160, loss=0.61  (1868.0  examples/sec;  0.069  sec/batch)
step 79170, loss=0.47  (1895.3  examples/sec;  0.068  sec/batch)
step 79180, loss=0.45  (1981.2  examples/sec;  0.065  sec/batch)
step 79190, loss=0.56  (1791.8  examples/sec;  0.071  sec/batch)
step 79200, loss=0.68  (1889.4  examples/sec;  0.068  sec/batch)
step 79210, loss=0.59  (1915.7  examples/sec;  0.067  sec/batch)
step 79220, loss=0.50  (1869.0  examples/sec;  0.068  sec/batch)
step 79230, loss=0.79  (1964.5  examples/sec;  0.065  sec/batch)
step 79240, loss=0.49  (1754.7  examples/sec;  0.073  sec/batch)
step 79250, loss=0.66  (1949.3  examples/sec;  0.066  sec/batch)
step 79260, loss=0.75  (1978.9  examples/sec;  0.065  sec/batch)
step 79270, loss=0.61  (1790.9  examples/sec;  0.071  sec/batch)
step 79280, loss=0.52  (1827.2  examples/sec;  0.070  sec/batch)
step 79290, loss=0.80  (1896.5  examples/sec;  0.067  sec/batch)
step 79300, loss=0.59  (1881.2  examples/sec;  0.068  sec/batch)
step 79310, loss=0.52  (1897.2  examples/sec;  0.067  sec/batch)
step 79320, loss=0.60  (1891.2  examples/sec;  0.068  sec/batch)
step 79330, loss=0.58  (1898.3  examples/sec;  0.067  sec/batch)
step 79340, loss=0.64  (1982.5  examples/sec;  0.065  sec/batch)
step 79350, loss=0.80  (1874.4  examples/sec;  0.068  sec/batch)
step 79360, loss=0.49  (1894.4  examples/sec;  0.068  sec/batch)
step 79370, loss=0.46  (1922.3  examples/sec;  0.067  sec/batch)
step 79380, loss=0.58  (1895.9  examples/sec;  0.068  sec/batch)
step 79390, loss=0.55  (1865.5  examples/sec;  0.069  sec/batch)
step 79400, loss=0.68  (1970.7  examples/sec;  0.065  sec/batch)
step 79410, loss=0.57  (1832.6  examples/sec;  0.070  sec/batch)
step 79420, loss=0.57  (1940.8  examples/sec;  0.066  sec/batch)
step 79430, loss=0.61  (1950.3  examples/sec;  0.066  sec/batch)
step 79440, loss=0.56  (1887.0  examples/sec;  0.068  sec/batch)
step 79450, loss=0.67  (1925.7  examples/sec;  0.066  sec/batch)
step 79460, loss=0.57  (2002.5  examples/sec;  0.064  sec/batch)
step 79470, loss=0.58  (1888.5  examples/sec;  0.068  sec/batch)
step 79480, loss=0.57  (1917.2  examples/sec;  0.067  sec/batch)
step 79490, loss=0.53  (1843.5  examples/sec;  0.069  sec/batch)
step 79500, loss=0.61  (1876.5  examples/sec;  0.068  sec/batch)
step 79510, loss=0.56  (1917.6  examples/sec;  0.067  sec/batch)
step 79520, loss=0.56  (1912.0  examples/sec;  0.067  sec/batch)
step 79530, loss=0.59  (1982.1  examples/sec;  0.065  sec/batch)
step 79540, loss=0.57  (1938.3  examples/sec;  0.066  sec/batch)
step 79550, loss=0.64  (1868.7  examples/sec;  0.068  sec/batch)
step 79560, loss=0.62  (1981.6  examples/sec;  0.065  sec/batch)
step 79570, loss=0.48  (1963.8  examples/sec;  0.065  sec/batch)
step 79580, loss=0.64  (1867.9  examples/sec;  0.069  sec/batch)
step 79590, loss=0.71  (1878.6  examples/sec;  0.068  sec/batch)
step 79600, loss=0.59  (1837.6  examples/sec;  0.070  sec/batch)
step 79610, loss=0.54  (1938.6  examples/sec;  0.066  sec/batch)
step 79620, loss=0.58  (1954.5  examples/sec;  0.065  sec/batch)
step 79630, loss=0.54  (1862.6  examples/sec;  0.069  sec/batch)
step 79640, loss=0.63  (1839.7  examples/sec;  0.070  sec/batch)
step 79650, loss=0.47  (1878.4  examples/sec;  0.068  sec/batch)
step 79660, loss=0.62  (1921.5  examples/sec;  0.067  sec/batch)
step 79670, loss=0.49  (1862.9  examples/sec;  0.069  sec/batch)
step 79680, loss=0.56  (1894.3  examples/sec;  0.068  sec/batch)
step 79690, loss=0.74  (1925.8  examples/sec;  0.066  sec/batch)
step 79700, loss=0.63  (1873.4  examples/sec;  0.068  sec/batch)
step 79710, loss=0.69  (2004.2  examples/sec;  0.064  sec/batch)
step 79720, loss=0.65  (1911.6  examples/sec;  0.067  sec/batch)
step 79730, loss=0.67  (1924.1  examples/sec;  0.067  sec/batch)
step 79740, loss=0.66  (1898.5  examples/sec;  0.067  sec/batch)
step 79750, loss=0.66  (1974.1  examples/sec;  0.065  sec/batch)
step 79760, loss=0.58  (1906.8  examples/sec;  0.067  sec/batch)
step 79770, loss=0.64  (1868.4  examples/sec;  0.069  sec/batch)
step 79780, loss=0.51  (1901.4  examples/sec;  0.067  sec/batch)
step 79790, loss=0.58  (1936.4  examples/sec;  0.066  sec/batch)
step 79800, loss=0.67  (1934.5  examples/sec;  0.066  sec/batch)
step 79810, loss=0.58  (1903.5  examples/sec;  0.067  sec/batch)
step 79820, loss=0.65  (1912.1  examples/sec;  0.067  sec/batch)
step 79830, loss=0.53  (1950.4  examples/sec;  0.066  sec/batch)
step 79840, loss=0.67  (1912.3  examples/sec;  0.067  sec/batch)
step 79850, loss=0.65  (1896.3  examples/sec;  0.068  sec/batch)
step 79860, loss=0.61  (1913.1  examples/sec;  0.067  sec/batch)
step 79870, loss=0.56  (1918.1  examples/sec;  0.067  sec/batch)
step 79880, loss=0.71  (1961.8  examples/sec;  0.065  sec/batch)
step 79890, loss=0.64  (1990.5  examples/sec;  0.064  sec/batch)
step 79900, loss=0.68  (1876.3  examples/sec;  0.068  sec/batch)
step 79910, loss=0.54  (1868.1  examples/sec;  0.069  sec/batch)
step 79920, loss=0.56  (1976.0  examples/sec;  0.065  sec/batch)
step 79930, loss=0.57  (1841.7  examples/sec;  0.070  sec/batch)
step 79940, loss=0.59  (1870.1  examples/sec;  0.068  sec/batch)
step 79950, loss=0.66  (2065.6  examples/sec;  0.062  sec/batch)
step 79960, loss=0.78  (2048.2  examples/sec;  0.062  sec/batch)
step 79970, loss=0.67  (1911.2  examples/sec;  0.067  sec/batch)
step 79980, loss=0.50  (1977.9  examples/sec;  0.065  sec/batch)
step 79990, loss=0.57  (1936.7  examples/sec;  0.066  sec/batch)
step 80000, loss=0.45  (1900.7  examples/sec;  0.067  sec/batch)
step 80010, loss=0.52  (1781.9  examples/sec;  0.072  sec/batch)
step 80020, loss=0.72  (1852.0  examples/sec;  0.069  sec/batch)
step 80030, loss=0.48  (1881.4  examples/sec;  0.068  sec/batch)
step 80040, loss=0.75  (1942.6  examples/sec;  0.066  sec/batch)
step 80050, loss=0.61  (1870.9  examples/sec;  0.068  sec/batch)
step 80060, loss=0.59  (1897.7  examples/sec;  0.067  sec/batch)
step 80070, loss=0.56  (1904.9  examples/sec;  0.067  sec/batch)
step 80080, loss=0.78  (1846.4  examples/sec;  0.069  sec/batch)
step 80090, loss=0.60  (1825.2  examples/sec;  0.070  sec/batch)
step 80100, loss=0.57  (1855.9  examples/sec;  0.069  sec/batch)
step 80110, loss=0.54  (1893.4  examples/sec;  0.068  sec/batch)
step 80120, loss=0.79  (1865.5  examples/sec;  0.069  sec/batch)
step 80130, loss=0.47  (1735.9  examples/sec;  0.074  sec/batch)
step 80140, loss=0.62  (1908.9  examples/sec;  0.067  sec/batch)
step 80150, loss=0.65  (1839.0  examples/sec;  0.070  sec/batch)
step 80160, loss=0.56  (1991.3  examples/sec;  0.064  sec/batch)
step 80170, loss=0.73  (1876.1  examples/sec;  0.068  sec/batch)
step 80180, loss=0.67  (1944.2  examples/sec;  0.066  sec/batch)
step 80190, loss=0.72  (1812.1  examples/sec;  0.071  sec/batch)
step 80200, loss=0.57  (1976.5  examples/sec;  0.065  sec/batch)
step 80210, loss=0.58  (1864.4  examples/sec;  0.069  sec/batch)
step 80220, loss=0.55  (1861.0  examples/sec;  0.069  sec/batch)
step 80230, loss=0.55  (1920.5  examples/sec;  0.067  sec/batch)
step 80240, loss=0.46  (1910.0  examples/sec;  0.067  sec/batch)
step 80250, loss=0.55  (1860.7  examples/sec;  0.069  sec/batch)
step 80260, loss=0.59  (1862.0  examples/sec;  0.069  sec/batch)
step 80270, loss=0.67  (1913.3  examples/sec;  0.067  sec/batch)
step 80280, loss=0.61  (1847.2  examples/sec;  0.069  sec/batch)
step 80290, loss=0.72  (1933.0  examples/sec;  0.066  sec/batch)
step 80300, loss=0.73  (2003.6  examples/sec;  0.064  sec/batch)
step 80310, loss=0.43  (1894.5  examples/sec;  0.068  sec/batch)
step 80320, loss=0.63  (1861.0  examples/sec;  0.069  sec/batch)
step 80330, loss=0.62  (1904.7  examples/sec;  0.067  sec/batch)
step 80340, loss=0.63  (1939.9  examples/sec;  0.066  sec/batch)
step 80350, loss=0.70  (1894.1  examples/sec;  0.068  sec/batch)
step 80360, loss=0.67  (2071.6  examples/sec;  0.062  sec/batch)
step 80370, loss=0.72  (1858.5  examples/sec;  0.069  sec/batch)
step 80380, loss=0.62  (1882.3  examples/sec;  0.068  sec/batch)
step 80390, loss=0.47  (1916.0  examples/sec;  0.067  sec/batch)
step 80400, loss=0.52  (1912.8  examples/sec;  0.067  sec/batch)
step 80410, loss=0.67  (1692.9  examples/sec;  0.076  sec/batch)
step 80420, loss=0.51  (1854.4  examples/sec;  0.069  sec/batch)
step 80430, loss=0.63  (1996.2  examples/sec;  0.064  sec/batch)
step 80440, loss=0.68  (1961.6  examples/sec;  0.065  sec/batch)
step 80450, loss=0.68  (1946.2  examples/sec;  0.066  sec/batch)
step 80460, loss=0.51  (1927.3  examples/sec;  0.066  sec/batch)
step 80470, loss=0.53  (1890.4  examples/sec;  0.068  sec/batch)
step 80480, loss=0.58  (1873.2  examples/sec;  0.068  sec/batch)
step 80490, loss=0.72  (1935.5  examples/sec;  0.066  sec/batch)
step 80500, loss=0.49  (1831.5  examples/sec;  0.070  sec/batch)
step 80510, loss=0.76  (1923.0  examples/sec;  0.067  sec/batch)
step 80520, loss=0.67  (1917.2  examples/sec;  0.067  sec/batch)
step 80530, loss=0.68  (1772.0  examples/sec;  0.072  sec/batch)
step 80540, loss=0.53  (1880.1  examples/sec;  0.068  sec/batch)
step 80550, loss=0.49  (1723.6  examples/sec;  0.074  sec/batch)
step 80560, loss=0.56  (1751.6  examples/sec;  0.073  sec/batch)
step 80570, loss=0.53  (1863.4  examples/sec;  0.069  sec/batch)
step 80580, loss=0.58  (1951.8  examples/sec;  0.066  sec/batch)
step 80590, loss=0.51  (1868.2  examples/sec;  0.069  sec/batch)
step 80600, loss=0.61  (1880.9  examples/sec;  0.068  sec/batch)
step 80610, loss=0.72  (1858.3  examples/sec;  0.069  sec/batch)
step 80620, loss=0.59  (1886.9  examples/sec;  0.068  sec/batch)
step 80630, loss=0.58  (1929.8  examples/sec;  0.066  sec/batch)
step 80640, loss=0.70  (1928.9  examples/sec;  0.066  sec/batch)
step 80650, loss=0.57  (1863.4  examples/sec;  0.069  sec/batch)
step 80660, loss=0.62  (1865.3  examples/sec;  0.069  sec/batch)
step 80670, loss=0.57  (1959.3  examples/sec;  0.065  sec/batch)
step 80680, loss=0.51  (1872.7  examples/sec;  0.068  sec/batch)
step 80690, loss=0.58  (1899.3  examples/sec;  0.067  sec/batch)
step 80700, loss=0.71  (1792.4  examples/sec;  0.071  sec/batch)
step 80710, loss=0.56  (1902.1  examples/sec;  0.067  sec/batch)
step 80720, loss=0.75  (1935.5  examples/sec;  0.066  sec/batch)
step 80730, loss=0.59  (1921.2  examples/sec;  0.067  sec/batch)
step 80740, loss=0.57  (1903.5  examples/sec;  0.067  sec/batch)
step 80750, loss=0.65  (1746.6  examples/sec;  0.073  sec/batch)
step 80760, loss=0.47  (1882.4  examples/sec;  0.068  sec/batch)
step 80770, loss=0.51  (1808.5  examples/sec;  0.071  sec/batch)
step 80780, loss=0.67  (1877.6  examples/sec;  0.068  sec/batch)
step 80790, loss=0.89  (1911.9  examples/sec;  0.067  sec/batch)
step 80800, loss=0.51  (1869.2  examples/sec;  0.068  sec/batch)
step 80810, loss=0.57  (1813.1  examples/sec;  0.071  sec/batch)
step 80820, loss=0.58  (1912.4  examples/sec;  0.067  sec/batch)
step 80830, loss=0.64  (1812.7  examples/sec;  0.071  sec/batch)
step 80840, loss=0.65  (1734.5  examples/sec;  0.074  sec/batch)
step 80850, loss=0.75  (1983.4  examples/sec;  0.065  sec/batch)
step 80860, loss=0.58  (1929.7  examples/sec;  0.066  sec/batch)
step 80870, loss=0.77  (1885.8  examples/sec;  0.068  sec/batch)
step 80880, loss=0.45  (1992.5  examples/sec;  0.064  sec/batch)
step 80890, loss=0.64  (1953.0  examples/sec;  0.066  sec/batch)
step 80900, loss=0.57  (1695.1  examples/sec;  0.076  sec/batch)
step 80910, loss=0.55  (1869.8  examples/sec;  0.068  sec/batch)
step 80920, loss=0.66  (1919.2  examples/sec;  0.067  sec/batch)
step 80930, loss=0.51  (1957.8  examples/sec;  0.065  sec/batch)
step 80940, loss=0.51  (1903.5  examples/sec;  0.067  sec/batch)
step 80950, loss=0.50  (1925.5  examples/sec;  0.066  sec/batch)
step 80960, loss=0.71  (1910.7  examples/sec;  0.067  sec/batch)
step 80970, loss=0.67  (1903.2  examples/sec;  0.067  sec/batch)
step 80980, loss=0.61  (1872.7  examples/sec;  0.068  sec/batch)
step 80990, loss=0.70  (1933.7  examples/sec;  0.066  sec/batch)
step 81000, loss=0.59  (1845.9  examples/sec;  0.069  sec/batch)
step 81010, loss=0.51  (1870.0  examples/sec;  0.068  sec/batch)
step 81020, loss=0.59  (1895.6  examples/sec;  0.068  sec/batch)
step 81030, loss=0.49  (1928.4  examples/sec;  0.066  sec/batch)
step 81040, loss=0.62  (1852.5  examples/sec;  0.069  sec/batch)
step 81050, loss=0.55  (1890.3  examples/sec;  0.068  sec/batch)
step 81060, loss=0.63  (1868.0  examples/sec;  0.069  sec/batch)
step 81070, loss=0.70  (1913.0  examples/sec;  0.067  sec/batch)
step 81080, loss=0.56  (1863.6  examples/sec;  0.069  sec/batch)
step 81090, loss=0.55  (1817.2  examples/sec;  0.070  sec/batch)
step 81100, loss=0.60  (2028.0  examples/sec;  0.063  sec/batch)
step 81110, loss=0.77  (1842.0  examples/sec;  0.069  sec/batch)
step 81120, loss=0.47  (1879.5  examples/sec;  0.068  sec/batch)
step 81130, loss=0.60  (1848.1  examples/sec;  0.069  sec/batch)
step 81140, loss=0.49  (1895.7  examples/sec;  0.068  sec/batch)
step 81150, loss=0.72  (1307.4  examples/sec;  0.098  sec/batch)
step 81160, loss=0.69  (1908.1  examples/sec;  0.067  sec/batch)
step 81170, loss=0.43  (1861.0  examples/sec;  0.069  sec/batch)
step 81180, loss=0.67  (1926.3  examples/sec;  0.066  sec/batch)
step 81190, loss=0.60  (1862.7  examples/sec;  0.069  sec/batch)
step 81200, loss=0.73  (1937.4  examples/sec;  0.066  sec/batch)
step 81210, loss=0.62  (1944.1  examples/sec;  0.066  sec/batch)
step 81220, loss=0.63  (1978.9  examples/sec;  0.065  sec/batch)
step 81230, loss=0.63  (1935.9  examples/sec;  0.066  sec/batch)
step 81240, loss=0.65  (1931.4  examples/sec;  0.066  sec/batch)
step 81250, loss=0.43  (1867.2  examples/sec;  0.069  sec/batch)
step 81260, loss=0.72  (1852.9  examples/sec;  0.069  sec/batch)
step 81270, loss=0.71  (1899.8  examples/sec;  0.067  sec/batch)
step 81280, loss=0.52  (1898.6  examples/sec;  0.067  sec/batch)
step 81290, loss=0.62  (1914.5  examples/sec;  0.067  sec/batch)
step 81300, loss=0.62  (1822.2  examples/sec;  0.070  sec/batch)
step 81310, loss=0.57  (1734.8  examples/sec;  0.074  sec/batch)
step 81320, loss=0.61  (1824.3  examples/sec;  0.070  sec/batch)
step 81330, loss=0.67  (1855.9  examples/sec;  0.069  sec/batch)
step 81340, loss=0.63  (1852.6  examples/sec;  0.069  sec/batch)
step 81350, loss=0.67  (1955.1  examples/sec;  0.065  sec/batch)
step 81360, loss=0.51  (1940.8  examples/sec;  0.066  sec/batch)
step 81370, loss=0.57  (1952.1  examples/sec;  0.066  sec/batch)
step 81380, loss=0.70  (1934.2  examples/sec;  0.066  sec/batch)
step 81390, loss=0.57  (2050.9  examples/sec;  0.062  sec/batch)
step 81400, loss=0.62  (1937.4  examples/sec;  0.066  sec/batch)
step 81410, loss=0.65  (1897.0  examples/sec;  0.067  sec/batch)
step 81420, loss=0.62  (1927.4  examples/sec;  0.066  sec/batch)
step 81430, loss=0.48  (1926.8  examples/sec;  0.066  sec/batch)
step 81440, loss=0.65  (1876.4  examples/sec;  0.068  sec/batch)
step 81450, loss=0.56  (1937.5  examples/sec;  0.066  sec/batch)
step 81460, loss=0.74  (1925.8  examples/sec;  0.066  sec/batch)
step 81470, loss=0.57  (1959.7  examples/sec;  0.065  sec/batch)
step 81480, loss=0.51  (1828.6  examples/sec;  0.070  sec/batch)
step 81490, loss=0.56  (1919.4  examples/sec;  0.067  sec/batch)
step 81500, loss=0.61  (1854.2  examples/sec;  0.069  sec/batch)
step 81510, loss=0.68  (1833.4  examples/sec;  0.070  sec/batch)
step 81520, loss=0.65  (1905.1  examples/sec;  0.067  sec/batch)
step 81530, loss=0.59  (1788.5  examples/sec;  0.072  sec/batch)
step 81540, loss=0.84  (1970.8  examples/sec;  0.065  sec/batch)
step 81550, loss=0.66  (1843.5  examples/sec;  0.069  sec/batch)
step 81560, loss=0.60  (1908.5  examples/sec;  0.067  sec/batch)
step 81570, loss=0.56  (2038.2  examples/sec;  0.063  sec/batch)
step 81580, loss=0.58  (1850.8  examples/sec;  0.069  sec/batch)
step 81590, loss=0.58  (1909.6  examples/sec;  0.067  sec/batch)
step 81600, loss=0.58  (1942.7  examples/sec;  0.066  sec/batch)
step 81610, loss=0.50  (1885.4  examples/sec;  0.068  sec/batch)
step 81620, loss=0.68  (1944.5  examples/sec;  0.066  sec/batch)
step 81630, loss=0.66  (1928.6  examples/sec;  0.066  sec/batch)
step 81640, loss=0.64  (1872.4  examples/sec;  0.068  sec/batch)
step 81650, loss=0.52  (1851.0  examples/sec;  0.069  sec/batch)
step 81660, loss=0.63  (1837.7  examples/sec;  0.070  sec/batch)
step 81670, loss=0.51  (1821.1  examples/sec;  0.070  sec/batch)
step 81680, loss=0.63  (1823.7  examples/sec;  0.070  sec/batch)
step 81690, loss=0.58  (1889.2  examples/sec;  0.068  sec/batch)
step 81700, loss=0.59  (1876.7  examples/sec;  0.068  sec/batch)
step 81710, loss=0.82  (1922.6  examples/sec;  0.067  sec/batch)
step 81720, loss=0.67  (1635.2  examples/sec;  0.078  sec/batch)
step 81730, loss=0.58  (1956.6  examples/sec;  0.065  sec/batch)
step 81740, loss=0.67  (1887.8  examples/sec;  0.068  sec/batch)
step 81750, loss=0.76  (1873.5  examples/sec;  0.068  sec/batch)
step 81760, loss=0.60  (1986.6  examples/sec;  0.064  sec/batch)
step 81770, loss=0.87  (1893.0  examples/sec;  0.068  sec/batch)
step 81780, loss=0.59  (1892.8  examples/sec;  0.068  sec/batch)
step 81790, loss=0.60  (1884.8  examples/sec;  0.068  sec/batch)
step 81800, loss=0.70  (1852.8  examples/sec;  0.069  sec/batch)
step 81810, loss=0.63  (1887.5  examples/sec;  0.068  sec/batch)
step 81820, loss=0.57  (1825.3  examples/sec;  0.070  sec/batch)
step 81830, loss=0.59  (1847.2  examples/sec;  0.069  sec/batch)
step 81840, loss=0.51  (1834.1  examples/sec;  0.070  sec/batch)
step 81850, loss=0.67  (1974.2  examples/sec;  0.065  sec/batch)
step 81860, loss=0.64  (1865.8  examples/sec;  0.069  sec/batch)
step 81870, loss=0.69  (1852.5  examples/sec;  0.069  sec/batch)
step 81880, loss=0.73  (2052.0  examples/sec;  0.062  sec/batch)
step 81890, loss=0.54  (1880.2  examples/sec;  0.068  sec/batch)
step 81900, loss=0.69  (1890.0  examples/sec;  0.068  sec/batch)
step 81910, loss=0.52  (1899.0  examples/sec;  0.067  sec/batch)
step 81920, loss=0.60  (1847.3  examples/sec;  0.069  sec/batch)
step 81930, loss=0.71  (1946.3  examples/sec;  0.066  sec/batch)
step 81940, loss=0.71  (1971.9  examples/sec;  0.065  sec/batch)
step 81950, loss=0.76  (1932.5  examples/sec;  0.066  sec/batch)
step 81960, loss=0.58  (1580.6  examples/sec;  0.081  sec/batch)
step 81970, loss=0.65  (1993.8  examples/sec;  0.064  sec/batch)
step 81980, loss=0.92  (1846.6  examples/sec;  0.069  sec/batch)
step 81990, loss=0.64  (1899.9  examples/sec;  0.067  sec/batch)
step 82000, loss=0.52  (1851.9  examples/sec;  0.069  sec/batch)
step 82010, loss=0.55  (2019.9  examples/sec;  0.063  sec/batch)
step 82020, loss=0.55  (1794.5  examples/sec;  0.071  sec/batch)
step 82030, loss=0.59  (1840.3  examples/sec;  0.070  sec/batch)
step 82040, loss=0.60  (1927.9  examples/sec;  0.066  sec/batch)
step 82050, loss=0.68  (1884.4  examples/sec;  0.068  sec/batch)
step 82060, loss=0.65  (1573.2  examples/sec;  0.081  sec/batch)
step 82070, loss=0.59  (1566.9  examples/sec;  0.082  sec/batch)
step 82080, loss=0.60  (1942.2  examples/sec;  0.066  sec/batch)
step 82090, loss=0.57  (1950.4  examples/sec;  0.066  sec/batch)
step 82100, loss=0.59  (1888.9  examples/sec;  0.068  sec/batch)
step 82110, loss=0.52  (1985.2  examples/sec;  0.064  sec/batch)
step 82120, loss=0.58  (1907.9  examples/sec;  0.067  sec/batch)
step 82130, loss=0.55  (1907.1  examples/sec;  0.067  sec/batch)
step 82140, loss=0.51  (1572.0  examples/sec;  0.081  sec/batch)
step 82150, loss=0.58  (1907.5  examples/sec;  0.067  sec/batch)
step 82160, loss=0.74  (1931.1  examples/sec;  0.066  sec/batch)
step 82170, loss=0.61  (1939.3  examples/sec;  0.066  sec/batch)
step 82180, loss=0.51  (1799.3  examples/sec;  0.071  sec/batch)
step 82190, loss=0.82  (1836.0  examples/sec;  0.070  sec/batch)
step 82200, loss=0.61  (1900.9  examples/sec;  0.067  sec/batch)
step 82210, loss=0.54  (1908.9  examples/sec;  0.067  sec/batch)
step 82220, loss=0.50  (1887.0  examples/sec;  0.068  sec/batch)
step 82230, loss=0.77  (1951.6  examples/sec;  0.066  sec/batch)
step 82240, loss=0.76  (1897.8  examples/sec;  0.067  sec/batch)
step 82250, loss=0.51  (1848.7  examples/sec;  0.069  sec/batch)
step 82260, loss=0.50  (1917.1  examples/sec;  0.067  sec/batch)
step 82270, loss=0.65  (1885.0  examples/sec;  0.068  sec/batch)
step 82280, loss=0.70  (1900.1  examples/sec;  0.067  sec/batch)
step 82290, loss=0.57  (1863.3  examples/sec;  0.069  sec/batch)
step 82300, loss=0.59  (1669.5  examples/sec;  0.077  sec/batch)
step 82310, loss=0.68  (1852.9  examples/sec;  0.069  sec/batch)
step 82320, loss=0.65  (1873.5  examples/sec;  0.068  sec/batch)
step 82330, loss=0.69  (1945.5  examples/sec;  0.066  sec/batch)
step 82340, loss=0.60  (1895.1  examples/sec;  0.068  sec/batch)
step 82350, loss=0.76  (1884.2  examples/sec;  0.068  sec/batch)
step 82360, loss=0.58  (1838.7  examples/sec;  0.070  sec/batch)
step 82370, loss=0.45  (1593.4  examples/sec;  0.080  sec/batch)
step 82380, loss=0.52  (1895.7  examples/sec;  0.068  sec/batch)
step 82390, loss=0.59  (1894.1  examples/sec;  0.068  sec/batch)
step 82400, loss=0.63  (1890.2  examples/sec;  0.068  sec/batch)
step 82410, loss=0.59  (1618.8  examples/sec;  0.079  sec/batch)
step 82420, loss=0.67  (1913.4  examples/sec;  0.067  sec/batch)
step 82430, loss=0.61  (1851.0  examples/sec;  0.069  sec/batch)
step 82440, loss=0.61  (1905.0  examples/sec;  0.067  sec/batch)
step 82450, loss=0.74  (1821.5  examples/sec;  0.070  sec/batch)
step 82460, loss=0.65  (1921.0  examples/sec;  0.067  sec/batch)
step 82470, loss=0.79  (1912.0  examples/sec;  0.067  sec/batch)
step 82480, loss=0.60  (1826.6  examples/sec;  0.070  sec/batch)
step 82490, loss=0.68  (1828.3  examples/sec;  0.070  sec/batch)
step 82500, loss=0.61  (1859.5  examples/sec;  0.069  sec/batch)
step 82510, loss=0.59  (1847.9  examples/sec;  0.069  sec/batch)
step 82520, loss=0.72  (1956.3  examples/sec;  0.065  sec/batch)
step 82530, loss=0.73  (1902.7  examples/sec;  0.067  sec/batch)
step 82540, loss=0.51  (1890.8  examples/sec;  0.068  sec/batch)
step 82550, loss=0.64  (1974.5  examples/sec;  0.065  sec/batch)
step 82560, loss=0.72  (1930.2  examples/sec;  0.066  sec/batch)
step 82570, loss=0.69  (1883.0  examples/sec;  0.068  sec/batch)
step 82580, loss=0.61  (1644.1  examples/sec;  0.078  sec/batch)
step 82590, loss=0.54  (1973.6  examples/sec;  0.065  sec/batch)
step 82600, loss=0.76  (1821.1  examples/sec;  0.070  sec/batch)
step 82610, loss=0.71  (1853.2  examples/sec;  0.069  sec/batch)
step 82620, loss=0.65  (1912.8  examples/sec;  0.067  sec/batch)
step 82630, loss=0.67  (1975.5  examples/sec;  0.065  sec/batch)
step 82640, loss=0.69  (1865.9  examples/sec;  0.069  sec/batch)
step 82650, loss=0.54  (1933.3  examples/sec;  0.066  sec/batch)
step 82660, loss=0.70  (1900.6  examples/sec;  0.067  sec/batch)
step 82670, loss=0.74  (1929.8  examples/sec;  0.066  sec/batch)
step 82680, loss=0.56  (1980.6  examples/sec;  0.065  sec/batch)
step 82690, loss=0.58  (1917.9  examples/sec;  0.067  sec/batch)
step 82700, loss=0.48  (1925.0  examples/sec;  0.066  sec/batch)
step 82710, loss=0.67  (1836.9  examples/sec;  0.070  sec/batch)
step 82720, loss=0.57  (1957.3  examples/sec;  0.065  sec/batch)
step 82730, loss=0.55  (1913.0  examples/sec;  0.067  sec/batch)
step 82740, loss=0.54  (1814.8  examples/sec;  0.071  sec/batch)
step 82750, loss=0.72  (1859.7  examples/sec;  0.069  sec/batch)
step 82760, loss=0.67  (1642.1  examples/sec;  0.078  sec/batch)
step 82770, loss=0.64  (1909.8  examples/sec;  0.067  sec/batch)
step 82780, loss=0.65  (1895.0  examples/sec;  0.068  sec/batch)
step 82790, loss=0.49  (1965.5  examples/sec;  0.065  sec/batch)
step 82800, loss=0.65  (1749.9  examples/sec;  0.073  sec/batch)
step 82810, loss=0.66  (1873.5  examples/sec;  0.068  sec/batch)
step 82820, loss=0.72  (1974.3  examples/sec;  0.065  sec/batch)
step 82830, loss=0.61  (1959.1  examples/sec;  0.065  sec/batch)
step 82840, loss=0.49  (1911.3  examples/sec;  0.067  sec/batch)
step 82850, loss=0.49  (1962.0  examples/sec;  0.065  sec/batch)
step 82860, loss=0.66  (1894.7  examples/sec;  0.068  sec/batch)
step 82870, loss=0.70  (1851.8  examples/sec;  0.069  sec/batch)
step 82880, loss=0.70  (1934.7  examples/sec;  0.066  sec/batch)
step 82890, loss=0.58  (1911.3  examples/sec;  0.067  sec/batch)
step 82900, loss=0.66  (1972.7  examples/sec;  0.065  sec/batch)
step 82910, loss=0.55  (1922.1  examples/sec;  0.067  sec/batch)
step 82920, loss=0.83  (1792.3  examples/sec;  0.071  sec/batch)
step 82930, loss=0.63  (1857.7  examples/sec;  0.069  sec/batch)
step 82940, loss=0.59  (1889.2  examples/sec;  0.068  sec/batch)
step 82950, loss=0.57  (1753.7  examples/sec;  0.073  sec/batch)
step 82960, loss=0.55  (1963.7  examples/sec;  0.065  sec/batch)
step 82970, loss=0.69  (1939.9  examples/sec;  0.066  sec/batch)
step 82980, loss=0.53  (1854.2  examples/sec;  0.069  sec/batch)
step 82990, loss=0.58  (1816.3  examples/sec;  0.070  sec/batch)
step 83000, loss=0.70  (1837.7  examples/sec;  0.070  sec/batch)
step 83010, loss=0.75  (1883.2  examples/sec;  0.068  sec/batch)
step 83020, loss=0.66  (1800.0  examples/sec;  0.071  sec/batch)
step 83030, loss=0.76  (1842.4  examples/sec;  0.069  sec/batch)
step 83040, loss=0.65  (1892.4  examples/sec;  0.068  sec/batch)
step 83050, loss=0.59  (2037.4  examples/sec;  0.063  sec/batch)
step 83060, loss=0.64  (1837.8  examples/sec;  0.070  sec/batch)
step 83070, loss=0.63  (1912.6  examples/sec;  0.067  sec/batch)
step 83080, loss=0.73  (1886.6  examples/sec;  0.068  sec/batch)
step 83090, loss=0.69  (1953.5  examples/sec;  0.066  sec/batch)
step 83100, loss=0.43  (1874.3  examples/sec;  0.068  sec/batch)
step 83110, loss=0.41  (1995.4  examples/sec;  0.064  sec/batch)
step 83120, loss=0.73  (1905.1  examples/sec;  0.067  sec/batch)
step 83130, loss=0.67  (1939.8  examples/sec;  0.066  sec/batch)
step 83140, loss=0.62  (1957.5  examples/sec;  0.065  sec/batch)
step 83150, loss=0.61  (1929.9  examples/sec;  0.066  sec/batch)
step 83160, loss=0.54  (1812.0  examples/sec;  0.071  sec/batch)
step 83170, loss=0.67  (1922.1  examples/sec;  0.067  sec/batch)
step 83180, loss=0.41  (1934.1  examples/sec;  0.066  sec/batch)
step 83190, loss=0.66  (1854.5  examples/sec;  0.069  sec/batch)
step 83200, loss=0.53  (1905.2  examples/sec;  0.067  sec/batch)
step 83210, loss=0.71  (1946.5  examples/sec;  0.066  sec/batch)
step 83220, loss=0.64  (1814.1  examples/sec;  0.071  sec/batch)
step 83230, loss=0.59  (1987.2  examples/sec;  0.064  sec/batch)
step 83240, loss=0.53  (1919.3  examples/sec;  0.067  sec/batch)
step 83250, loss=0.60  (1894.6  examples/sec;  0.068  sec/batch)
step 83260, loss=0.57  (1725.0  examples/sec;  0.074  sec/batch)
step 83270, loss=0.54  (1873.1  examples/sec;  0.068  sec/batch)
step 83280, loss=0.54  (1956.2  examples/sec;  0.065  sec/batch)
step 83290, loss=0.49  (1872.3  examples/sec;  0.068  sec/batch)
step 83300, loss=0.69  (1851.3  examples/sec;  0.069  sec/batch)
step 83310, loss=0.37  (1815.7  examples/sec;  0.070  sec/batch)
step 83320, loss=0.64  (1906.9  examples/sec;  0.067  sec/batch)
step 83330, loss=0.58  (1815.1  examples/sec;  0.071  sec/batch)
step 83340, loss=0.47  (1894.6  examples/sec;  0.068  sec/batch)
step 83350, loss=0.61  (1799.7  examples/sec;  0.071  sec/batch)
step 83360, loss=0.60  (1969.7  examples/sec;  0.065  sec/batch)
step 83370, loss=0.54  (1801.0  examples/sec;  0.071  sec/batch)
step 83380, loss=0.61  (1853.4  examples/sec;  0.069  sec/batch)
step 83390, loss=0.56  (1938.7  examples/sec;  0.066  sec/batch)
step 83400, loss=0.54  (1881.1  examples/sec;  0.068  sec/batch)
step 83410, loss=0.64  (1937.1  examples/sec;  0.066  sec/batch)
step 83420, loss=0.65  (1998.4  examples/sec;  0.064  sec/batch)
step 83430, loss=0.62  (1902.6  examples/sec;  0.067  sec/batch)
step 83440, loss=0.64  (1899.1  examples/sec;  0.067  sec/batch)
step 83450, loss=0.61  (1929.5  examples/sec;  0.066  sec/batch)
step 83460, loss=0.61  (1887.2  examples/sec;  0.068  sec/batch)
step 83470, loss=0.74  (1846.7  examples/sec;  0.069  sec/batch)
step 83480, loss=0.62  (1831.4  examples/sec;  0.070  sec/batch)
step 83490, loss=0.68  (1937.2  examples/sec;  0.066  sec/batch)
step 83500, loss=0.65  (1895.0  examples/sec;  0.068  sec/batch)
step 83510, loss=0.61  (1891.3  examples/sec;  0.068  sec/batch)
step 83520, loss=0.69  (2013.6  examples/sec;  0.064  sec/batch)
step 83530, loss=0.61  (1924.8  examples/sec;  0.067  sec/batch)
step 83540, loss=0.66  (1935.1  examples/sec;  0.066  sec/batch)
step 83550, loss=0.62  (1883.6  examples/sec;  0.068  sec/batch)
step 83560, loss=0.77  (1934.6  examples/sec;  0.066  sec/batch)
step 83570, loss=0.67  (1863.4  examples/sec;  0.069  sec/batch)
step 83580, loss=0.63  (1808.7  examples/sec;  0.071  sec/batch)
step 83590, loss=0.68  (1916.5  examples/sec;  0.067  sec/batch)
step 83600, loss=0.59  (1863.1  examples/sec;  0.069  sec/batch)
step 83610, loss=0.45  (1920.6  examples/sec;  0.067  sec/batch)
step 83620, loss=0.76  (1820.6  examples/sec;  0.070  sec/batch)
step 83630, loss=0.75  (1862.1  examples/sec;  0.069  sec/batch)
step 83640, loss=0.61  (1919.6  examples/sec;  0.067  sec/batch)
step 83650, loss=0.57  (1872.1  examples/sec;  0.068  sec/batch)
step 83660, loss=0.63  (1947.5  examples/sec;  0.066  sec/batch)
step 83670, loss=0.61  (1884.0  examples/sec;  0.068  sec/batch)
step 83680, loss=0.73  (1888.9  examples/sec;  0.068  sec/batch)
step 83690, loss=0.46  (1909.4  examples/sec;  0.067  sec/batch)
step 83700, loss=0.58  (1931.3  examples/sec;  0.066  sec/batch)
step 83710, loss=0.52  (1956.5  examples/sec;  0.065  sec/batch)
step 83720, loss=0.63  (1926.4  examples/sec;  0.066  sec/batch)
step 83730, loss=0.62  (1903.3  examples/sec;  0.067  sec/batch)
step 83740, loss=0.61  (1917.0  examples/sec;  0.067  sec/batch)
step 83750, loss=0.52  (1920.2  examples/sec;  0.067  sec/batch)
step 83760, loss=0.55  (1901.1  examples/sec;  0.067  sec/batch)
step 83770, loss=0.45  (1878.5  examples/sec;  0.068  sec/batch)
step 83780, loss=0.61  (1939.8  examples/sec;  0.066  sec/batch)
step 83790, loss=0.64  (1887.9  examples/sec;  0.068  sec/batch)
step 83800, loss=0.57  (2028.1  examples/sec;  0.063  sec/batch)
step 83810, loss=0.56  (1852.6  examples/sec;  0.069  sec/batch)
step 83820, loss=0.58  (1980.8  examples/sec;  0.065  sec/batch)
step 83830, loss=0.54  (1876.7  examples/sec;  0.068  sec/batch)
step 83840, loss=0.50  (1851.3  examples/sec;  0.069  sec/batch)
step 83850, loss=0.70  (1854.1  examples/sec;  0.069  sec/batch)
step 83860, loss=0.66  (1896.2  examples/sec;  0.068  sec/batch)
step 83870, loss=0.47  (1951.9  examples/sec;  0.066  sec/batch)
step 83880, loss=0.64  (1958.5  examples/sec;  0.065  sec/batch)
step 83890, loss=0.53  (1856.8  examples/sec;  0.069  sec/batch)
step 83900, loss=0.44  (1895.1  examples/sec;  0.068  sec/batch)
step 83910, loss=0.70  (1889.4  examples/sec;  0.068  sec/batch)
step 83920, loss=0.66  (1941.3  examples/sec;  0.066  sec/batch)
step 83930, loss=0.52  (1914.7  examples/sec;  0.067  sec/batch)
step 83940, loss=0.54  (1902.4  examples/sec;  0.067  sec/batch)
step 83950, loss=0.68  (1596.5  examples/sec;  0.080  sec/batch)
step 83960, loss=0.56  (1924.9  examples/sec;  0.066  sec/batch)
step 83970, loss=0.65  (1894.1  examples/sec;  0.068  sec/batch)
step 83980, loss=0.55  (1905.8  examples/sec;  0.067  sec/batch)
step 83990, loss=0.63  (1918.2  examples/sec;  0.067  sec/batch)
step 84000, loss=0.52  (1791.7  examples/sec;  0.071  sec/batch)
step 84010, loss=0.60  (1777.2  examples/sec;  0.072  sec/batch)
step 84020, loss=0.67  (1878.2  examples/sec;  0.068  sec/batch)
step 84030, loss=0.62  (1980.4  examples/sec;  0.065  sec/batch)
step 84040, loss=0.60  (1912.3  examples/sec;  0.067  sec/batch)
step 84050, loss=0.63  (1903.6  examples/sec;  0.067  sec/batch)
step 84060, loss=0.64  (1927.9  examples/sec;  0.066  sec/batch)
step 84070, loss=0.56  (1909.6  examples/sec;  0.067  sec/batch)
step 84080, loss=0.59  (1879.9  examples/sec;  0.068  sec/batch)
step 84090, loss=0.45  (1930.5  examples/sec;  0.066  sec/batch)
step 84100, loss=0.53  (1901.6  examples/sec;  0.067  sec/batch)
step 84110, loss=0.66  (1944.5  examples/sec;  0.066  sec/batch)
step 84120, loss=0.70  (1796.6  examples/sec;  0.071  sec/batch)
step 84130, loss=0.59  (1954.0  examples/sec;  0.066  sec/batch)
step 84140, loss=0.51  (1919.9  examples/sec;  0.067  sec/batch)
step 84150, loss=0.69  (1720.1  examples/sec;  0.074  sec/batch)
step 84160, loss=0.66  (1881.1  examples/sec;  0.068  sec/batch)
step 84170, loss=0.59  (1887.3  examples/sec;  0.068  sec/batch)
step 84180, loss=0.63  (1931.9  examples/sec;  0.066  sec/batch)
step 84190, loss=0.57  (1933.2  examples/sec;  0.066  sec/batch)
step 84200, loss=0.70  (1860.6  examples/sec;  0.069  sec/batch)
step 84210, loss=0.65  (1770.8  examples/sec;  0.072  sec/batch)
step 84220, loss=0.51  (1887.9  examples/sec;  0.068  sec/batch)
step 84230, loss=0.76  (1995.9  examples/sec;  0.064  sec/batch)
step 84240, loss=0.49  (1888.1  examples/sec;  0.068  sec/batch)
step 84250, loss=0.42  (1909.2  examples/sec;  0.067  sec/batch)
step 84260, loss=0.53  (1614.8  examples/sec;  0.079  sec/batch)
step 84270, loss=0.60  (1907.4  examples/sec;  0.067  sec/batch)
step 84280, loss=0.69  (1911.8  examples/sec;  0.067  sec/batch)
step 84290, loss=0.69  (1968.6  examples/sec;  0.065  sec/batch)
step 84300, loss=0.55  (1850.5  examples/sec;  0.069  sec/batch)
step 84310, loss=0.66  (1927.8  examples/sec;  0.066  sec/batch)
step 84320, loss=0.54  (1863.8  examples/sec;  0.069  sec/batch)
step 84330, loss=0.59  (2013.2  examples/sec;  0.064  sec/batch)
step 84340, loss=0.61  (1885.1  examples/sec;  0.068  sec/batch)
step 84350, loss=0.66  (1851.2  examples/sec;  0.069  sec/batch)
step 84360, loss=0.68  (1914.5  examples/sec;  0.067  sec/batch)
step 84370, loss=0.60  (1910.4  examples/sec;  0.067  sec/batch)
step 84380, loss=0.67  (1852.0  examples/sec;  0.069  sec/batch)
step 84390, loss=0.63  (1837.0  examples/sec;  0.070  sec/batch)
step 84400, loss=0.57  (1895.3  examples/sec;  0.068  sec/batch)
step 84410, loss=0.55  (1842.3  examples/sec;  0.069  sec/batch)
step 84420, loss=0.51  (1845.0  examples/sec;  0.069  sec/batch)
step 84430, loss=0.65  (1865.7  examples/sec;  0.069  sec/batch)
step 84440, loss=0.62  (1644.0  examples/sec;  0.078  sec/batch)
step 84450, loss=0.71  (1927.6  examples/sec;  0.066  sec/batch)
step 84460, loss=0.54  (1878.9  examples/sec;  0.068  sec/batch)
step 84470, loss=0.47  (1889.0  examples/sec;  0.068  sec/batch)
step 84480, loss=0.62  (1941.6  examples/sec;  0.066  sec/batch)
step 84490, loss=0.64  (1958.6  examples/sec;  0.065  sec/batch)
step 84500, loss=0.66  (1827.5  examples/sec;  0.070  sec/batch)
step 84510, loss=0.56  (1694.4  examples/sec;  0.076  sec/batch)
step 84520, loss=0.74  (1986.3  examples/sec;  0.064  sec/batch)
step 84530, loss=0.66  (1900.9  examples/sec;  0.067  sec/batch)
step 84540, loss=0.53  (1861.3  examples/sec;  0.069  sec/batch)
step 84550, loss=0.60  (1954.1  examples/sec;  0.066  sec/batch)
step 84560, loss=0.67  (1921.2  examples/sec;  0.067  sec/batch)
step 84570, loss=0.53  (1914.9  examples/sec;  0.067  sec/batch)
step 84580, loss=0.48  (1950.0  examples/sec;  0.066  sec/batch)
step 84590, loss=0.61  (1928.0  examples/sec;  0.066  sec/batch)
step 84600, loss=0.58  (1854.1  examples/sec;  0.069  sec/batch)
step 84610, loss=0.70  (1908.4  examples/sec;  0.067  sec/batch)
step 84620, loss=0.52  (1885.7  examples/sec;  0.068  sec/batch)
step 84630, loss=0.61  (1949.0  examples/sec;  0.066  sec/batch)
step 84640, loss=0.62  (1944.6  examples/sec;  0.066  sec/batch)
step 84650, loss=0.62  (1793.1  examples/sec;  0.071  sec/batch)
step 84660, loss=0.53  (1806.8  examples/sec;  0.071  sec/batch)
step 84670, loss=0.67  (1879.4  examples/sec;  0.068  sec/batch)
step 84680, loss=0.45  (1742.2  examples/sec;  0.073  sec/batch)
step 84690, loss=0.69  (1838.5  examples/sec;  0.070  sec/batch)
step 84700, loss=0.58  (1859.4  examples/sec;  0.069  sec/batch)
step 84710, loss=0.61  (1982.8  examples/sec;  0.065  sec/batch)
step 84720, loss=0.52  (1880.1  examples/sec;  0.068  sec/batch)
step 84730, loss=0.55  (1922.8  examples/sec;  0.067  sec/batch)
step 84740, loss=0.50  (1836.4  examples/sec;  0.070  sec/batch)
step 84750, loss=0.79  (1912.0  examples/sec;  0.067  sec/batch)
step 84760, loss=0.72  (1792.9  examples/sec;  0.071  sec/batch)
step 84770, loss=0.57  (1834.0  examples/sec;  0.070  sec/batch)
step 84780, loss=0.60  (1888.3  examples/sec;  0.068  sec/batch)
step 84790, loss=0.53  (1903.0  examples/sec;  0.067  sec/batch)
step 84800, loss=0.68  (1814.0  examples/sec;  0.071  sec/batch)
step 84810, loss=0.60  (1958.1  examples/sec;  0.065  sec/batch)
step 84820, loss=0.69  (1900.4  examples/sec;  0.067  sec/batch)
step 84830, loss=0.62  (1838.9  examples/sec;  0.070  sec/batch)
step 84840, loss=0.68  (1822.6  examples/sec;  0.070  sec/batch)
step 84850, loss=0.60  (1943.7  examples/sec;  0.066  sec/batch)
step 84860, loss=0.66  (1902.8  examples/sec;  0.067  sec/batch)
step 84870, loss=0.64  (1876.7  examples/sec;  0.068  sec/batch)
step 84880, loss=0.67  (1863.3  examples/sec;  0.069  sec/batch)
step 84890, loss=0.42  (1868.1  examples/sec;  0.069  sec/batch)
step 84900, loss=0.62  (1943.5  examples/sec;  0.066  sec/batch)
step 84910, loss=0.56  (1939.2  examples/sec;  0.066  sec/batch)
step 84920, loss=0.90  (1889.6  examples/sec;  0.068  sec/batch)
step 84930, loss=0.80  (1836.5  examples/sec;  0.070  sec/batch)
step 84940, loss=0.72  (1896.4  examples/sec;  0.067  sec/batch)
step 84950, loss=0.74  (1963.5  examples/sec;  0.065  sec/batch)
step 84960, loss=0.57  (1967.0  examples/sec;  0.065  sec/batch)
step 84970, loss=0.73  (1875.8  examples/sec;  0.068  sec/batch)
step 84980, loss=0.61  (1882.7  examples/sec;  0.068  sec/batch)
step 84990, loss=0.69  (1932.9  examples/sec;  0.066  sec/batch)
step 85000, loss=0.62  (1854.3  examples/sec;  0.069  sec/batch)
step 85010, loss=0.60  (1887.8  examples/sec;  0.068  sec/batch)
step 85020, loss=0.64  (1819.8  examples/sec;  0.070  sec/batch)
step 85030, loss=0.71  (1861.2  examples/sec;  0.069  sec/batch)
step 85040, loss=0.84  (1873.8  examples/sec;  0.068  sec/batch)
step 85050, loss=0.67  (1683.4  examples/sec;  0.076  sec/batch)
step 85060, loss=0.56  (1906.9  examples/sec;  0.067  sec/batch)
step 85070, loss=0.46  (1850.5  examples/sec;  0.069  sec/batch)
step 85080, loss=0.53  (1891.6  examples/sec;  0.068  sec/batch)
step 85090, loss=0.64  (1947.7  examples/sec;  0.066  sec/batch)
step 85100, loss=0.57  (1868.6  examples/sec;  0.069  sec/batch)
step 85110, loss=0.63  (1899.2  examples/sec;  0.067  sec/batch)
step 85120, loss=0.52  (1765.9  examples/sec;  0.072  sec/batch)
step 85130, loss=0.52  (1858.5  examples/sec;  0.069  sec/batch)
step 85140, loss=0.56  (1900.8  examples/sec;  0.067  sec/batch)
step 85150, loss=0.56  (1940.0  examples/sec;  0.066  sec/batch)
step 85160, loss=0.47  (1876.9  examples/sec;  0.068  sec/batch)
step 85170, loss=0.50  (1905.4  examples/sec;  0.067  sec/batch)
step 85180, loss=0.48  (1866.1  examples/sec;  0.069  sec/batch)
step 85190, loss=0.58  (1869.5  examples/sec;  0.068  sec/batch)
step 85200, loss=0.58  (1964.5  examples/sec;  0.065  sec/batch)
step 85210, loss=0.71  (1895.5  examples/sec;  0.068  sec/batch)
step 85220, loss=0.36  (1928.4  examples/sec;  0.066  sec/batch)
step 85230, loss=0.62  (1903.9  examples/sec;  0.067  sec/batch)
step 85240, loss=0.63  (1887.1  examples/sec;  0.068  sec/batch)
step 85250, loss=0.64  (1885.6  examples/sec;  0.068  sec/batch)
step 85260, loss=0.56  (1808.8  examples/sec;  0.071  sec/batch)
step 85270, loss=0.47  (1893.6  examples/sec;  0.068  sec/batch)
step 85280, loss=0.54  (1864.1  examples/sec;  0.069  sec/batch)
step 85290, loss=0.52  (1957.1  examples/sec;  0.065  sec/batch)
step 85300, loss=0.61  (1936.0  examples/sec;  0.066  sec/batch)
step 85310, loss=0.61  (1877.9  examples/sec;  0.068  sec/batch)
step 85320, loss=0.59  (1948.0  examples/sec;  0.066  sec/batch)
step 85330, loss=0.46  (1919.4  examples/sec;  0.067  sec/batch)
step 85340, loss=0.60  (1853.5  examples/sec;  0.069  sec/batch)
step 85350, loss=0.72  (1958.4  examples/sec;  0.065  sec/batch)
step 85360, loss=0.51  (1938.8  examples/sec;  0.066  sec/batch)
step 85370, loss=0.53  (1835.2  examples/sec;  0.070  sec/batch)
step 85380, loss=0.64  (1964.4  examples/sec;  0.065  sec/batch)
step 85390, loss=0.67  (1874.7  examples/sec;  0.068  sec/batch)
step 85400, loss=0.55  (1846.6  examples/sec;  0.069  sec/batch)
step 85410, loss=0.55  (1940.2  examples/sec;  0.066  sec/batch)
step 85420, loss=0.45  (1952.4  examples/sec;  0.066  sec/batch)
step 85430, loss=0.68  (1902.9  examples/sec;  0.067  sec/batch)
step 85440, loss=0.47  (1801.3  examples/sec;  0.071  sec/batch)
step 85450, loss=0.59  (1874.0  examples/sec;  0.068  sec/batch)
step 85460, loss=0.66  (1883.4  examples/sec;  0.068  sec/batch)
step 85470, loss=0.69  (1893.3  examples/sec;  0.068  sec/batch)
step 85480, loss=0.73  (1879.4  examples/sec;  0.068  sec/batch)
step 85490, loss=0.62  (1993.3  examples/sec;  0.064  sec/batch)
step 85500, loss=0.54  (1836.9  examples/sec;  0.070  sec/batch)
step 85510, loss=0.50  (1852.2  examples/sec;  0.069  sec/batch)
step 85520, loss=0.72  (1867.8  examples/sec;  0.069  sec/batch)
step 85530, loss=0.55  (1897.4  examples/sec;  0.067  sec/batch)
step 85540, loss=0.75  (1947.9  examples/sec;  0.066  sec/batch)
step 85550, loss=0.57  (1836.0  examples/sec;  0.070  sec/batch)
step 85560, loss=0.60  (1964.8  examples/sec;  0.065  sec/batch)
step 85570, loss=0.62  (1909.5  examples/sec;  0.067  sec/batch)
step 85580, loss=0.59  (1821.0  examples/sec;  0.070  sec/batch)
step 85590, loss=0.54  (1903.5  examples/sec;  0.067  sec/batch)
step 85600, loss=0.66  (1975.6  examples/sec;  0.065  sec/batch)
step 85610, loss=0.69  (1954.6  examples/sec;  0.065  sec/batch)
step 85620, loss=0.43  (1919.4  examples/sec;  0.067  sec/batch)
step 85630, loss=0.60  (1861.0  examples/sec;  0.069  sec/batch)
step 85640, loss=0.50  (1810.9  examples/sec;  0.071  sec/batch)
step 85650, loss=0.64  (1618.7  examples/sec;  0.079  sec/batch)
step 85660, loss=0.54  (1899.4  examples/sec;  0.067  sec/batch)
step 85670, loss=0.57  (1953.6  examples/sec;  0.066  sec/batch)
step 85680, loss=0.62  (1917.8  examples/sec;  0.067  sec/batch)
step 85690, loss=0.62  (1955.1  examples/sec;  0.065  sec/batch)
step 85700, loss=0.51  (1881.7  examples/sec;  0.068  sec/batch)
step 85710, loss=0.57  (1885.9  examples/sec;  0.068  sec/batch)
step 85720, loss=0.61  (1921.8  examples/sec;  0.067  sec/batch)
step 85730, loss=0.67  (1895.3  examples/sec;  0.068  sec/batch)
step 85740, loss=0.64  (1891.9  examples/sec;  0.068  sec/batch)
step 85750, loss=0.58  (1904.2  examples/sec;  0.067  sec/batch)
step 85760, loss=0.48  (1900.1  examples/sec;  0.067  sec/batch)
step 85770, loss=0.60  (2040.3  examples/sec;  0.063  sec/batch)
step 85780, loss=0.64  (1997.8  examples/sec;  0.064  sec/batch)
step 85790, loss=0.69  (1995.4  examples/sec;  0.064  sec/batch)
step 85800, loss=0.63  (1983.9  examples/sec;  0.065  sec/batch)
step 85810, loss=0.71  (1897.3  examples/sec;  0.067  sec/batch)
step 85820, loss=0.59  (1832.8  examples/sec;  0.070  sec/batch)
step 85830, loss=0.65  (1994.5  examples/sec;  0.064  sec/batch)
step 85840, loss=0.63  (1878.2  examples/sec;  0.068  sec/batch)
step 85850, loss=0.45  (1956.6  examples/sec;  0.065  sec/batch)
step 85860, loss=0.61  (1905.5  examples/sec;  0.067  sec/batch)
step 85870, loss=0.57  (1823.8  examples/sec;  0.070  sec/batch)
step 85880, loss=0.76  (1951.4  examples/sec;  0.066  sec/batch)
step 85890, loss=0.47  (1893.2  examples/sec;  0.068  sec/batch)
step 85900, loss=0.68  (1760.4  examples/sec;  0.073  sec/batch)
step 85910, loss=0.62  (1676.5  examples/sec;  0.076  sec/batch)
step 85920, loss=0.69  (1997.2  examples/sec;  0.064  sec/batch)
step 85930, loss=0.61  (1894.9  examples/sec;  0.068  sec/batch)
step 85940, loss=0.72  (1976.3  examples/sec;  0.065  sec/batch)
step 85950, loss=0.44  (1831.1  examples/sec;  0.070  sec/batch)
step 85960, loss=0.63  (1927.8  examples/sec;  0.066  sec/batch)
step 85970, loss=0.52  (1960.7  examples/sec;  0.065  sec/batch)
step 85980, loss=0.73  (1751.2  examples/sec;  0.073  sec/batch)
step 85990, loss=0.66  (1933.2  examples/sec;  0.066  sec/batch)
step 86000, loss=0.50  (1963.6  examples/sec;  0.065  sec/batch)
step 86010, loss=0.60  (1907.1  examples/sec;  0.067  sec/batch)
step 86020, loss=0.48  (1913.5  examples/sec;  0.067  sec/batch)
step 86030, loss=0.60  (1921.1  examples/sec;  0.067  sec/batch)
step 86040, loss=0.64  (1844.8  examples/sec;  0.069  sec/batch)
step 86050, loss=0.71  (1911.3  examples/sec;  0.067  sec/batch)
step 86060, loss=0.50  (1886.4  examples/sec;  0.068  sec/batch)
step 86070, loss=0.80  (1952.5  examples/sec;  0.066  sec/batch)
step 86080, loss=0.65  (1876.1  examples/sec;  0.068  sec/batch)
step 86090, loss=0.60  (2103.5  examples/sec;  0.061  sec/batch)
step 86100, loss=0.53  (1876.3  examples/sec;  0.068  sec/batch)
step 86110, loss=0.61  (1910.7  examples/sec;  0.067  sec/batch)
step 86120, loss=0.66  (1970.0  examples/sec;  0.065  sec/batch)
step 86130, loss=0.84  (1872.3  examples/sec;  0.068  sec/batch)
step 86140, loss=0.65  (1591.6  examples/sec;  0.080  sec/batch)
step 86150, loss=0.60  (1862.4  examples/sec;  0.069  sec/batch)
step 86160, loss=0.55  (1887.4  examples/sec;  0.068  sec/batch)
step 86170, loss=0.59  (1810.9  examples/sec;  0.071  sec/batch)
step 86180, loss=0.60  (1809.1  examples/sec;  0.071  sec/batch)
step 86190, loss=0.63  (1882.9  examples/sec;  0.068  sec/batch)
step 86200, loss=0.55  (1963.3  examples/sec;  0.065  sec/batch)
step 86210, loss=0.76  (1889.3  examples/sec;  0.068  sec/batch)
step 86220, loss=0.49  (1891.9  examples/sec;  0.068  sec/batch)
step 86230, loss=0.67  (1806.5  examples/sec;  0.071  sec/batch)
step 86240, loss=0.59  (1837.7  examples/sec;  0.070  sec/batch)
step 86250, loss=0.61  (1900.0  examples/sec;  0.067  sec/batch)
step 86260, loss=0.56  (1890.5  examples/sec;  0.068  sec/batch)
step 86270, loss=0.46  (1965.3  examples/sec;  0.065  sec/batch)
step 86280, loss=0.63  (1840.8  examples/sec;  0.070  sec/batch)
step 86290, loss=0.65  (1865.6  examples/sec;  0.069  sec/batch)
step 86300, loss=0.70  (1927.2  examples/sec;  0.066  sec/batch)
step 86310, loss=0.65  (1892.4  examples/sec;  0.068  sec/batch)
step 86320, loss=0.58  (1877.3  examples/sec;  0.068  sec/batch)
step 86330, loss=0.57  (1978.9  examples/sec;  0.065  sec/batch)
step 86340, loss=0.61  (1802.4  examples/sec;  0.071  sec/batch)
step 86350, loss=0.63  (1870.4  examples/sec;  0.068  sec/batch)
step 86360, loss=0.59  (1876.0  examples/sec;  0.068  sec/batch)
step 86370, loss=0.52  (1841.7  examples/sec;  0.070  sec/batch)
step 86380, loss=0.51  (1934.4  examples/sec;  0.066  sec/batch)
step 86390, loss=0.65  (1918.9  examples/sec;  0.067  sec/batch)
step 86400, loss=0.80  (1928.1  examples/sec;  0.066  sec/batch)
step 86410, loss=0.54  (1846.2  examples/sec;  0.069  sec/batch)
step 86420, loss=0.51  (1809.3  examples/sec;  0.071  sec/batch)
step 86430, loss=0.59  (1883.0  examples/sec;  0.068  sec/batch)
step 86440, loss=0.58  (1914.4  examples/sec;  0.067  sec/batch)
step 86450, loss=0.63  (1859.6  examples/sec;  0.069  sec/batch)
step 86460, loss=0.65  (1764.1  examples/sec;  0.073  sec/batch)
step 86470, loss=0.62  (1951.8  examples/sec;  0.066  sec/batch)
step 86480, loss=0.65  (1923.1  examples/sec;  0.067  sec/batch)
step 86490, loss=0.63  (1910.9  examples/sec;  0.067  sec/batch)
step 86500, loss=0.64  (1988.6  examples/sec;  0.064  sec/batch)
step 86510, loss=0.48  (1913.8  examples/sec;  0.067  sec/batch)
step 86520, loss=0.50  (1967.1  examples/sec;  0.065  sec/batch)
step 86530, loss=0.50  (1911.9  examples/sec;  0.067  sec/batch)
step 86540, loss=0.50  (2006.8  examples/sec;  0.064  sec/batch)
step 86550, loss=0.80  (1999.4  examples/sec;  0.064  sec/batch)
step 86560, loss=0.63  (1897.4  examples/sec;  0.067  sec/batch)
step 86570, loss=0.62  (1844.3  examples/sec;  0.069  sec/batch)
step 86580, loss=0.54  (1987.6  examples/sec;  0.064  sec/batch)
step 86590, loss=0.72  (1989.0  examples/sec;  0.064  sec/batch)
step 86600, loss=0.74  (1853.5  examples/sec;  0.069  sec/batch)
step 86610, loss=0.60  (1912.7  examples/sec;  0.067  sec/batch)
step 86620, loss=0.53  (1883.1  examples/sec;  0.068  sec/batch)
step 86630, loss=0.47  (1747.3  examples/sec;  0.073  sec/batch)
step 86640, loss=0.43  (1842.1  examples/sec;  0.069  sec/batch)
step 86650, loss=0.55  (1844.2  examples/sec;  0.069  sec/batch)
step 86660, loss=0.60  (1864.4  examples/sec;  0.069  sec/batch)
step 86670, loss=0.48  (1915.4  examples/sec;  0.067  sec/batch)
step 86680, loss=0.68  (1909.1  examples/sec;  0.067  sec/batch)
step 86690, loss=0.61  (1966.5  examples/sec;  0.065  sec/batch)
step 86700, loss=0.59  (1958.7  examples/sec;  0.065  sec/batch)
step 86710, loss=0.41  (1913.0  examples/sec;  0.067  sec/batch)
step 86720, loss=0.48  (1876.7  examples/sec;  0.068  sec/batch)
step 86730, loss=0.73  (1963.3  examples/sec;  0.065  sec/batch)
step 86740, loss=0.53  (1820.0  examples/sec;  0.070  sec/batch)
step 86750, loss=0.63  (1968.4  examples/sec;  0.065  sec/batch)
step 86760, loss=0.72  (1941.7  examples/sec;  0.066  sec/batch)
step 86770, loss=0.70  (1923.7  examples/sec;  0.067  sec/batch)
step 86780, loss=0.62  (1926.3  examples/sec;  0.066  sec/batch)
step 86790, loss=0.49  (1907.8  examples/sec;  0.067  sec/batch)
step 86800, loss=0.53  (1858.0  examples/sec;  0.069  sec/batch)
step 86810, loss=0.64  (1848.3  examples/sec;  0.069  sec/batch)
step 86820, loss=0.48  (1924.6  examples/sec;  0.067  sec/batch)
step 86830, loss=0.60  (1834.5  examples/sec;  0.070  sec/batch)
step 86840, loss=0.62  (1991.6  examples/sec;  0.064  sec/batch)
step 86850, loss=0.54  (1878.1  examples/sec;  0.068  sec/batch)
step 86860, loss=0.44  (1909.2  examples/sec;  0.067  sec/batch)
step 86870, loss=0.48  (1965.8  examples/sec;  0.065  sec/batch)
step 86880, loss=0.59  (1894.3  examples/sec;  0.068  sec/batch)
step 86890, loss=0.73  (1891.2  examples/sec;  0.068  sec/batch)
step 86900, loss=0.57  (1866.5  examples/sec;  0.069  sec/batch)
step 86910, loss=0.68  (1946.0  examples/sec;  0.066  sec/batch)
step 86920, loss=0.66  (1949.9  examples/sec;  0.066  sec/batch)
step 86930, loss=0.64  (1881.4  examples/sec;  0.068  sec/batch)
step 86940, loss=0.56  (1890.0  examples/sec;  0.068  sec/batch)
step 86950, loss=0.47  (2022.3  examples/sec;  0.063  sec/batch)
step 86960, loss=0.69  (1883.0  examples/sec;  0.068  sec/batch)
step 86970, loss=0.75  (1856.3  examples/sec;  0.069  sec/batch)
step 86980, loss=0.71  (1933.4  examples/sec;  0.066  sec/batch)
step 86990, loss=0.60  (1882.6  examples/sec;  0.068  sec/batch)
step 87000, loss=0.51  (1826.0  examples/sec;  0.070  sec/batch)
step 87010, loss=0.60  (1887.1  examples/sec;  0.068  sec/batch)
step 87020, loss=0.53  (1934.1  examples/sec;  0.066  sec/batch)
step 87030, loss=0.75  (1903.3  examples/sec;  0.067  sec/batch)
step 87040, loss=0.56  (1913.7  examples/sec;  0.067  sec/batch)
step 87050, loss=0.61  (1899.9  examples/sec;  0.067  sec/batch)
step 87060, loss=0.59  (1924.1  examples/sec;  0.067  sec/batch)
step 87070, loss=0.60  (1881.8  examples/sec;  0.068  sec/batch)
step 87080, loss=0.63  (1885.6  examples/sec;  0.068  sec/batch)
step 87090, loss=0.66  (1913.7  examples/sec;  0.067  sec/batch)
step 87100, loss=0.49  (1905.4  examples/sec;  0.067  sec/batch)
step 87110, loss=0.57  (1909.4  examples/sec;  0.067  sec/batch)
step 87120, loss=0.63  (1959.3  examples/sec;  0.065  sec/batch)
step 87130, loss=0.62  (1899.3  examples/sec;  0.067  sec/batch)
step 87140, loss=0.66  (1803.2  examples/sec;  0.071  sec/batch)
step 87150, loss=0.57  (1861.8  examples/sec;  0.069  sec/batch)
step 87160, loss=0.61  (1972.6  examples/sec;  0.065  sec/batch)
step 87170, loss=0.58  (1838.6  examples/sec;  0.070  sec/batch)
step 87180, loss=0.65  (1891.7  examples/sec;  0.068  sec/batch)
step 87190, loss=0.57  (1877.1  examples/sec;  0.068  sec/batch)
step 87200, loss=0.55  (1967.0  examples/sec;  0.065  sec/batch)
step 87210, loss=0.67  (1830.8  examples/sec;  0.070  sec/batch)
step 87220, loss=0.53  (1909.6  examples/sec;  0.067  sec/batch)
step 87230, loss=0.51  (1346.1  examples/sec;  0.095  sec/batch)
step 87240, loss=0.56  (1870.1  examples/sec;  0.068  sec/batch)
step 87250, loss=0.67  (1936.0  examples/sec;  0.066  sec/batch)
step 87260, loss=0.58  (1842.0  examples/sec;  0.069  sec/batch)
step 87270, loss=0.62  (1938.9  examples/sec;  0.066  sec/batch)
step 87280, loss=0.59  (1918.5  examples/sec;  0.067  sec/batch)
step 87290, loss=0.58  (1926.0  examples/sec;  0.066  sec/batch)
step 87300, loss=0.69  (1934.7  examples/sec;  0.066  sec/batch)
step 87310, loss=0.45  (1947.1  examples/sec;  0.066  sec/batch)
step 87320, loss=0.69  (1863.5  examples/sec;  0.069  sec/batch)
step 87330, loss=0.65  (1931.9  examples/sec;  0.066  sec/batch)
step 87340, loss=0.58  (1821.7  examples/sec;  0.070  sec/batch)
step 87350, loss=0.55  (1863.6  examples/sec;  0.069  sec/batch)
step 87360, loss=0.57  (1873.9  examples/sec;  0.068  sec/batch)
step 87370, loss=0.55  (1786.3  examples/sec;  0.072  sec/batch)
step 87380, loss=0.70  (1918.3  examples/sec;  0.067  sec/batch)
step 87390, loss=0.48  (1885.5  examples/sec;  0.068  sec/batch)
step 87400, loss=0.56  (1910.3  examples/sec;  0.067  sec/batch)
step 87410, loss=0.57  (1852.2  examples/sec;  0.069  sec/batch)
step 87420, loss=0.51  (1874.1  examples/sec;  0.068  sec/batch)
step 87430, loss=0.55  (1966.7  examples/sec;  0.065  sec/batch)
step 87440, loss=0.54  (1939.2  examples/sec;  0.066  sec/batch)
step 87450, loss=0.67  (1880.9  examples/sec;  0.068  sec/batch)
step 87460, loss=0.67  (1855.3  examples/sec;  0.069  sec/batch)
step 87470, loss=0.70  (1886.2  examples/sec;  0.068  sec/batch)
step 87480, loss=0.69  (1848.3  examples/sec;  0.069  sec/batch)
step 87490, loss=0.49  (1787.5  examples/sec;  0.072  sec/batch)
step 87500, loss=0.65  (1879.3  examples/sec;  0.068  sec/batch)
step 87510, loss=0.56  (1958.7  examples/sec;  0.065  sec/batch)
step 87520, loss=0.62  (1843.1  examples/sec;  0.069  sec/batch)
step 87530, loss=0.42  (1913.3  examples/sec;  0.067  sec/batch)
step 87540, loss=0.59  (1880.6  examples/sec;  0.068  sec/batch)
step 87550, loss=0.67  (1825.3  examples/sec;  0.070  sec/batch)
step 87560, loss=0.59  (1762.8  examples/sec;  0.073  sec/batch)
step 87570, loss=0.62  (1920.6  examples/sec;  0.067  sec/batch)
step 87580, loss=0.59  (1927.0  examples/sec;  0.066  sec/batch)
step 87590, loss=0.45  (1919.1  examples/sec;  0.067  sec/batch)
step 87600, loss=0.53  (1848.9  examples/sec;  0.069  sec/batch)
step 87610, loss=0.59  (1875.0  examples/sec;  0.068  sec/batch)
step 87620, loss=0.56  (1856.6  examples/sec;  0.069  sec/batch)
step 87630, loss=0.47  (1554.6  examples/sec;  0.082  sec/batch)
step 87640, loss=0.71  (1863.2  examples/sec;  0.069  sec/batch)
step 87650, loss=0.54  (1917.2  examples/sec;  0.067  sec/batch)
step 87660, loss=0.59  (1947.2  examples/sec;  0.066  sec/batch)
step 87670, loss=0.72  (1852.4  examples/sec;  0.069  sec/batch)
step 87680, loss=0.55  (1909.8  examples/sec;  0.067  sec/batch)
step 87690, loss=0.79  (1913.8  examples/sec;  0.067  sec/batch)
step 87700, loss=0.66  (1856.9  examples/sec;  0.069  sec/batch)
step 87710, loss=0.57  (1973.1  examples/sec;  0.065  sec/batch)
step 87720, loss=0.44  (1896.0  examples/sec;  0.068  sec/batch)
step 87730, loss=0.46  (1838.7  examples/sec;  0.070  sec/batch)
step 87740, loss=0.59  (1945.3  examples/sec;  0.066  sec/batch)
step 87750, loss=0.66  (1988.0  examples/sec;  0.064  sec/batch)
step 87760, loss=0.56  (1897.8  examples/sec;  0.067  sec/batch)
step 87770, loss=0.54  (1857.9  examples/sec;  0.069  sec/batch)
step 87780, loss=0.65  (1861.9  examples/sec;  0.069  sec/batch)
step 87790, loss=0.53  (1784.6  examples/sec;  0.072  sec/batch)
step 87800, loss=0.58  (1925.2  examples/sec;  0.066  sec/batch)
step 87810, loss=0.51  (1889.0  examples/sec;  0.068  sec/batch)
step 87820, loss=0.53  (1950.8  examples/sec;  0.066  sec/batch)
step 87830, loss=0.61  (1831.8  examples/sec;  0.070  sec/batch)
step 87840, loss=0.59  (1867.5  examples/sec;  0.069  sec/batch)
step 87850, loss=0.47  (1824.3  examples/sec;  0.070  sec/batch)
step 87860, loss=0.66  (1894.6  examples/sec;  0.068  sec/batch)
step 87870, loss=0.58  (1863.9  examples/sec;  0.069  sec/batch)
step 87880, loss=0.61  (1778.0  examples/sec;  0.072  sec/batch)
step 87890, loss=0.56  (1961.8  examples/sec;  0.065  sec/batch)
step 87900, loss=0.64  (1884.3  examples/sec;  0.068  sec/batch)
step 87910, loss=0.55  (1845.4  examples/sec;  0.069  sec/batch)
step 87920, loss=0.54  (1916.8  examples/sec;  0.067  sec/batch)
step 87930, loss=0.71  (1892.6  examples/sec;  0.068  sec/batch)
step 87940, loss=0.51  (1856.8  examples/sec;  0.069  sec/batch)
step 87950, loss=0.64  (1873.6  examples/sec;  0.068  sec/batch)
step 87960, loss=0.55  (1956.1  examples/sec;  0.065  sec/batch)
step 87970, loss=0.51  (1930.9  examples/sec;  0.066  sec/batch)
step 87980, loss=0.59  (1780.1  examples/sec;  0.072  sec/batch)
step 87990, loss=0.57  (1890.9  examples/sec;  0.068  sec/batch)
step 88000, loss=0.59  (1857.3  examples/sec;  0.069  sec/batch)
step 88010, loss=0.62  (1863.1  examples/sec;  0.069  sec/batch)
step 88020, loss=0.49  (1935.4  examples/sec;  0.066  sec/batch)
step 88030, loss=0.71  (1950.2  examples/sec;  0.066  sec/batch)
step 88040, loss=0.63  (1927.5  examples/sec;  0.066  sec/batch)
step 88050, loss=0.74  (1746.2  examples/sec;  0.073  sec/batch)
step 88060, loss=0.55  (1863.7  examples/sec;  0.069  sec/batch)
step 88070, loss=0.54  (1951.4  examples/sec;  0.066  sec/batch)
step 88080, loss=0.62  (1993.9  examples/sec;  0.064  sec/batch)
step 88090, loss=0.51  (1741.9  examples/sec;  0.073  sec/batch)
step 88100, loss=0.44  (1938.1  examples/sec;  0.066  sec/batch)
step 88110, loss=0.58  (1936.3  examples/sec;  0.066  sec/batch)
step 88120, loss=0.52  (1871.9  examples/sec;  0.068  sec/batch)
step 88130, loss=0.55  (1927.1  examples/sec;  0.066  sec/batch)
step 88140, loss=0.67  (1758.6  examples/sec;  0.073  sec/batch)
step 88150, loss=0.51  (1894.8  examples/sec;  0.068  sec/batch)
step 88160, loss=0.58  (1995.5  examples/sec;  0.064  sec/batch)
step 88170, loss=0.55  (2003.8  examples/sec;  0.064  sec/batch)
step 88180, loss=0.71  (1849.9  examples/sec;  0.069  sec/batch)
step 88190, loss=0.56  (1894.6  examples/sec;  0.068  sec/batch)
step 88200, loss=0.55  (1887.2  examples/sec;  0.068  sec/batch)
step 88210, loss=0.61  (1848.2  examples/sec;  0.069  sec/batch)
step 88220, loss=0.44  (1870.9  examples/sec;  0.068  sec/batch)
step 88230, loss=0.70  (1869.7  examples/sec;  0.068  sec/batch)
step 88240, loss=0.44  (1859.5  examples/sec;  0.069  sec/batch)
step 88250, loss=0.58  (1886.0  examples/sec;  0.068  sec/batch)
step 88260, loss=0.61  (1827.3  examples/sec;  0.070  sec/batch)
step 88270, loss=0.63  (1800.9  examples/sec;  0.071  sec/batch)
step 88280, loss=0.69  (1961.5  examples/sec;  0.065  sec/batch)
step 88290, loss=0.66  (1887.0  examples/sec;  0.068  sec/batch)
step 88300, loss=0.59  (1977.4  examples/sec;  0.065  sec/batch)
step 88310, loss=0.52  (1932.2  examples/sec;  0.066  sec/batch)
step 88320, loss=0.66  (1913.8  examples/sec;  0.067  sec/batch)
step 88330, loss=0.48  (1883.6  examples/sec;  0.068  sec/batch)
step 88340, loss=0.63  (1917.2  examples/sec;  0.067  sec/batch)
step 88350, loss=0.57  (1883.2  examples/sec;  0.068  sec/batch)
step 88360, loss=0.57  (1926.1  examples/sec;  0.066  sec/batch)
step 88370, loss=0.52  (1885.5  examples/sec;  0.068  sec/batch)
step 88380, loss=0.75  (1934.1  examples/sec;  0.066  sec/batch)
step 88390, loss=0.54  (1959.9  examples/sec;  0.065  sec/batch)
step 88400, loss=0.52  (1871.3  examples/sec;  0.068  sec/batch)
step 88410, loss=0.61  (1927.5  examples/sec;  0.066  sec/batch)
step 88420, loss=0.63  (1933.4  examples/sec;  0.066  sec/batch)
step 88430, loss=0.72  (1923.4  examples/sec;  0.067  sec/batch)
step 88440, loss=0.52  (1870.9  examples/sec;  0.068  sec/batch)
step 88450, loss=0.51  (1893.9  examples/sec;  0.068  sec/batch)
step 88460, loss=0.61  (1972.6  examples/sec;  0.065  sec/batch)
step 88470, loss=0.73  (1735.5  examples/sec;  0.074  sec/batch)
step 88480, loss=0.54  (1907.2  examples/sec;  0.067  sec/batch)
step 88490, loss=0.58  (1941.5  examples/sec;  0.066  sec/batch)
step 88500, loss=0.54  (1931.4  examples/sec;  0.066  sec/batch)
step 88510, loss=0.50  (2015.5  examples/sec;  0.064  sec/batch)
step 88520, loss=0.63  (1855.0  examples/sec;  0.069  sec/batch)
step 88530, loss=0.44  (1791.9  examples/sec;  0.071  sec/batch)
step 88540, loss=0.48  (1955.0  examples/sec;  0.065  sec/batch)
step 88550, loss=0.56  (1845.7  examples/sec;  0.069  sec/batch)
step 88560, loss=0.54  (1877.7  examples/sec;  0.068  sec/batch)
step 88570, loss=0.80  (1997.1  examples/sec;  0.064  sec/batch)
step 88580, loss=0.78  (1911.5  examples/sec;  0.067  sec/batch)
step 88590, loss=0.60  (1920.6  examples/sec;  0.067  sec/batch)
step 88600, loss=0.71  (1813.8  examples/sec;  0.071  sec/batch)
step 88610, loss=0.64  (1896.3  examples/sec;  0.067  sec/batch)
step 88620, loss=0.59  (1863.4  examples/sec;  0.069  sec/batch)
step 88630, loss=0.67  (1975.3  examples/sec;  0.065  sec/batch)
step 88640, loss=0.49  (1886.9  examples/sec;  0.068  sec/batch)
step 88650, loss=0.52  (1836.9  examples/sec;  0.070  sec/batch)
step 88660, loss=0.51  (1933.2  examples/sec;  0.066  sec/batch)
step 88670, loss=0.60  (1927.0  examples/sec;  0.066  sec/batch)
step 88680, loss=0.63  (2031.3  examples/sec;  0.063  sec/batch)
step 88690, loss=0.62  (1903.0  examples/sec;  0.067  sec/batch)
step 88700, loss=0.73  (1949.8  examples/sec;  0.066  sec/batch)
step 88710, loss=0.84  (1873.6  examples/sec;  0.068  sec/batch)
step 88720, loss=0.57  (1938.3  examples/sec;  0.066  sec/batch)
step 88730, loss=0.57  (1890.8  examples/sec;  0.068  sec/batch)
step 88740, loss=0.63  (1890.3  examples/sec;  0.068  sec/batch)
step 88750, loss=0.50  (2034.4  examples/sec;  0.063  sec/batch)
step 88760, loss=0.68  (1971.4  examples/sec;  0.065  sec/batch)
step 88770, loss=0.65  (1854.0  examples/sec;  0.069  sec/batch)
step 88780, loss=0.50  (1874.1  examples/sec;  0.068  sec/batch)
step 88790, loss=0.67  (2040.4  examples/sec;  0.063  sec/batch)
step 88800, loss=0.57  (1887.1  examples/sec;  0.068  sec/batch)
step 88810, loss=0.64  (1766.0  examples/sec;  0.072  sec/batch)
step 88820, loss=0.65  (1798.6  examples/sec;  0.071  sec/batch)
step 88830, loss=0.87  (1937.0  examples/sec;  0.066  sec/batch)
step 88840, loss=0.60  (1837.5  examples/sec;  0.070  sec/batch)
step 88850, loss=0.65  (1924.0  examples/sec;  0.067  sec/batch)
step 88860, loss=0.72  (1949.5  examples/sec;  0.066  sec/batch)
step 88870, loss=0.56  (1814.4  examples/sec;  0.071  sec/batch)
step 88880, loss=0.56  (1919.5  examples/sec;  0.067  sec/batch)
step 88890, loss=0.49  (1815.6  examples/sec;  0.071  sec/batch)
step 88900, loss=0.59  (1850.4  examples/sec;  0.069  sec/batch)
step 88910, loss=0.45  (1640.6  examples/sec;  0.078  sec/batch)
step 88920, loss=0.51  (1727.8  examples/sec;  0.074  sec/batch)
step 88930, loss=0.65  (1939.3  examples/sec;  0.066  sec/batch)
step 88940, loss=0.69  (1963.9  examples/sec;  0.065  sec/batch)
step 88950, loss=0.61  (1918.0  examples/sec;  0.067  sec/batch)
step 88960, loss=0.62  (1847.4  examples/sec;  0.069  sec/batch)
step 88970, loss=0.47  (1655.0  examples/sec;  0.077  sec/batch)
step 88980, loss=0.52  (1839.7  examples/sec;  0.070  sec/batch)
step 88990, loss=0.56  (1947.1  examples/sec;  0.066  sec/batch)
step 89000, loss=0.48  (1845.4  examples/sec;  0.069  sec/batch)
step 89010, loss=0.54  (1955.3  examples/sec;  0.065  sec/batch)
step 89020, loss=0.43  (1908.0  examples/sec;  0.067  sec/batch)
step 89030, loss=0.79  (1934.9  examples/sec;  0.066  sec/batch)
step 89040, loss=0.67  (1859.1  examples/sec;  0.069  sec/batch)
step 89050, loss=0.76  (1911.6  examples/sec;  0.067  sec/batch)
step 89060, loss=0.71  (1883.5  examples/sec;  0.068  sec/batch)
step 89070, loss=0.50  (2019.0  examples/sec;  0.063  sec/batch)
step 89080, loss=0.53  (1893.6  examples/sec;  0.068  sec/batch)
step 89090, loss=0.48  (1835.5  examples/sec;  0.070  sec/batch)
step 89100, loss=0.55  (1965.2  examples/sec;  0.065  sec/batch)
step 89110, loss=0.55  (1873.6  examples/sec;  0.068  sec/batch)
step 89120, loss=0.58  (1908.2  examples/sec;  0.067  sec/batch)
step 89130, loss=0.40  (1839.3  examples/sec;  0.070  sec/batch)
step 89140, loss=0.61  (1917.5  examples/sec;  0.067  sec/batch)
step 89150, loss=0.59  (1881.2  examples/sec;  0.068  sec/batch)
step 89160, loss=0.61  (1714.6  examples/sec;  0.075  sec/batch)
step 89170, loss=0.62  (1919.7  examples/sec;  0.067  sec/batch)
step 89180, loss=0.72  (1878.8  examples/sec;  0.068  sec/batch)
step 89190, loss=0.59  (1596.8  examples/sec;  0.080  sec/batch)
step 89200, loss=0.69  (1922.0  examples/sec;  0.067  sec/batch)
step 89210, loss=0.58  (1863.3  examples/sec;  0.069  sec/batch)
step 89220, loss=0.50  (1833.5  examples/sec;  0.070  sec/batch)
step 89230, loss=0.59  (1921.2  examples/sec;  0.067  sec/batch)
step 89240, loss=0.56  (1905.6  examples/sec;  0.067  sec/batch)
step 89250, loss=0.66  (1864.0  examples/sec;  0.069  sec/batch)
step 89260, loss=0.57  (1898.0  examples/sec;  0.067  sec/batch)
step 89270, loss=0.63  (1886.9  examples/sec;  0.068  sec/batch)
step 89280, loss=0.50  (1824.2  examples/sec;  0.070  sec/batch)
step 89290, loss=0.54  (1969.1  examples/sec;  0.065  sec/batch)
step 89300, loss=0.63  (1910.1  examples/sec;  0.067  sec/batch)
step 89310, loss=0.82  (1889.0  examples/sec;  0.068  sec/batch)
step 89320, loss=0.80  (1912.6  examples/sec;  0.067  sec/batch)
step 89330, loss=0.50  (1889.0  examples/sec;  0.068  sec/batch)
step 89340, loss=0.65  (1939.1  examples/sec;  0.066  sec/batch)
step 89350, loss=0.57  (1765.2  examples/sec;  0.073  sec/batch)
step 89360, loss=0.56  (1884.3  examples/sec;  0.068  sec/batch)
step 89370, loss=0.46  (1901.8  examples/sec;  0.067  sec/batch)
step 89380, loss=0.70  (1859.8  examples/sec;  0.069  sec/batch)
step 89390, loss=0.72  (1895.3  examples/sec;  0.068  sec/batch)
step 89400, loss=0.66  (1951.2  examples/sec;  0.066  sec/batch)
step 89410, loss=0.59  (1777.5  examples/sec;  0.072  sec/batch)
step 89420, loss=0.45  (1852.2  examples/sec;  0.069  sec/batch)
step 89430, loss=0.57  (1899.8  examples/sec;  0.067  sec/batch)
step 89440, loss=0.49  (2027.2  examples/sec;  0.063  sec/batch)
step 89450, loss=0.66  (1870.9  examples/sec;  0.068  sec/batch)
step 89460, loss=0.69  (1900.7  examples/sec;  0.067  sec/batch)
step 89470, loss=0.63  (1906.9  examples/sec;  0.067  sec/batch)
step 89480, loss=0.70  (1884.6  examples/sec;  0.068  sec/batch)
step 89490, loss=0.68  (1871.7  examples/sec;  0.068  sec/batch)
step 89500, loss=0.57  (1845.7  examples/sec;  0.069  sec/batch)
step 89510, loss=0.75  (1892.7  examples/sec;  0.068  sec/batch)
step 89520, loss=0.51  (1880.2  examples/sec;  0.068  sec/batch)
step 89530, loss=0.52  (1914.3  examples/sec;  0.067  sec/batch)
step 89540, loss=0.61  (1856.2  examples/sec;  0.069  sec/batch)
step 89550, loss=0.60  (1924.8  examples/sec;  0.066  sec/batch)
step 89560, loss=0.69  (1938.3  examples/sec;  0.066  sec/batch)
step 89570, loss=0.64  (1915.1  examples/sec;  0.067  sec/batch)
step 89580, loss=0.51  (1891.4  examples/sec;  0.068  sec/batch)
step 89590, loss=0.59  (1895.4  examples/sec;  0.068  sec/batch)
step 89600, loss=0.60  (1961.1  examples/sec;  0.065  sec/batch)
step 89610, loss=0.57  (1887.0  examples/sec;  0.068  sec/batch)
step 89620, loss=0.48  (1844.2  examples/sec;  0.069  sec/batch)
step 89630, loss=0.73  (1861.6  examples/sec;  0.069  sec/batch)
step 89640, loss=0.50  (1935.1  examples/sec;  0.066  sec/batch)
step 89650, loss=0.68  (1874.2  examples/sec;  0.068  sec/batch)
step 89660, loss=0.62  (1997.4  examples/sec;  0.064  sec/batch)
step 89670, loss=0.61  (1896.3  examples/sec;  0.067  sec/batch)
step 89680, loss=0.50  (1798.1  examples/sec;  0.071  sec/batch)
step 89690, loss=0.53  (1981.3  examples/sec;  0.065  sec/batch)
step 89700, loss=0.57  (1889.6  examples/sec;  0.068  sec/batch)
step 89710, loss=0.66  (1814.3  examples/sec;  0.071  sec/batch)
step 89720, loss=0.62  (1853.1  examples/sec;  0.069  sec/batch)
step 89730, loss=0.60  (1900.0  examples/sec;  0.067  sec/batch)
step 89740, loss=0.54  (1899.6  examples/sec;  0.067  sec/batch)
step 89750, loss=0.73  (1868.8  examples/sec;  0.068  sec/batch)
step 89760, loss=0.65  (2009.8  examples/sec;  0.064  sec/batch)
step 89770, loss=0.62  (1868.1  examples/sec;  0.069  sec/batch)
step 89780, loss=0.61  (1912.5  examples/sec;  0.067  sec/batch)
step 89790, loss=0.66  (1836.1  examples/sec;  0.070  sec/batch)
step 89800, loss=0.65  (1673.0  examples/sec;  0.077  sec/batch)
step 89810, loss=0.76  (1978.4  examples/sec;  0.065  sec/batch)
step 89820, loss=0.62  (1956.3  examples/sec;  0.065  sec/batch)
step 89830, loss=0.56  (1890.4  examples/sec;  0.068  sec/batch)
step 89840, loss=0.65  (1849.8  examples/sec;  0.069  sec/batch)
step 89850, loss=0.54  (1880.9  examples/sec;  0.068  sec/batch)
step 89860, loss=0.70  (1862.2  examples/sec;  0.069  sec/batch)
step 89870, loss=0.54  (1937.7  examples/sec;  0.066  sec/batch)
step 89880, loss=0.67  (1755.2  examples/sec;  0.073  sec/batch)
step 89890, loss=0.60  (1736.0  examples/sec;  0.074  sec/batch)
step 89900, loss=0.58  (1896.1  examples/sec;  0.068  sec/batch)
step 89910, loss=0.57  (1887.6  examples/sec;  0.068  sec/batch)
step 89920, loss=0.49  (1772.4  examples/sec;  0.072  sec/batch)
step 89930, loss=0.67  (1980.5  examples/sec;  0.065  sec/batch)
step 89940, loss=0.61  (1913.0  examples/sec;  0.067  sec/batch)
step 89950, loss=0.43  (1878.7  examples/sec;  0.068  sec/batch)
step 89960, loss=0.71  (1861.2  examples/sec;  0.069  sec/batch)
step 89970, loss=0.69  (1972.2  examples/sec;  0.065  sec/batch)
step 89980, loss=0.66  (1917.5  examples/sec;  0.067  sec/batch)
step 89990, loss=0.52  (1985.7  examples/sec;  0.064  sec/batch)
step 90000, loss=0.58  (1695.7  examples/sec;  0.075  sec/batch)
step 90010, loss=0.64  (1923.6  examples/sec;  0.067  sec/batch)
step 90020, loss=0.51  (1930.4  examples/sec;  0.066  sec/batch)
step 90030, loss=0.51  (1934.2  examples/sec;  0.066  sec/batch)
step 90040, loss=0.57  (1957.5  examples/sec;  0.065  sec/batch)
step 90050, loss=0.63  (2015.9  examples/sec;  0.063  sec/batch)
step 90060, loss=0.64  (1984.8  examples/sec;  0.064  sec/batch)
step 90070, loss=0.55  (1989.0  examples/sec;  0.064  sec/batch)
step 90080, loss=0.56  (1842.7  examples/sec;  0.069  sec/batch)
step 90090, loss=0.65  (1859.3  examples/sec;  0.069  sec/batch)
step 90100, loss=0.60  (1829.4  examples/sec;  0.070  sec/batch)
step 90110, loss=0.64  (1870.5  examples/sec;  0.068  sec/batch)
step 90120, loss=0.48  (1927.7  examples/sec;  0.066  sec/batch)
step 90130, loss=0.56  (1890.9  examples/sec;  0.068  sec/batch)
step 90140, loss=0.71  (1855.4  examples/sec;  0.069  sec/batch)
step 90150, loss=0.73  (1945.0  examples/sec;  0.066  sec/batch)
step 90160, loss=0.69  (1926.5  examples/sec;  0.066  sec/batch)
step 90170, loss=0.55  (1945.4  examples/sec;  0.066  sec/batch)
step 90180, loss=0.60  (1890.3  examples/sec;  0.068  sec/batch)
step 90190, loss=0.61  (1692.1  examples/sec;  0.076  sec/batch)
step 90200, loss=0.64  (1883.7  examples/sec;  0.068  sec/batch)
step 90210, loss=0.50  (1880.3  examples/sec;  0.068  sec/batch)
step 90220, loss=0.49  (1859.1  examples/sec;  0.069  sec/batch)
step 90230, loss=0.69  (1841.5  examples/sec;  0.070  sec/batch)
step 90240, loss=0.61  (1898.2  examples/sec;  0.067  sec/batch)
step 90250, loss=0.60  (1944.3  examples/sec;  0.066  sec/batch)
step 90260, loss=0.55  (1650.8  examples/sec;  0.078  sec/batch)
step 90270, loss=0.63  (1886.7  examples/sec;  0.068  sec/batch)
step 90280, loss=0.58  (1866.7  examples/sec;  0.069  sec/batch)
step 90290, loss=0.60  (1886.5  examples/sec;  0.068  sec/batch)
step 90300, loss=0.55  (1899.3  examples/sec;  0.067  sec/batch)
step 90310, loss=0.73  (1833.7  examples/sec;  0.070  sec/batch)
step 90320, loss=0.51  (1848.4  examples/sec;  0.069  sec/batch)
step 90330, loss=0.71  (1841.7  examples/sec;  0.070  sec/batch)
step 90340, loss=0.64  (1870.1  examples/sec;  0.068  sec/batch)
step 90350, loss=0.59  (1864.6  examples/sec;  0.069  sec/batch)
step 90360, loss=0.54  (1927.2  examples/sec;  0.066  sec/batch)
step 90370, loss=0.53  (1835.8  examples/sec;  0.070  sec/batch)
step 90380, loss=0.64  (1923.3  examples/sec;  0.067  sec/batch)
step 90390, loss=0.47  (1886.5  examples/sec;  0.068  sec/batch)
step 90400, loss=0.62  (1899.1  examples/sec;  0.067  sec/batch)
step 90410, loss=0.60  (1814.9  examples/sec;  0.071  sec/batch)
step 90420, loss=0.58  (1919.3  examples/sec;  0.067  sec/batch)
step 90430, loss=0.61  (1811.4  examples/sec;  0.071  sec/batch)
step 90440, loss=0.75  (1898.5  examples/sec;  0.067  sec/batch)
step 90450, loss=0.74  (1902.1  examples/sec;  0.067  sec/batch)
step 90460, loss=0.65  (1980.5  examples/sec;  0.065  sec/batch)
step 90470, loss=0.57  (1865.4  examples/sec;  0.069  sec/batch)
step 90480, loss=0.74  (1866.7  examples/sec;  0.069  sec/batch)
step 90490, loss=0.51  (1934.2  examples/sec;  0.066  sec/batch)
step 90500, loss=0.52  (1945.6  examples/sec;  0.066  sec/batch)
step 90510, loss=0.51  (1844.0  examples/sec;  0.069  sec/batch)
step 90520, loss=0.50  (2023.1  examples/sec;  0.063  sec/batch)
step 90530, loss=0.66  (1900.3  examples/sec;  0.067  sec/batch)
step 90540, loss=0.43  (1884.0  examples/sec;  0.068  sec/batch)
step 90550, loss=0.58  (1987.7  examples/sec;  0.064  sec/batch)
step 90560, loss=0.58  (1934.5  examples/sec;  0.066  sec/batch)
step 90570, loss=0.69  (1780.7  examples/sec;  0.072  sec/batch)
step 90580, loss=0.62  (1829.1  examples/sec;  0.070  sec/batch)
step 90590, loss=0.60  (1868.8  examples/sec;  0.068  sec/batch)
step 90600, loss=0.71  (1822.9  examples/sec;  0.070  sec/batch)
step 90610, loss=0.42  (1908.8  examples/sec;  0.067  sec/batch)
step 90620, loss=0.61  (1829.9  examples/sec;  0.070  sec/batch)
step 90630, loss=0.58  (1982.8  examples/sec;  0.065  sec/batch)
step 90640, loss=0.59  (1904.2  examples/sec;  0.067  sec/batch)
step 90650, loss=0.60  (1945.1  examples/sec;  0.066  sec/batch)
step 90660, loss=0.53  (1866.5  examples/sec;  0.069  sec/batch)
step 90670, loss=0.62  (1877.7  examples/sec;  0.068  sec/batch)
step 90680, loss=0.53  (1906.7  examples/sec;  0.067  sec/batch)
step 90690, loss=0.57  (1972.6  examples/sec;  0.065  sec/batch)
step 90700, loss=0.63  (1920.3  examples/sec;  0.067  sec/batch)
step 90710, loss=0.45  (1780.9  examples/sec;  0.072  sec/batch)
step 90720, loss=0.67  (1897.3  examples/sec;  0.067  sec/batch)
step 90730, loss=0.65  (1937.5  examples/sec;  0.066  sec/batch)
step 90740, loss=0.59  (1878.5  examples/sec;  0.068  sec/batch)
step 90750, loss=0.71  (1944.3  examples/sec;  0.066  sec/batch)
step 90760, loss=0.61  (1832.6  examples/sec;  0.070  sec/batch)
step 90770, loss=0.67  (1819.5  examples/sec;  0.070  sec/batch)
step 90780, loss=0.56  (1910.6  examples/sec;  0.067  sec/batch)
step 90790, loss=0.62  (1980.0  examples/sec;  0.065  sec/batch)
step 90800, loss=0.81  (1772.0  examples/sec;  0.072  sec/batch)
step 90810, loss=0.49  (1895.8  examples/sec;  0.068  sec/batch)
step 90820, loss=0.61  (1659.0  examples/sec;  0.077  sec/batch)
step 90830, loss=0.46  (1825.9  examples/sec;  0.070  sec/batch)
step 90840, loss=0.57  (1823.4  examples/sec;  0.070  sec/batch)
step 90850, loss=0.47  (1974.0  examples/sec;  0.065  sec/batch)
step 90860, loss=0.70  (1862.6  examples/sec;  0.069  sec/batch)
step 90870, loss=0.61  (1860.8  examples/sec;  0.069  sec/batch)
step 90880, loss=0.64  (1905.6  examples/sec;  0.067  sec/batch)
step 90890, loss=0.59  (1809.8  examples/sec;  0.071  sec/batch)
step 90900, loss=0.66  (1887.5  examples/sec;  0.068  sec/batch)
step 90910, loss=0.59  (1841.3  examples/sec;  0.070  sec/batch)
step 90920, loss=0.65  (1830.6  examples/sec;  0.070  sec/batch)
step 90930, loss=0.51  (1872.5  examples/sec;  0.068  sec/batch)
step 90940, loss=0.71  (1893.4  examples/sec;  0.068  sec/batch)
step 90950, loss=0.67  (1914.0  examples/sec;  0.067  sec/batch)
step 90960, loss=0.66  (1839.1  examples/sec;  0.070  sec/batch)
step 90970, loss=0.55  (1925.7  examples/sec;  0.066  sec/batch)
step 90980, loss=0.53  (1968.5  examples/sec;  0.065  sec/batch)
step 90990, loss=0.52  (1836.9  examples/sec;  0.070  sec/batch)
step 91000, loss=0.63  (1916.5  examples/sec;  0.067  sec/batch)
step 91010, loss=0.59  (1808.6  examples/sec;  0.071  sec/batch)
step 91020, loss=0.57  (1956.4  examples/sec;  0.065  sec/batch)
step 91030, loss=0.74  (1970.6  examples/sec;  0.065  sec/batch)
step 91040, loss=0.54  (1882.8  examples/sec;  0.068  sec/batch)
step 91050, loss=0.64  (1902.9  examples/sec;  0.067  sec/batch)
step 91060, loss=0.68  (1932.2  examples/sec;  0.066  sec/batch)
step 91070, loss=0.58  (1935.9  examples/sec;  0.066  sec/batch)
step 91080, loss=0.63  (1857.7  examples/sec;  0.069  sec/batch)
step 91090, loss=0.60  (1990.4  examples/sec;  0.064  sec/batch)
step 91100, loss=0.66  (1935.6  examples/sec;  0.066  sec/batch)
step 91110, loss=0.43  (1913.9  examples/sec;  0.067  sec/batch)
step 91120, loss=0.79  (1875.0  examples/sec;  0.068  sec/batch)
step 91130, loss=0.61  (1913.8  examples/sec;  0.067  sec/batch)
step 91140, loss=0.66  (1877.7  examples/sec;  0.068  sec/batch)
step 91150, loss=0.60  (1914.8  examples/sec;  0.067  sec/batch)
step 91160, loss=0.50  (1983.7  examples/sec;  0.065  sec/batch)
step 91170, loss=0.57  (1951.0  examples/sec;  0.066  sec/batch)
step 91180, loss=0.71  (1911.3  examples/sec;  0.067  sec/batch)
step 91190, loss=0.61  (1938.4  examples/sec;  0.066  sec/batch)
step 91200, loss=0.61  (1870.8  examples/sec;  0.068  sec/batch)
step 91210, loss=0.62  (1951.3  examples/sec;  0.066  sec/batch)
step 91220, loss=0.79  (1859.7  examples/sec;  0.069  sec/batch)
step 91230, loss=0.57  (1657.9  examples/sec;  0.077  sec/batch)
step 91240, loss=0.52  (1925.4  examples/sec;  0.066  sec/batch)
step 91250, loss=0.59  (1900.9  examples/sec;  0.067  sec/batch)
step 91260, loss=0.66  (1865.2  examples/sec;  0.069  sec/batch)
step 91270, loss=0.48  (1891.4  examples/sec;  0.068  sec/batch)
step 91280, loss=0.51  (1939.2  examples/sec;  0.066  sec/batch)
step 91290, loss=0.55  (1890.2  examples/sec;  0.068  sec/batch)
step 91300, loss=0.59  (1948.1  examples/sec;  0.066  sec/batch)
step 91310, loss=0.57  (1648.9  examples/sec;  0.078  sec/batch)
step 91320, loss=0.58  (1789.9  examples/sec;  0.072  sec/batch)
step 91330, loss=0.64  (1851.7  examples/sec;  0.069  sec/batch)
step 91340, loss=0.63  (1977.4  examples/sec;  0.065  sec/batch)
step 91350, loss=0.52  (1964.6  examples/sec;  0.065  sec/batch)
step 91360, loss=0.57  (1936.6  examples/sec;  0.066  sec/batch)
step 91370, loss=0.80  (1837.1  examples/sec;  0.070  sec/batch)
step 91380, loss=0.53  (1879.2  examples/sec;  0.068  sec/batch)
step 91390, loss=0.81  (1852.9  examples/sec;  0.069  sec/batch)
step 91400, loss=0.70  (1810.1  examples/sec;  0.071  sec/batch)
step 91410, loss=0.55  (1945.1  examples/sec;  0.066  sec/batch)
step 91420, loss=0.55  (1900.7  examples/sec;  0.067  sec/batch)
step 91430, loss=0.68  (1861.4  examples/sec;  0.069  sec/batch)
step 91440, loss=0.39  (1856.3  examples/sec;  0.069  sec/batch)
step 91450, loss=0.52  (1892.2  examples/sec;  0.068  sec/batch)
step 91460, loss=0.67  (1814.4  examples/sec;  0.071  sec/batch)
step 91470, loss=0.78  (1904.8  examples/sec;  0.067  sec/batch)
step 91480, loss=0.61  (1923.0  examples/sec;  0.067  sec/batch)
step 91490, loss=0.68  (1911.3  examples/sec;  0.067  sec/batch)
step 91500, loss=0.58  (1841.4  examples/sec;  0.070  sec/batch)
step 91510, loss=0.50  (1952.4  examples/sec;  0.066  sec/batch)
step 91520, loss=0.52  (1894.2  examples/sec;  0.068  sec/batch)
step 91530, loss=0.42  (1878.0  examples/sec;  0.068  sec/batch)
step 91540, loss=0.54  (1952.9  examples/sec;  0.066  sec/batch)
step 91550, loss=0.57  (1925.1  examples/sec;  0.066  sec/batch)
step 91560, loss=0.53  (1949.8  examples/sec;  0.066  sec/batch)
step 91570, loss=0.58  (1854.2  examples/sec;  0.069  sec/batch)
step 91580, loss=0.53  (1937.0  examples/sec;  0.066  sec/batch)
step 91590, loss=0.45  (1914.1  examples/sec;  0.067  sec/batch)
step 91600, loss=0.67  (1943.9  examples/sec;  0.066  sec/batch)
step 91610, loss=0.67  (1891.6  examples/sec;  0.068  sec/batch)
step 91620, loss=0.56  (1982.6  examples/sec;  0.065  sec/batch)
step 91630, loss=0.58  (1959.6  examples/sec;  0.065  sec/batch)
step 91640, loss=0.53  (1912.3  examples/sec;  0.067  sec/batch)
step 91650, loss=0.61  (1973.1  examples/sec;  0.065  sec/batch)
step 91660, loss=0.48  (1914.6  examples/sec;  0.067  sec/batch)
step 91670, loss=0.68  (1918.1  examples/sec;  0.067  sec/batch)
step 91680, loss=0.54  (1910.7  examples/sec;  0.067  sec/batch)
step 91690, loss=0.55  (1993.3  examples/sec;  0.064  sec/batch)
step 91700, loss=0.83  (1957.6  examples/sec;  0.065  sec/batch)
step 91710, loss=0.52  (1608.5  examples/sec;  0.080  sec/batch)
step 91720, loss=0.47  (1920.0  examples/sec;  0.067  sec/batch)
step 91730, loss=0.62  (1827.7  examples/sec;  0.070  sec/batch)
step 91740, loss=0.58  (1909.0  examples/sec;  0.067  sec/batch)
step 91750, loss=0.60  (1857.4  examples/sec;  0.069  sec/batch)
step 91760, loss=0.66  (1769.2  examples/sec;  0.072  sec/batch)
step 91770, loss=0.53  (1810.3  examples/sec;  0.071  sec/batch)
step 91780, loss=0.51  (1907.5  examples/sec;  0.067  sec/batch)
step 91790, loss=0.55  (1853.5  examples/sec;  0.069  sec/batch)
step 91800, loss=0.69  (1925.6  examples/sec;  0.066  sec/batch)
step 91810, loss=0.60  (1932.0  examples/sec;  0.066  sec/batch)
step 91820, loss=0.59  (1812.0  examples/sec;  0.071  sec/batch)
step 91830, loss=0.68  (1907.1  examples/sec;  0.067  sec/batch)
step 91840, loss=0.60  (1985.0  examples/sec;  0.064  sec/batch)
step 91850, loss=0.61  (1836.5  examples/sec;  0.070  sec/batch)
step 91860, loss=0.51  (1897.2  examples/sec;  0.067  sec/batch)
step 91870, loss=0.53  (1909.5  examples/sec;  0.067  sec/batch)
step 91880, loss=0.58  (1984.7  examples/sec;  0.064  sec/batch)
step 91890, loss=0.61  (1919.2  examples/sec;  0.067  sec/batch)
step 91900, loss=0.77  (2011.1  examples/sec;  0.064  sec/batch)
step 91910, loss=0.73  (1862.3  examples/sec;  0.069  sec/batch)
step 91920, loss=0.55  (1818.3  examples/sec;  0.070  sec/batch)
step 91930, loss=0.61  (1917.1  examples/sec;  0.067  sec/batch)
step 91940, loss=0.67  (1882.4  examples/sec;  0.068  sec/batch)
step 91950, loss=0.72  (1866.7  examples/sec;  0.069  sec/batch)
step 91960, loss=0.74  (1879.2  examples/sec;  0.068  sec/batch)
step 91970, loss=0.59  (1927.2  examples/sec;  0.066  sec/batch)
step 91980, loss=0.63  (1868.5  examples/sec;  0.069  sec/batch)
step 91990, loss=0.61  (1882.0  examples/sec;  0.068  sec/batch)
step 92000, loss=0.69  (1844.9  examples/sec;  0.069  sec/batch)
step 92010, loss=0.59  (1921.3  examples/sec;  0.067  sec/batch)
step 92020, loss=0.48  (1886.5  examples/sec;  0.068  sec/batch)
step 92030, loss=0.52  (1915.2  examples/sec;  0.067  sec/batch)
step 92040, loss=0.77  (1895.3  examples/sec;  0.068  sec/batch)
step 92050, loss=0.54  (1819.3  examples/sec;  0.070  sec/batch)
step 92060, loss=0.54  (1874.7  examples/sec;  0.068  sec/batch)
step 92070, loss=0.54  (1916.1  examples/sec;  0.067  sec/batch)
step 92080, loss=0.59  (2029.3  examples/sec;  0.063  sec/batch)
step 92090, loss=0.72  (1896.0  examples/sec;  0.068  sec/batch)
step 92100, loss=0.55  (1870.2  examples/sec;  0.068  sec/batch)
step 92110, loss=0.76  (1918.7  examples/sec;  0.067  sec/batch)
step 92120, loss=0.53  (1854.8  examples/sec;  0.069  sec/batch)
step 92130, loss=0.69  (1866.8  examples/sec;  0.069  sec/batch)
step 92140, loss=0.60  (1877.1  examples/sec;  0.068  sec/batch)
step 92150, loss=0.81  (1860.1  examples/sec;  0.069  sec/batch)
step 92160, loss=0.57  (1902.0  examples/sec;  0.067  sec/batch)
step 92170, loss=0.68  (1849.0  examples/sec;  0.069  sec/batch)
step 92180, loss=0.66  (1835.2  examples/sec;  0.070  sec/batch)
step 92190, loss=0.47  (1920.2  examples/sec;  0.067  sec/batch)
step 92200, loss=0.52  (1835.2  examples/sec;  0.070  sec/batch)
step 92210, loss=0.59  (1952.3  examples/sec;  0.066  sec/batch)
step 92220, loss=0.61  (1829.7  examples/sec;  0.070  sec/batch)
step 92230, loss=0.60  (1910.1  examples/sec;  0.067  sec/batch)
step 92240, loss=0.78  (1780.2  examples/sec;  0.072  sec/batch)
step 92250, loss=0.47  (1835.9  examples/sec;  0.070  sec/batch)
step 92260, loss=0.76  (1876.8  examples/sec;  0.068  sec/batch)
step 92270, loss=0.53  (1895.0  examples/sec;  0.068  sec/batch)
step 92280, loss=0.75  (1832.9  examples/sec;  0.070  sec/batch)
step 92290, loss=0.60  (1912.8  examples/sec;  0.067  sec/batch)
step 92300, loss=0.64  (1910.2  examples/sec;  0.067  sec/batch)
step 92310, loss=0.64  (1925.0  examples/sec;  0.066  sec/batch)
step 92320, loss=0.63  (1877.4  examples/sec;  0.068  sec/batch)
step 92330, loss=0.60  (1864.0  examples/sec;  0.069  sec/batch)
step 92340, loss=0.65  (1955.8  examples/sec;  0.065  sec/batch)
step 92350, loss=0.58  (1941.4  examples/sec;  0.066  sec/batch)
step 92360, loss=0.57  (1801.0  examples/sec;  0.071  sec/batch)
step 92370, loss=0.64  (2014.0  examples/sec;  0.064  sec/batch)
step 92380, loss=0.77  (1961.3  examples/sec;  0.065  sec/batch)
step 92390, loss=0.51  (1897.9  examples/sec;  0.067  sec/batch)
step 92400, loss=0.50  (1834.7  examples/sec;  0.070  sec/batch)
step 92410, loss=0.49  (1885.8  examples/sec;  0.068  sec/batch)
step 92420, loss=0.60  (1849.9  examples/sec;  0.069  sec/batch)
step 92430, loss=0.55  (1850.0  examples/sec;  0.069  sec/batch)
step 92440, loss=0.68  (1951.6  examples/sec;  0.066  sec/batch)
step 92450, loss=0.52  (1887.5  examples/sec;  0.068  sec/batch)
step 92460, loss=0.52  (1960.5  examples/sec;  0.065  sec/batch)
step 92470, loss=0.52  (1887.8  examples/sec;  0.068  sec/batch)
step 92480, loss=0.78  (2014.8  examples/sec;  0.064  sec/batch)
step 92490, loss=0.82  (1866.8  examples/sec;  0.069  sec/batch)
step 92500, loss=0.72  (1919.9  examples/sec;  0.067  sec/batch)
step 92510, loss=0.50  (1862.4  examples/sec;  0.069  sec/batch)
step 92520, loss=0.50  (1967.7  examples/sec;  0.065  sec/batch)
step 92530, loss=0.61  (1798.4  examples/sec;  0.071  sec/batch)
step 92540, loss=0.55  (1888.4  examples/sec;  0.068  sec/batch)
step 92550, loss=0.52  (1913.9  examples/sec;  0.067  sec/batch)
step 92560, loss=0.48  (1909.2  examples/sec;  0.067  sec/batch)
step 92570, loss=0.54  (1876.1  examples/sec;  0.068  sec/batch)
step 92580, loss=0.55  (1978.7  examples/sec;  0.065  sec/batch)
step 92590, loss=0.53  (1941.4  examples/sec;  0.066  sec/batch)
step 92600, loss=0.54  (1892.4  examples/sec;  0.068  sec/batch)
step 92610, loss=0.59  (1958.1  examples/sec;  0.065  sec/batch)
step 92620, loss=0.52  (1997.9  examples/sec;  0.064  sec/batch)
step 92630, loss=0.58  (1974.4  examples/sec;  0.065  sec/batch)
step 92640, loss=0.52  (1887.5  examples/sec;  0.068  sec/batch)
step 92650, loss=0.49  (1900.3  examples/sec;  0.067  sec/batch)
step 92660, loss=0.62  (1855.0  examples/sec;  0.069  sec/batch)
step 92670, loss=0.50  (1999.2  examples/sec;  0.064  sec/batch)
step 92680, loss=0.53  (1870.8  examples/sec;  0.068  sec/batch)
step 92690, loss=0.66  (1893.1  examples/sec;  0.068  sec/batch)
step 92700, loss=0.55  (1910.1  examples/sec;  0.067  sec/batch)
step 92710, loss=0.58  (1915.5  examples/sec;  0.067  sec/batch)
step 92720, loss=0.62  (2001.2  examples/sec;  0.064  sec/batch)
step 92730, loss=0.53  (1876.9  examples/sec;  0.068  sec/batch)
step 92740, loss=0.57  (1868.5  examples/sec;  0.069  sec/batch)
step 92750, loss=0.55  (1975.8  examples/sec;  0.065  sec/batch)
step 92760, loss=0.55  (1979.9  examples/sec;  0.065  sec/batch)
step 92770, loss=0.77  (1968.1  examples/sec;  0.065  sec/batch)
step 92780, loss=0.63  (1953.3  examples/sec;  0.066  sec/batch)
step 92790, loss=0.56  (1895.0  examples/sec;  0.068  sec/batch)
step 92800, loss=0.55  (1837.0  examples/sec;  0.070  sec/batch)
step 92810, loss=0.75  (1823.9  examples/sec;  0.070  sec/batch)
step 92820, loss=0.61  (1903.1  examples/sec;  0.067  sec/batch)
step 92830, loss=0.60  (1852.4  examples/sec;  0.069  sec/batch)
step 92840, loss=0.59  (1951.7  examples/sec;  0.066  sec/batch)
step 92850, loss=0.54  (1814.8  examples/sec;  0.071  sec/batch)
step 92860, loss=0.55  (1871.3  examples/sec;  0.068  sec/batch)
step 92870, loss=0.59  (1895.9  examples/sec;  0.068  sec/batch)
step 92880, loss=0.52  (1880.4  examples/sec;  0.068  sec/batch)
step 92890, loss=0.57  (1925.0  examples/sec;  0.066  sec/batch)
step 92900, loss=0.57  (1577.8  examples/sec;  0.081  sec/batch)
step 92910, loss=0.49  (1909.9  examples/sec;  0.067  sec/batch)
step 92920, loss=0.64  (1851.2  examples/sec;  0.069  sec/batch)
step 92930, loss=0.71  (2030.5  examples/sec;  0.063  sec/batch)
step 92940, loss=0.51  (1880.8  examples/sec;  0.068  sec/batch)
step 92950, loss=0.64  (1944.1  examples/sec;  0.066  sec/batch)
step 92960, loss=0.72  (1912.3  examples/sec;  0.067  sec/batch)
step 92970, loss=0.60  (1933.8  examples/sec;  0.066  sec/batch)
step 92980, loss=0.60  (1920.6  examples/sec;  0.067  sec/batch)
step 92990, loss=0.62  (1868.3  examples/sec;  0.069  sec/batch)
step 93000, loss=0.62  (1858.4  examples/sec;  0.069  sec/batch)
step 93010, loss=0.59  (1893.0  examples/sec;  0.068  sec/batch)
step 93020, loss=0.54  (1991.6  examples/sec;  0.064  sec/batch)
step 93030, loss=0.62  (1868.1  examples/sec;  0.069  sec/batch)
step 93040, loss=0.62  (1887.3  examples/sec;  0.068  sec/batch)
step 93050, loss=0.57  (1930.2  examples/sec;  0.066  sec/batch)
step 93060, loss=0.58  (1905.8  examples/sec;  0.067  sec/batch)
step 93070, loss=0.62  (1889.0  examples/sec;  0.068  sec/batch)
step 93080, loss=0.41  (1816.6  examples/sec;  0.070  sec/batch)
step 93090, loss=0.52  (1939.5  examples/sec;  0.066  sec/batch)
step 93100, loss=0.58  (1844.0  examples/sec;  0.069  sec/batch)
step 93110, loss=0.58  (1854.6  examples/sec;  0.069  sec/batch)
step 93120, loss=0.49  (1895.4  examples/sec;  0.068  sec/batch)
step 93130, loss=0.72  (1898.9  examples/sec;  0.067  sec/batch)
step 93140, loss=0.50  (1883.4  examples/sec;  0.068  sec/batch)
step 93150, loss=0.61  (1922.4  examples/sec;  0.067  sec/batch)
step 93160, loss=0.63  (1890.5  examples/sec;  0.068  sec/batch)
step 93170, loss=0.49  (1952.4  examples/sec;  0.066  sec/batch)
step 93180, loss=0.53  (1865.0  examples/sec;  0.069  sec/batch)
step 93190, loss=0.70  (1892.7  examples/sec;  0.068  sec/batch)
step 93200, loss=0.53  (1817.3  examples/sec;  0.070  sec/batch)
step 93210, loss=0.58  (2011.8  examples/sec;  0.064  sec/batch)
step 93220, loss=0.50  (1928.6  examples/sec;  0.066  sec/batch)
step 93230, loss=0.53  (1893.5  examples/sec;  0.068  sec/batch)
step 93240, loss=0.48  (1912.4  examples/sec;  0.067  sec/batch)
step 93250, loss=0.54  (1860.3  examples/sec;  0.069  sec/batch)
step 93260, loss=0.72  (1903.7  examples/sec;  0.067  sec/batch)
step 93270, loss=0.74  (1869.6  examples/sec;  0.068  sec/batch)
step 93280, loss=0.52  (1929.2  examples/sec;  0.066  sec/batch)
step 93290, loss=0.52  (1890.8  examples/sec;  0.068  sec/batch)
step 93300, loss=0.59  (1945.4  examples/sec;  0.066  sec/batch)
step 93310, loss=0.56  (1937.1  examples/sec;  0.066  sec/batch)
step 93320, loss=0.46  (1882.7  examples/sec;  0.068  sec/batch)
step 93330, loss=0.61  (1944.8  examples/sec;  0.066  sec/batch)
step 93340, loss=0.62  (1954.2  examples/sec;  0.066  sec/batch)
step 93350, loss=0.58  (1893.1  examples/sec;  0.068  sec/batch)
step 93360, loss=0.50  (1896.5  examples/sec;  0.067  sec/batch)
step 93370, loss=0.50  (1933.1  examples/sec;  0.066  sec/batch)
step 93380, loss=0.66  (1852.6  examples/sec;  0.069  sec/batch)
step 93390, loss=0.66  (1768.4  examples/sec;  0.072  sec/batch)
step 93400, loss=0.46  (1969.4  examples/sec;  0.065  sec/batch)
step 93410, loss=0.89  (1933.6  examples/sec;  0.066  sec/batch)
step 93420, loss=0.58  (1924.0  examples/sec;  0.067  sec/batch)
step 93430, loss=0.47  (1813.4  examples/sec;  0.071  sec/batch)
step 93440, loss=0.58  (1918.4  examples/sec;  0.067  sec/batch)
step 93450, loss=0.61  (1826.2  examples/sec;  0.070  sec/batch)
step 93460, loss=0.71  (1839.0  examples/sec;  0.070  sec/batch)
step 93470, loss=0.60  (1951.9  examples/sec;  0.066  sec/batch)
step 93480, loss=0.44  (1949.3  examples/sec;  0.066  sec/batch)
step 93490, loss=0.72  (1975.4  examples/sec;  0.065  sec/batch)
step 93500, loss=0.67  (1934.0  examples/sec;  0.066  sec/batch)
step 93510, loss=0.54  (1868.6  examples/sec;  0.069  sec/batch)
step 93520, loss=0.57  (1923.5  examples/sec;  0.067  sec/batch)
step 93530, loss=0.57  (1831.6  examples/sec;  0.070  sec/batch)
step 93540, loss=0.55  (1931.7  examples/sec;  0.066  sec/batch)
step 93550, loss=0.65  (1922.8  examples/sec;  0.067  sec/batch)
step 93560, loss=0.66  (1979.4  examples/sec;  0.065  sec/batch)
step 93570, loss=0.63  (1948.3  examples/sec;  0.066  sec/batch)
step 93580, loss=0.55  (1907.9  examples/sec;  0.067  sec/batch)
step 93590, loss=0.61  (1904.9  examples/sec;  0.067  sec/batch)
step 93600, loss=0.60  (1880.3  examples/sec;  0.068  sec/batch)
step 93610, loss=0.54  (1933.5  examples/sec;  0.066  sec/batch)
step 93620, loss=0.59  (1875.6  examples/sec;  0.068  sec/batch)
step 93630, loss=0.61  (1877.9  examples/sec;  0.068  sec/batch)
step 93640, loss=0.54  (1869.0  examples/sec;  0.068  sec/batch)
step 93650, loss=0.62  (1793.0  examples/sec;  0.071  sec/batch)
step 93660, loss=0.53  (1947.2  examples/sec;  0.066  sec/batch)
step 93670, loss=0.55  (1894.3  examples/sec;  0.068  sec/batch)
step 93680, loss=0.60  (1853.8  examples/sec;  0.069  sec/batch)
step 93690, loss=0.59  (1896.8  examples/sec;  0.067  sec/batch)
step 93700, loss=0.43  (1940.5  examples/sec;  0.066  sec/batch)
step 93710, loss=0.52  (1941.1  examples/sec;  0.066  sec/batch)
step 93720, loss=0.70  (1901.9  examples/sec;  0.067  sec/batch)
step 93730, loss=0.55  (1910.1  examples/sec;  0.067  sec/batch)
step 93740, loss=0.65  (1910.7  examples/sec;  0.067  sec/batch)
step 93750, loss=0.72  (1895.0  examples/sec;  0.068  sec/batch)
step 93760, loss=0.57  (1821.9  examples/sec;  0.070  sec/batch)
step 93770, loss=0.56  (1837.0  examples/sec;  0.070  sec/batch)
step 93780, loss=0.51  (1946.6  examples/sec;  0.066  sec/batch)
step 93790, loss=0.56  (1965.1  examples/sec;  0.065  sec/batch)
step 93800, loss=0.64  (1986.8  examples/sec;  0.064  sec/batch)
step 93810, loss=0.63  (1812.5  examples/sec;  0.071  sec/batch)
step 93820, loss=0.49  (1825.6  examples/sec;  0.070  sec/batch)
step 93830, loss=0.53  (1835.4  examples/sec;  0.070  sec/batch)
step 93840, loss=0.55  (1924.1  examples/sec;  0.067  sec/batch)
step 93850, loss=0.72  (1952.6  examples/sec;  0.066  sec/batch)
step 93860, loss=0.61  (1827.2  examples/sec;  0.070  sec/batch)
step 93870, loss=0.56  (1831.0  examples/sec;  0.070  sec/batch)
step 93880, loss=0.63  (1932.2  examples/sec;  0.066  sec/batch)
step 93890, loss=0.66  (1906.6  examples/sec;  0.067  sec/batch)
step 93900, loss=0.50  (1950.7  examples/sec;  0.066  sec/batch)
step 93910, loss=0.52  (1911.1  examples/sec;  0.067  sec/batch)
step 93920, loss=0.47  (1970.5  examples/sec;  0.065  sec/batch)
step 93930, loss=0.62  (1874.4  examples/sec;  0.068  sec/batch)
step 93940, loss=0.55  (1911.7  examples/sec;  0.067  sec/batch)
step 93950, loss=0.51  (1852.2  examples/sec;  0.069  sec/batch)
step 93960, loss=0.44  (1891.6  examples/sec;  0.068  sec/batch)
step 93970, loss=0.54  (1828.6  examples/sec;  0.070  sec/batch)
step 93980, loss=0.49  (1843.1  examples/sec;  0.069  sec/batch)
step 93990, loss=0.72  (1975.6  examples/sec;  0.065  sec/batch)
step 94000, loss=0.50  (1864.5  examples/sec;  0.069  sec/batch)
step 94010, loss=0.66  (1888.4  examples/sec;  0.068  sec/batch)
step 94020, loss=0.49  (1876.6  examples/sec;  0.068  sec/batch)
step 94030, loss=0.64  (1919.5  examples/sec;  0.067  sec/batch)
step 94040, loss=0.69  (1891.3  examples/sec;  0.068  sec/batch)
step 94050, loss=0.57  (1933.7  examples/sec;  0.066  sec/batch)
step 94060, loss=0.61  (1912.8  examples/sec;  0.067  sec/batch)
step 94070, loss=0.68  (1830.0  examples/sec;  0.070  sec/batch)
step 94080, loss=0.56  (1888.3  examples/sec;  0.068  sec/batch)
step 94090, loss=0.70  (1961.1  examples/sec;  0.065  sec/batch)
step 94100, loss=0.49  (1878.7  examples/sec;  0.068  sec/batch)
step 94110, loss=0.54  (1888.5  examples/sec;  0.068  sec/batch)
step 94120, loss=0.71  (1847.2  examples/sec;  0.069  sec/batch)
step 94130, loss=0.73  (1973.8  examples/sec;  0.065  sec/batch)
step 94140, loss=0.53  (1772.6  examples/sec;  0.072  sec/batch)
step 94150, loss=0.45  (1819.5  examples/sec;  0.070  sec/batch)
step 94160, loss=0.56  (1921.5  examples/sec;  0.067  sec/batch)
step 94170, loss=0.56  (1932.5  examples/sec;  0.066  sec/batch)
step 94180, loss=0.68  (1882.4  examples/sec;  0.068  sec/batch)
step 94190, loss=0.70  (1800.8  examples/sec;  0.071  sec/batch)
step 94200, loss=0.74  (1879.0  examples/sec;  0.068  sec/batch)
step 94210, loss=0.63  (1878.8  examples/sec;  0.068  sec/batch)
step 94220, loss=0.58  (1943.0  examples/sec;  0.066  sec/batch)
step 94230, loss=0.64  (1883.8  examples/sec;  0.068  sec/batch)
step 94240, loss=0.61  (1923.5  examples/sec;  0.067  sec/batch)
step 94250, loss=0.64  (1951.7  examples/sec;  0.066  sec/batch)
step 94260, loss=0.59  (1933.2  examples/sec;  0.066  sec/batch)
step 94270, loss=0.54  (1896.4  examples/sec;  0.067  sec/batch)
step 94280, loss=0.54  (1829.8  examples/sec;  0.070  sec/batch)
step 94290, loss=0.51  (1942.5  examples/sec;  0.066  sec/batch)
step 94300, loss=0.61  (1942.4  examples/sec;  0.066  sec/batch)
step 94310, loss=0.72  (1853.9  examples/sec;  0.069  sec/batch)
step 94320, loss=0.68  (1922.5  examples/sec;  0.067  sec/batch)
step 94330, loss=0.57  (1896.2  examples/sec;  0.068  sec/batch)
step 94340, loss=0.54  (1957.6  examples/sec;  0.065  sec/batch)
step 94350, loss=0.61  (1902.6  examples/sec;  0.067  sec/batch)
step 94360, loss=0.56  (1925.2  examples/sec;  0.066  sec/batch)
step 94370, loss=0.49  (1896.4  examples/sec;  0.067  sec/batch)
step 94380, loss=0.54  (1951.2  examples/sec;  0.066  sec/batch)
step 94390, loss=0.69  (1814.3  examples/sec;  0.071  sec/batch)
step 94400, loss=0.58  (1884.6  examples/sec;  0.068  sec/batch)
step 94410, loss=0.54  (1878.9  examples/sec;  0.068  sec/batch)
step 94420, loss=0.58  (1885.6  examples/sec;  0.068  sec/batch)
step 94430, loss=0.53  (1953.2  examples/sec;  0.066  sec/batch)
step 94440, loss=0.60  (1873.0  examples/sec;  0.068  sec/batch)
step 94450, loss=0.54  (1845.2  examples/sec;  0.069  sec/batch)
step 94460, loss=0.60  (1887.2  examples/sec;  0.068  sec/batch)
step 94470, loss=0.52  (1875.3  examples/sec;  0.068  sec/batch)
step 94480, loss=0.58  (1876.1  examples/sec;  0.068  sec/batch)
step 94490, loss=0.63  (1837.0  examples/sec;  0.070  sec/batch)
step 94500, loss=0.67  (1871.8  examples/sec;  0.068  sec/batch)
step 94510, loss=0.57  (1881.6  examples/sec;  0.068  sec/batch)
step 94520, loss=0.56  (1830.6  examples/sec;  0.070  sec/batch)
step 94530, loss=0.61  (1715.5  examples/sec;  0.075  sec/batch)
step 94540, loss=0.66  (1881.7  examples/sec;  0.068  sec/batch)
step 94550, loss=0.53  (1926.3  examples/sec;  0.066  sec/batch)
step 94560, loss=0.69  (1874.5  examples/sec;  0.068  sec/batch)
step 94570, loss=0.67  (1807.1  examples/sec;  0.071  sec/batch)
step 94580, loss=0.58  (1880.3  examples/sec;  0.068  sec/batch)
step 94590, loss=0.59  (1883.1  examples/sec;  0.068  sec/batch)
step 94600, loss=0.64  (1880.6  examples/sec;  0.068  sec/batch)
step 94610, loss=0.65  (1827.1  examples/sec;  0.070  sec/batch)
step 94620, loss=0.55  (1941.7  examples/sec;  0.066  sec/batch)
step 94630, loss=0.65  (1878.0  examples/sec;  0.068  sec/batch)
step 94640, loss=0.57  (1868.2  examples/sec;  0.069  sec/batch)
step 94650, loss=0.60  (1838.6  examples/sec;  0.070  sec/batch)
step 94660, loss=0.49  (1957.8  examples/sec;  0.065  sec/batch)
step 94670, loss=0.65  (1921.8  examples/sec;  0.067  sec/batch)
step 94680, loss=0.57  (1873.2  examples/sec;  0.068  sec/batch)
step 94690, loss=0.60  (1917.0  examples/sec;  0.067  sec/batch)
step 94700, loss=0.66  (1912.2  examples/sec;  0.067  sec/batch)
step 94710, loss=0.84  (1862.5  examples/sec;  0.069  sec/batch)
step 94720, loss=0.61  (1876.6  examples/sec;  0.068  sec/batch)
step 94730, loss=0.64  (1929.4  examples/sec;  0.066  sec/batch)
step 94740, loss=0.63  (1881.7  examples/sec;  0.068  sec/batch)
step 94750, loss=0.55  (1883.5  examples/sec;  0.068  sec/batch)
step 94760, loss=0.51  (1976.0  examples/sec;  0.065  sec/batch)
step 94770, loss=0.47  (1986.7  examples/sec;  0.064  sec/batch)
step 94780, loss=0.68  (1892.0  examples/sec;  0.068  sec/batch)
step 94790, loss=0.49  (1956.5  examples/sec;  0.065  sec/batch)
step 94800, loss=0.63  (1608.5  examples/sec;  0.080  sec/batch)
step 94810, loss=0.54  (1899.1  examples/sec;  0.067  sec/batch)
step 94820, loss=0.57  (1614.7  examples/sec;  0.079  sec/batch)
step 94830, loss=0.62  (1799.5  examples/sec;  0.071  sec/batch)
step 94840, loss=0.56  (1966.0  examples/sec;  0.065  sec/batch)
step 94850, loss=0.58  (1872.6  examples/sec;  0.068  sec/batch)
step 94860, loss=0.62  (1650.9  examples/sec;  0.078  sec/batch)
step 94870, loss=0.70  (1798.6  examples/sec;  0.071  sec/batch)
step 94880, loss=0.61  (1875.2  examples/sec;  0.068  sec/batch)
step 94890, loss=0.61  (1719.0  examples/sec;  0.074  sec/batch)
step 94900, loss=0.58  (1911.0  examples/sec;  0.067  sec/batch)
step 94910, loss=0.43  (1874.1  examples/sec;  0.068  sec/batch)
step 94920, loss=0.41  (1896.6  examples/sec;  0.067  sec/batch)
step 94930, loss=0.75  (1890.5  examples/sec;  0.068  sec/batch)
step 94940, loss=0.63  (1863.7  examples/sec;  0.069  sec/batch)
step 94950, loss=0.55  (1746.2  examples/sec;  0.073  sec/batch)
step 94960, loss=0.53  (1835.9  examples/sec;  0.070  sec/batch)
step 94970, loss=0.48  (1912.3  examples/sec;  0.067  sec/batch)
step 94980, loss=0.82  (1912.0  examples/sec;  0.067  sec/batch)
step 94990, loss=0.62  (1879.3  examples/sec;  0.068  sec/batch)
step 95000, loss=0.51  (1852.1  examples/sec;  0.069  sec/batch)
step 95010, loss=0.47  (1912.8  examples/sec;  0.067  sec/batch)
step 95020, loss=0.58  (1944.3  examples/sec;  0.066  sec/batch)
step 95030, loss=0.62  (1841.4  examples/sec;  0.070  sec/batch)
step 95040, loss=0.66  (1847.7  examples/sec;  0.069  sec/batch)
step 95050, loss=0.65  (1915.2  examples/sec;  0.067  sec/batch)
step 95060, loss=0.69  (2010.8  examples/sec;  0.064  sec/batch)
step 95070, loss=0.65  (1927.0  examples/sec;  0.066  sec/batch)
step 95080, loss=0.54  (1882.1  examples/sec;  0.068  sec/batch)
step 95090, loss=0.59  (1765.1  examples/sec;  0.073  sec/batch)
step 95100, loss=0.65  (1948.5  examples/sec;  0.066  sec/batch)
step 95110, loss=0.60  (1812.5  examples/sec;  0.071  sec/batch)
step 95120, loss=0.60  (1989.4  examples/sec;  0.064  sec/batch)
step 95130, loss=0.47  (1987.4  examples/sec;  0.064  sec/batch)
step 95140, loss=0.46  (1884.5  examples/sec;  0.068  sec/batch)
step 95150, loss=0.52  (1830.7  examples/sec;  0.070  sec/batch)
step 95160, loss=0.62  (1932.1  examples/sec;  0.066  sec/batch)
step 95170, loss=0.63  (1972.2  examples/sec;  0.065  sec/batch)
step 95180, loss=0.55  (1865.6  examples/sec;  0.069  sec/batch)
step 95190, loss=0.54  (1912.9  examples/sec;  0.067  sec/batch)
step 95200, loss=0.59  (1787.1  examples/sec;  0.072  sec/batch)
step 95210, loss=0.59  (1813.5  examples/sec;  0.071  sec/batch)
step 95220, loss=0.53  (1884.2  examples/sec;  0.068  sec/batch)
step 95230, loss=0.48  (1854.2  examples/sec;  0.069  sec/batch)
step 95240, loss=0.72  (1794.5  examples/sec;  0.071  sec/batch)
step 95250, loss=0.72  (1877.9  examples/sec;  0.068  sec/batch)
step 95260, loss=0.51  (1912.1  examples/sec;  0.067  sec/batch)
step 95270, loss=0.64  (1876.8  examples/sec;  0.068  sec/batch)
step 95280, loss=0.61  (1929.9  examples/sec;  0.066  sec/batch)
step 95290, loss=0.64  (1735.9  examples/sec;  0.074  sec/batch)
step 95300, loss=0.63  (1963.7  examples/sec;  0.065  sec/batch)
step 95310, loss=0.53  (1859.0  examples/sec;  0.069  sec/batch)
step 95320, loss=0.52  (1978.6  examples/sec;  0.065  sec/batch)
step 95330, loss=0.56  (2045.7  examples/sec;  0.063  sec/batch)
step 95340, loss=0.42  (1884.5  examples/sec;  0.068  sec/batch)
step 95350, loss=0.56  (1938.8  examples/sec;  0.066  sec/batch)
step 95360, loss=0.62  (1949.8  examples/sec;  0.066  sec/batch)
step 95370, loss=0.57  (1727.4  examples/sec;  0.074  sec/batch)
step 95380, loss=0.55  (1854.2  examples/sec;  0.069  sec/batch)
step 95390, loss=0.54  (1580.7  examples/sec;  0.081  sec/batch)
step 95400, loss=0.49  (1999.8  examples/sec;  0.064  sec/batch)
step 95410, loss=0.54  (1951.0  examples/sec;  0.066  sec/batch)
step 95420, loss=0.51  (1863.9  examples/sec;  0.069  sec/batch)
step 95430, loss=0.49  (1916.7  examples/sec;  0.067  sec/batch)
step 95440, loss=0.58  (1913.3  examples/sec;  0.067  sec/batch)
step 95450, loss=0.61  (1927.2  examples/sec;  0.066  sec/batch)
step 95460, loss=0.60  (2007.8  examples/sec;  0.064  sec/batch)
step 95470, loss=0.50  (1775.8  examples/sec;  0.072  sec/batch)
step 95480, loss=0.57  (1892.8  examples/sec;  0.068  sec/batch)
step 95490, loss=0.55  (1801.3  examples/sec;  0.071  sec/batch)
step 95500, loss=0.65  (1945.0  examples/sec;  0.066  sec/batch)
step 95510, loss=0.61  (1867.6  examples/sec;  0.069  sec/batch)
step 95520, loss=0.59  (1887.9  examples/sec;  0.068  sec/batch)
step 95530, loss=0.57  (1881.2  examples/sec;  0.068  sec/batch)
step 95540, loss=0.52  (1954.3  examples/sec;  0.065  sec/batch)
step 95550, loss=0.47  (1881.5  examples/sec;  0.068  sec/batch)
step 95560, loss=0.65  (1886.5  examples/sec;  0.068  sec/batch)
step 95570, loss=0.56  (1847.6  examples/sec;  0.069  sec/batch)
step 95580, loss=0.55  (1839.5  examples/sec;  0.070  sec/batch)
step 95590, loss=0.56  (1921.5  examples/sec;  0.067  sec/batch)
step 95600, loss=0.52  (1872.9  examples/sec;  0.068  sec/batch)
step 95610, loss=0.66  (1922.2  examples/sec;  0.067  sec/batch)
step 95620, loss=0.60  (1803.1  examples/sec;  0.071  sec/batch)
step 95630, loss=0.57  (1927.5  examples/sec;  0.066  sec/batch)
step 95640, loss=0.49  (1861.1  examples/sec;  0.069  sec/batch)
step 95650, loss=0.55  (1899.8  examples/sec;  0.067  sec/batch)
step 95660, loss=0.50  (1868.8  examples/sec;  0.068  sec/batch)
step 95670, loss=0.69  (1854.3  examples/sec;  0.069  sec/batch)
step 95680, loss=0.55  (1939.1  examples/sec;  0.066  sec/batch)
step 95690, loss=0.54  (1861.2  examples/sec;  0.069  sec/batch)
step 95700, loss=0.56  (1932.1  examples/sec;  0.066  sec/batch)
step 95710, loss=0.63  (1896.8  examples/sec;  0.067  sec/batch)
step 95720, loss=0.50  (1973.8  examples/sec;  0.065  sec/batch)
step 95730, loss=0.68  (1663.7  examples/sec;  0.077  sec/batch)
step 95740, loss=0.46  (1939.9  examples/sec;  0.066  sec/batch)
step 95750, loss=0.47  (1885.4  examples/sec;  0.068  sec/batch)
step 95760, loss=0.67  (1879.8  examples/sec;  0.068  sec/batch)
step 95770, loss=0.55  (1870.7  examples/sec;  0.068  sec/batch)
step 95780, loss=0.71  (1861.4  examples/sec;  0.069  sec/batch)
step 95790, loss=0.47  (1776.9  examples/sec;  0.072  sec/batch)
step 95800, loss=0.41  (1850.9  examples/sec;  0.069  sec/batch)
step 95810, loss=0.59  (1816.3  examples/sec;  0.070  sec/batch)
step 95820, loss=0.56  (2015.8  examples/sec;  0.063  sec/batch)
step 95830, loss=0.55  (1856.9  examples/sec;  0.069  sec/batch)
step 95840, loss=0.72  (1893.2  examples/sec;  0.068  sec/batch)
step 95850, loss=0.59  (1902.5  examples/sec;  0.067  sec/batch)
step 95860, loss=0.57  (1878.9  examples/sec;  0.068  sec/batch)
step 95870, loss=0.58  (1896.2  examples/sec;  0.068  sec/batch)
step 95880, loss=0.64  (1924.3  examples/sec;  0.067  sec/batch)
step 95890, loss=0.64  (1842.4  examples/sec;  0.069  sec/batch)
step 95900, loss=0.63  (1851.6  examples/sec;  0.069  sec/batch)
step 95910, loss=0.77  (1938.2  examples/sec;  0.066  sec/batch)
step 95920, loss=0.70  (1550.4  examples/sec;  0.083  sec/batch)
step 95930, loss=0.62  (1958.4  examples/sec;  0.065  sec/batch)
step 95940, loss=0.57  (1957.8  examples/sec;  0.065  sec/batch)
step 95950, loss=0.63  (1838.9  examples/sec;  0.070  sec/batch)
step 95960, loss=0.55  (1807.1  examples/sec;  0.071  sec/batch)
step 95970, loss=0.72  (1910.9  examples/sec;  0.067  sec/batch)
step 95980, loss=0.56  (1829.4  examples/sec;  0.070  sec/batch)
step 95990, loss=0.75  (1859.1  examples/sec;  0.069  sec/batch)
step 96000, loss=0.54  (1864.7  examples/sec;  0.069  sec/batch)
step 96010, loss=0.53  (1922.5  examples/sec;  0.067  sec/batch)
step 96020, loss=0.56  (1972.4  examples/sec;  0.065  sec/batch)
step 96030, loss=0.59  (1951.6  examples/sec;  0.066  sec/batch)
step 96040, loss=0.55  (1841.6  examples/sec;  0.070  sec/batch)
step 96050, loss=0.69  (2037.0  examples/sec;  0.063  sec/batch)
step 96060, loss=0.52  (1911.6  examples/sec;  0.067  sec/batch)
step 96070, loss=0.57  (1855.8  examples/sec;  0.069  sec/batch)
step 96080, loss=0.64  (1915.0  examples/sec;  0.067  sec/batch)
step 96090, loss=0.57  (1881.0  examples/sec;  0.068  sec/batch)
step 96100, loss=0.53  (1965.8  examples/sec;  0.065  sec/batch)
step 96110, loss=0.37  (1958.0  examples/sec;  0.065  sec/batch)
step 96120, loss=0.52  (1894.6  examples/sec;  0.068  sec/batch)
step 96130, loss=0.52  (1892.2  examples/sec;  0.068  sec/batch)
step 96140, loss=0.65  (1891.5  examples/sec;  0.068  sec/batch)
step 96150, loss=0.61  (1976.8  examples/sec;  0.065  sec/batch)
step 96160, loss=0.65  (1905.4  examples/sec;  0.067  sec/batch)
step 96170, loss=0.49  (1925.8  examples/sec;  0.066  sec/batch)
step 96180, loss=0.48  (1942.9  examples/sec;  0.066  sec/batch)
step 96190, loss=0.63  (1886.4  examples/sec;  0.068  sec/batch)
step 96200, loss=0.43  (1831.8  examples/sec;  0.070  sec/batch)
step 96210, loss=0.56  (1921.5  examples/sec;  0.067  sec/batch)
step 96220, loss=0.57  (1854.9  examples/sec;  0.069  sec/batch)
step 96230, loss=0.37  (1873.8  examples/sec;  0.068  sec/batch)
step 96240, loss=0.41  (1973.6  examples/sec;  0.065  sec/batch)
step 96250, loss=0.49  (1852.0  examples/sec;  0.069  sec/batch)
step 96260, loss=0.62  (1968.7  examples/sec;  0.065  sec/batch)
step 96270, loss=0.46  (1922.8  examples/sec;  0.067  sec/batch)
step 96280, loss=0.67  (1965.8  examples/sec;  0.065  sec/batch)
step 96290, loss=0.55  (1859.0  examples/sec;  0.069  sec/batch)
step 96300, loss=0.62  (1850.8  examples/sec;  0.069  sec/batch)
step 96310, loss=0.64  (1865.0  examples/sec;  0.069  sec/batch)
step 96320, loss=0.59  (1567.1  examples/sec;  0.082  sec/batch)
step 96330, loss=0.71  (1952.5  examples/sec;  0.066  sec/batch)
step 96340, loss=0.52  (1897.1  examples/sec;  0.067  sec/batch)
step 96350, loss=0.48  (1814.1  examples/sec;  0.071  sec/batch)
step 96360, loss=0.56  (1894.0  examples/sec;  0.068  sec/batch)
step 96370, loss=0.60  (1945.5  examples/sec;  0.066  sec/batch)
step 96380, loss=0.59  (1782.6  examples/sec;  0.072  sec/batch)
step 96390, loss=0.76  (1982.9  examples/sec;  0.065  sec/batch)
step 96400, loss=0.67  (1995.9  examples/sec;  0.064  sec/batch)
step 96410, loss=0.69  (1874.8  examples/sec;  0.068  sec/batch)
step 96420, loss=0.59  (1943.3  examples/sec;  0.066  sec/batch)
step 96430, loss=0.49  (1817.9  examples/sec;  0.070  sec/batch)
step 96440, loss=0.67  (1822.7  examples/sec;  0.070  sec/batch)
step 96450, loss=0.64  (1830.8  examples/sec;  0.070  sec/batch)
step 96460, loss=0.55  (1869.7  examples/sec;  0.068  sec/batch)
step 96470, loss=0.67  (1842.5  examples/sec;  0.069  sec/batch)
step 96480, loss=0.66  (1993.9  examples/sec;  0.064  sec/batch)
step 96490, loss=0.58  (1866.7  examples/sec;  0.069  sec/batch)
step 96500, loss=0.47  (1935.6  examples/sec;  0.066  sec/batch)
step 96510, loss=0.66  (1877.7  examples/sec;  0.068  sec/batch)
step 96520, loss=0.66  (1887.8  examples/sec;  0.068  sec/batch)
step 96530, loss=0.54  (1949.8  examples/sec;  0.066  sec/batch)
step 96540, loss=0.63  (1973.3  examples/sec;  0.065  sec/batch)
step 96550, loss=0.63  (1897.5  examples/sec;  0.067  sec/batch)
step 96560, loss=0.45  (1935.3  examples/sec;  0.066  sec/batch)
step 96570, loss=0.67  (1932.3  examples/sec;  0.066  sec/batch)
step 96580, loss=0.54  (1893.0  examples/sec;  0.068  sec/batch)
step 96590, loss=0.55  (1934.2  examples/sec;  0.066  sec/batch)
step 96600, loss=0.65  (1823.3  examples/sec;  0.070  sec/batch)
step 96610, loss=0.65  (1823.4  examples/sec;  0.070  sec/batch)
step 96620, loss=0.57  (1879.6  examples/sec;  0.068  sec/batch)
step 96630, loss=0.62  (1908.6  examples/sec;  0.067  sec/batch)
step 96640, loss=0.53  (1871.3  examples/sec;  0.068  sec/batch)
step 96650, loss=0.67  (1930.1  examples/sec;  0.066  sec/batch)
step 96660, loss=0.41  (1706.0  examples/sec;  0.075  sec/batch)
step 96670, loss=0.71  (1918.8  examples/sec;  0.067  sec/batch)
step 96680, loss=0.44  (1953.7  examples/sec;  0.066  sec/batch)
step 96690, loss=0.57  (1860.0  examples/sec;  0.069  sec/batch)
step 96700, loss=0.55  (1879.9  examples/sec;  0.068  sec/batch)
step 96710, loss=0.50  (1882.4  examples/sec;  0.068  sec/batch)
step 96720, loss=0.49  (1932.2  examples/sec;  0.066  sec/batch)
step 96730, loss=0.82  (1883.9  examples/sec;  0.068  sec/batch)
step 96740, loss=0.63  (1979.2  examples/sec;  0.065  sec/batch)
step 96750, loss=0.51  (1958.7  examples/sec;  0.065  sec/batch)
step 96760, loss=0.70  (1887.0  examples/sec;  0.068  sec/batch)
step 96770, loss=0.57  (1917.0  examples/sec;  0.067  sec/batch)
step 96780, loss=0.64  (1862.5  examples/sec;  0.069  sec/batch)
step 96790, loss=0.62  (1935.5  examples/sec;  0.066  sec/batch)
step 96800, loss=0.45  (1908.5  examples/sec;  0.067  sec/batch)
step 96810, loss=0.49  (1940.5  examples/sec;  0.066  sec/batch)
step 96820, loss=0.64  (1945.3  examples/sec;  0.066  sec/batch)
step 96830, loss=0.63  (1912.4  examples/sec;  0.067  sec/batch)
step 96840, loss=0.49  (1929.3  examples/sec;  0.066  sec/batch)
step 96850, loss=0.53  (1933.2  examples/sec;  0.066  sec/batch)
step 96860, loss=0.68  (1887.8  examples/sec;  0.068  sec/batch)
step 96870, loss=0.51  (1887.0  examples/sec;  0.068  sec/batch)
step 96880, loss=0.62  (1865.3  examples/sec;  0.069  sec/batch)
step 96890, loss=0.72  (1924.0  examples/sec;  0.067  sec/batch)
step 96900, loss=0.69  (1911.7  examples/sec;  0.067  sec/batch)
step 96910, loss=0.46  (2030.1  examples/sec;  0.063  sec/batch)
step 96920, loss=0.43  (1850.2  examples/sec;  0.069  sec/batch)
step 96930, loss=0.66  (1948.1  examples/sec;  0.066  sec/batch)
step 96940, loss=0.68  (1875.6  examples/sec;  0.068  sec/batch)
step 96950, loss=0.56  (1919.4  examples/sec;  0.067  sec/batch)
step 96960, loss=0.61  (1944.8  examples/sec;  0.066  sec/batch)
step 96970, loss=0.63  (1954.6  examples/sec;  0.065  sec/batch)
step 96980, loss=0.54  (1876.4  examples/sec;  0.068  sec/batch)
step 96990, loss=0.54  (1966.6  examples/sec;  0.065  sec/batch)
step 97000, loss=0.70  (1999.3  examples/sec;  0.064  sec/batch)
step 97010, loss=0.46  (1898.0  examples/sec;  0.067  sec/batch)
step 97020, loss=0.68  (1852.5  examples/sec;  0.069  sec/batch)
step 97030, loss=0.55  (1897.6  examples/sec;  0.067  sec/batch)
step 97040, loss=0.66  (1883.3  examples/sec;  0.068  sec/batch)
step 97050, loss=0.55  (1933.2  examples/sec;  0.066  sec/batch)
step 97060, loss=0.57  (1954.7  examples/sec;  0.065  sec/batch)
step 97070, loss=0.57  (1930.9  examples/sec;  0.066  sec/batch)
step 97080, loss=0.62  (1907.1  examples/sec;  0.067  sec/batch)
step 97090, loss=0.64  (1844.2  examples/sec;  0.069  sec/batch)
step 97100, loss=0.47  (1927.4  examples/sec;  0.066  sec/batch)
step 97110, loss=0.45  (1827.1  examples/sec;  0.070  sec/batch)
step 97120, loss=0.60  (1804.4  examples/sec;  0.071  sec/batch)
step 97130, loss=0.54  (1978.5  examples/sec;  0.065  sec/batch)
step 97140, loss=0.65  (1980.0  examples/sec;  0.065  sec/batch)
step 97150, loss=0.64  (1892.7  examples/sec;  0.068  sec/batch)
step 97160, loss=0.75  (1916.3  examples/sec;  0.067  sec/batch)
step 97170, loss=0.64  (1950.6  examples/sec;  0.066  sec/batch)
step 97180, loss=0.59  (1969.4  examples/sec;  0.065  sec/batch)
step 97190, loss=0.60  (1852.1  examples/sec;  0.069  sec/batch)
step 97200, loss=0.43  (1804.5  examples/sec;  0.071  sec/batch)
step 97210, loss=0.49  (1956.6  examples/sec;  0.065  sec/batch)
step 97220, loss=0.62  (1835.6  examples/sec;  0.070  sec/batch)
step 97230, loss=0.72  (1928.1  examples/sec;  0.066  sec/batch)
step 97240, loss=0.53  (1845.2  examples/sec;  0.069  sec/batch)
step 97250, loss=0.57  (1946.3  examples/sec;  0.066  sec/batch)
step 97260, loss=0.49  (1908.8  examples/sec;  0.067  sec/batch)
step 97270, loss=0.61  (1954.2  examples/sec;  0.066  sec/batch)
step 97280, loss=0.61  (1878.4  examples/sec;  0.068  sec/batch)
step 97290, loss=0.57  (1726.2  examples/sec;  0.074  sec/batch)
step 97300, loss=0.62  (1894.5  examples/sec;  0.068  sec/batch)
step 97310, loss=0.48  (1993.7  examples/sec;  0.064  sec/batch)
step 97320, loss=0.59  (1693.9  examples/sec;  0.076  sec/batch)
step 97330, loss=0.49  (1867.3  examples/sec;  0.069  sec/batch)
step 97340, loss=0.70  (1895.4  examples/sec;  0.068  sec/batch)
step 97350, loss=0.52  (1892.4  examples/sec;  0.068  sec/batch)
step 97360, loss=0.45  (1924.7  examples/sec;  0.067  sec/batch)
step 97370, loss=0.67  (1833.6  examples/sec;  0.070  sec/batch)
step 97380, loss=0.70  (1929.9  examples/sec;  0.066  sec/batch)
step 97390, loss=0.67  (1829.4  examples/sec;  0.070  sec/batch)
step 97400, loss=0.49  (1911.0  examples/sec;  0.067  sec/batch)
step 97410, loss=0.55  (1986.6  examples/sec;  0.064  sec/batch)
step 97420, loss=0.67  (1823.8  examples/sec;  0.070  sec/batch)
step 97430, loss=0.44  (1880.4  examples/sec;  0.068  sec/batch)
step 97440, loss=0.57  (1629.2  examples/sec;  0.079  sec/batch)
step 97450, loss=0.52  (1872.7  examples/sec;  0.068  sec/batch)
step 97460, loss=0.66  (1948.2  examples/sec;  0.066  sec/batch)
step 97470, loss=0.61  (1849.0  examples/sec;  0.069  sec/batch)
step 97480, loss=0.63  (1904.9  examples/sec;  0.067  sec/batch)
step 97490, loss=0.59  (1942.9  examples/sec;  0.066  sec/batch)
step 97500, loss=0.61  (1966.7  examples/sec;  0.065  sec/batch)
step 97510, loss=0.54  (1961.9  examples/sec;  0.065  sec/batch)
step 97520, loss=0.59  (1820.6  examples/sec;  0.070  sec/batch)
step 97530, loss=0.59  (1933.6  examples/sec;  0.066  sec/batch)
step 97540, loss=0.59  (1869.2  examples/sec;  0.068  sec/batch)
step 97550, loss=0.52  (1884.1  examples/sec;  0.068  sec/batch)
step 97560, loss=0.52  (1892.5  examples/sec;  0.068  sec/batch)
step 97570, loss=0.58  (1965.7  examples/sec;  0.065  sec/batch)
step 97580, loss=0.48  (1857.0  examples/sec;  0.069  sec/batch)
step 97590, loss=0.52  (1915.1  examples/sec;  0.067  sec/batch)
step 97600, loss=0.63  (1913.4  examples/sec;  0.067  sec/batch)
step 97610, loss=0.67  (1863.9  examples/sec;  0.069  sec/batch)
step 97620, loss=0.57  (1887.9  examples/sec;  0.068  sec/batch)
step 97630, loss=0.47  (1870.6  examples/sec;  0.068  sec/batch)
step 97640, loss=0.69  (1918.3  examples/sec;  0.067  sec/batch)
step 97650, loss=0.61  (1912.0  examples/sec;  0.067  sec/batch)
step 97660, loss=0.63  (1711.9  examples/sec;  0.075  sec/batch)
step 97670, loss=0.64  (1800.0  examples/sec;  0.071  sec/batch)
step 97680, loss=0.50  (1812.4  examples/sec;  0.071  sec/batch)
step 97690, loss=0.56  (1831.7  examples/sec;  0.070  sec/batch)
step 97700, loss=0.64  (1926.6  examples/sec;  0.066  sec/batch)
step 97710, loss=0.61  (1876.1  examples/sec;  0.068  sec/batch)
step 97720, loss=0.62  (1910.6  examples/sec;  0.067  sec/batch)
step 97730, loss=0.70  (1952.6  examples/sec;  0.066  sec/batch)
step 97740, loss=0.65  (1898.6  examples/sec;  0.067  sec/batch)
step 97750, loss=0.64  (1818.6  examples/sec;  0.070  sec/batch)
step 97760, loss=0.70  (1968.8  examples/sec;  0.065  sec/batch)
step 97770, loss=0.62  (1906.9  examples/sec;  0.067  sec/batch)
step 97780, loss=0.60  (1897.3  examples/sec;  0.067  sec/batch)
step 97790, loss=0.46  (1980.2  examples/sec;  0.065  sec/batch)
step 97800, loss=0.52  (1955.2  examples/sec;  0.065  sec/batch)
step 97810, loss=0.54  (1632.0  examples/sec;  0.078  sec/batch)
step 97820, loss=0.67  (1914.1  examples/sec;  0.067  sec/batch)
step 97830, loss=0.50  (1875.0  examples/sec;  0.068  sec/batch)
step 97840, loss=0.67  (1919.2  examples/sec;  0.067  sec/batch)
step 97850, loss=0.64  (1876.8  examples/sec;  0.068  sec/batch)
step 97860, loss=0.61  (1846.6  examples/sec;  0.069  sec/batch)
step 97870, loss=0.56  (1859.5  examples/sec;  0.069  sec/batch)
step 97880, loss=0.54  (1999.1  examples/sec;  0.064  sec/batch)
step 97890, loss=0.70  (1842.9  examples/sec;  0.069  sec/batch)
step 97900, loss=0.69  (1825.5  examples/sec;  0.070  sec/batch)
step 97910, loss=0.65  (1919.5  examples/sec;  0.067  sec/batch)
step 97920, loss=0.65  (1860.0  examples/sec;  0.069  sec/batch)
step 97930, loss=0.63  (1885.0  examples/sec;  0.068  sec/batch)
step 97940, loss=0.44  (1952.5  examples/sec;  0.066  sec/batch)
step 97950, loss=0.56  (1914.9  examples/sec;  0.067  sec/batch)
step 97960, loss=0.56  (1981.0  examples/sec;  0.065  sec/batch)
step 97970, loss=0.51  (1922.8  examples/sec;  0.067  sec/batch)
step 97980, loss=0.46  (1958.8  examples/sec;  0.065  sec/batch)
step 97990, loss=0.77  (1581.0  examples/sec;  0.081  sec/batch)
step 98000, loss=0.62  (1845.8  examples/sec;  0.069  sec/batch)
step 98010, loss=0.52  (1895.5  examples/sec;  0.068  sec/batch)
step 98020, loss=0.51  (1879.8  examples/sec;  0.068  sec/batch)
step 98030, loss=0.62  (1941.8  examples/sec;  0.066  sec/batch)
step 98040, loss=0.65  (1883.0  examples/sec;  0.068  sec/batch)
step 98050, loss=0.54  (2028.3  examples/sec;  0.063  sec/batch)
step 98060, loss=0.69  (1948.7  examples/sec;  0.066  sec/batch)
step 98070, loss=0.58  (1857.8  examples/sec;  0.069  sec/batch)
step 98080, loss=0.58  (1827.6  examples/sec;  0.070  sec/batch)
step 98090, loss=0.66  (1898.7  examples/sec;  0.067  sec/batch)
step 98100, loss=0.59  (1851.3  examples/sec;  0.069  sec/batch)
step 98110, loss=0.56  (1941.1  examples/sec;  0.066  sec/batch)
step 98120, loss=0.51  (1850.7  examples/sec;  0.069  sec/batch)
step 98130, loss=0.55  (1899.1  examples/sec;  0.067  sec/batch)
step 98140, loss=0.45  (1888.6  examples/sec;  0.068  sec/batch)
step 98150, loss=0.57  (1909.2  examples/sec;  0.067  sec/batch)
step 98160, loss=0.72  (1887.8  examples/sec;  0.068  sec/batch)
step 98170, loss=0.66  (1824.5  examples/sec;  0.070  sec/batch)
step 98180, loss=0.51  (1968.9  examples/sec;  0.065  sec/batch)
step 98190, loss=0.49  (1968.6  examples/sec;  0.065  sec/batch)
step 98200, loss=0.51  (1934.0  examples/sec;  0.066  sec/batch)
step 98210, loss=0.67  (1905.0  examples/sec;  0.067  sec/batch)
step 98220, loss=0.69  (1982.5  examples/sec;  0.065  sec/batch)
step 98230, loss=0.63  (1873.9  examples/sec;  0.068  sec/batch)
step 98240, loss=0.45  (1854.0  examples/sec;  0.069  sec/batch)
step 98250, loss=0.65  (1904.3  examples/sec;  0.067  sec/batch)
step 98260, loss=0.48  (1783.7  examples/sec;  0.072  sec/batch)
step 98270, loss=0.90  (1920.3  examples/sec;  0.067  sec/batch)
step 98280, loss=0.45  (1801.0  examples/sec;  0.071  sec/batch)
step 98290, loss=0.70  (1988.3  examples/sec;  0.064  sec/batch)
step 98300, loss=0.60  (1856.3  examples/sec;  0.069  sec/batch)
step 98310, loss=0.50  (1909.9  examples/sec;  0.067  sec/batch)
step 98320, loss=0.53  (1762.6  examples/sec;  0.073  sec/batch)
step 98330, loss=0.66  (1877.3  examples/sec;  0.068  sec/batch)
step 98340, loss=0.66  (1834.9  examples/sec;  0.070  sec/batch)
step 98350, loss=0.55  (1801.6  examples/sec;  0.071  sec/batch)
step 98360, loss=0.48  (1907.7  examples/sec;  0.067  sec/batch)
step 98370, loss=0.69  (2072.9  examples/sec;  0.062  sec/batch)
step 98380, loss=0.64  (1866.7  examples/sec;  0.069  sec/batch)
step 98390, loss=0.55  (1893.2  examples/sec;  0.068  sec/batch)
step 98400, loss=0.45  (1970.6  examples/sec;  0.065  sec/batch)
step 98410, loss=0.54  (1912.2  examples/sec;  0.067  sec/batch)
step 98420, loss=0.63  (1852.7  examples/sec;  0.069  sec/batch)
step 98430, loss=0.53  (1908.2  examples/sec;  0.067  sec/batch)
step 98440, loss=0.45  (1885.6  examples/sec;  0.068  sec/batch)
step 98450, loss=0.44  (1910.4  examples/sec;  0.067  sec/batch)
step 98460, loss=0.58  (1980.1  examples/sec;  0.065  sec/batch)
step 98470, loss=0.66  (1838.8  examples/sec;  0.070  sec/batch)
step 98480, loss=0.60  (1873.8  examples/sec;  0.068  sec/batch)
step 98490, loss=0.48  (1923.8  examples/sec;  0.067  sec/batch)
step 98500, loss=0.52  (1912.6  examples/sec;  0.067  sec/batch)
step 98510, loss=0.59  (1859.8  examples/sec;  0.069  sec/batch)
step 98520, loss=0.45  (1890.8  examples/sec;  0.068  sec/batch)
step 98530, loss=0.64  (1877.2  examples/sec;  0.068  sec/batch)
step 98540, loss=0.63  (2092.6  examples/sec;  0.061  sec/batch)
step 98550, loss=0.47  (1888.8  examples/sec;  0.068  sec/batch)
step 98560, loss=0.57  (1873.0  examples/sec;  0.068  sec/batch)
step 98570, loss=0.71  (1800.3  examples/sec;  0.071  sec/batch)
step 98580, loss=0.53  (1818.5  examples/sec;  0.070  sec/batch)
step 98590, loss=0.55  (1969.4  examples/sec;  0.065  sec/batch)
step 98600, loss=0.60  (1841.0  examples/sec;  0.070  sec/batch)
step 98610, loss=0.65  (1956.8  examples/sec;  0.065  sec/batch)
step 98620, loss=0.54  (1870.7  examples/sec;  0.068  sec/batch)
step 98630, loss=0.48  (1875.7  examples/sec;  0.068  sec/batch)
step 98640, loss=0.59  (1958.7  examples/sec;  0.065  sec/batch)
step 98650, loss=0.51  (1899.1  examples/sec;  0.067  sec/batch)
step 98660, loss=0.67  (1913.9  examples/sec;  0.067  sec/batch)
step 98670, loss=0.53  (1970.1  examples/sec;  0.065  sec/batch)
step 98680, loss=0.59  (1823.4  examples/sec;  0.070  sec/batch)
step 98690, loss=0.53  (1832.8  examples/sec;  0.070  sec/batch)
step 98700, loss=0.62  (1866.8  examples/sec;  0.069  sec/batch)
step 98710, loss=0.63  (1890.1  examples/sec;  0.068  sec/batch)
step 98720, loss=0.78  (1981.1  examples/sec;  0.065  sec/batch)
step 98730, loss=0.60  (1887.7  examples/sec;  0.068  sec/batch)
step 98740, loss=0.60  (1848.0  examples/sec;  0.069  sec/batch)
step 98750, loss=0.70  (1956.3  examples/sec;  0.065  sec/batch)
step 98760, loss=0.66  (1872.1  examples/sec;  0.068  sec/batch)
step 98770, loss=0.68  (1932.0  examples/sec;  0.066  sec/batch)
step 98780, loss=0.54  (1915.7  examples/sec;  0.067  sec/batch)
step 98790, loss=0.59  (1951.9  examples/sec;  0.066  sec/batch)
step 98800, loss=0.68  (1857.0  examples/sec;  0.069  sec/batch)
step 98810, loss=0.58  (1863.8  examples/sec;  0.069  sec/batch)
step 98820, loss=0.58  (1908.2  examples/sec;  0.067  sec/batch)
step 98830, loss=0.44  (1786.2  examples/sec;  0.072  sec/batch)
step 98840, loss=0.56  (1585.7  examples/sec;  0.081  sec/batch)
step 98850, loss=0.51  (1945.4  examples/sec;  0.066  sec/batch)
step 98860, loss=0.60  (1910.6  examples/sec;  0.067  sec/batch)
step 98870, loss=0.69  (1915.7  examples/sec;  0.067  sec/batch)
step 98880, loss=0.44  (1862.1  examples/sec;  0.069  sec/batch)
step 98890, loss=0.50  (1887.0  examples/sec;  0.068  sec/batch)
step 98900, loss=0.73  (1934.9  examples/sec;  0.066  sec/batch)
step 98910, loss=0.66  (1906.7  examples/sec;  0.067  sec/batch)
step 98920, loss=0.55  (1834.8  examples/sec;  0.070  sec/batch)
step 98930, loss=0.60  (1914.5  examples/sec;  0.067  sec/batch)
step 98940, loss=0.61  (1856.4  examples/sec;  0.069  sec/batch)
step 98950, loss=0.75  (1845.1  examples/sec;  0.069  sec/batch)
step 98960, loss=0.61  (1873.0  examples/sec;  0.068  sec/batch)
step 98970, loss=0.46  (1963.6  examples/sec;  0.065  sec/batch)
step 98980, loss=0.67  (1911.0  examples/sec;  0.067  sec/batch)
step 98990, loss=0.67  (2003.9  examples/sec;  0.064  sec/batch)
step 99000, loss=0.61  (1835.9  examples/sec;  0.070  sec/batch)
step 99010, loss=0.55  (1949.6  examples/sec;  0.066  sec/batch)
step 99020, loss=0.55  (1880.7  examples/sec;  0.068  sec/batch)
step 99030, loss=0.56  (1911.9  examples/sec;  0.067  sec/batch)
step 99040, loss=0.56  (1959.6  examples/sec;  0.065  sec/batch)
step 99050, loss=0.63  (1883.7  examples/sec;  0.068  sec/batch)
step 99060, loss=0.67  (1596.7  examples/sec;  0.080  sec/batch)
step 99070, loss=0.68  (1900.3  examples/sec;  0.067  sec/batch)
step 99080, loss=0.53  (1927.5  examples/sec;  0.066  sec/batch)
step 99090, loss=0.57  (1957.1  examples/sec;  0.065  sec/batch)
step 99100, loss=0.48  (1857.5  examples/sec;  0.069  sec/batch)
step 99110, loss=0.64  (1890.6  examples/sec;  0.068  sec/batch)
step 99120, loss=0.68  (1931.8  examples/sec;  0.066  sec/batch)
step 99130, loss=0.57  (1938.3  examples/sec;  0.066  sec/batch)
step 99140, loss=0.62  (1908.1  examples/sec;  0.067  sec/batch)
step 99150, loss=0.75  (1823.8  examples/sec;  0.070  sec/batch)
step 99160, loss=0.76  (1831.1  examples/sec;  0.070  sec/batch)
step 99170, loss=0.48  (2006.8  examples/sec;  0.064  sec/batch)
step 99180, loss=0.54  (1853.2  examples/sec;  0.069  sec/batch)
step 99190, loss=0.48  (1895.7  examples/sec;  0.068  sec/batch)
step 99200, loss=0.82  (1881.8  examples/sec;  0.068  sec/batch)
step 99210, loss=0.56  (1983.7  examples/sec;  0.065  sec/batch)
step 99220, loss=0.60  (2006.3  examples/sec;  0.064  sec/batch)
step 99230, loss=0.51  (1881.6  examples/sec;  0.068  sec/batch)
step 99240, loss=0.52  (1938.8  examples/sec;  0.066  sec/batch)
step 99250, loss=0.54  (1956.4  examples/sec;  0.065  sec/batch)
step 99260, loss=0.47  (1980.4  examples/sec;  0.065  sec/batch)
step 99270, loss=0.56  (2021.5  examples/sec;  0.063  sec/batch)
step 99280, loss=0.53  (1898.5  examples/sec;  0.067  sec/batch)
step 99290, loss=0.42  (1956.9  examples/sec;  0.065  sec/batch)
step 99300, loss=0.52  (1843.3  examples/sec;  0.069  sec/batch)
step 99310, loss=0.64  (1931.9  examples/sec;  0.066  sec/batch)
step 99320, loss=0.66  (1872.5  examples/sec;  0.068  sec/batch)
step 99330, loss=0.56  (1882.5  examples/sec;  0.068  sec/batch)
step 99340, loss=0.60  (1898.9  examples/sec;  0.067  sec/batch)
step 99350, loss=0.67  (1983.7  examples/sec;  0.065  sec/batch)
step 99360, loss=0.53  (1981.5  examples/sec;  0.065  sec/batch)
step 99370, loss=0.67  (1930.6  examples/sec;  0.066  sec/batch)
step 99380, loss=0.58  (1974.7  examples/sec;  0.065  sec/batch)
step 99390, loss=0.45  (1897.5  examples/sec;  0.067  sec/batch)
step 99400, loss=0.72  (1773.7  examples/sec;  0.072  sec/batch)
step 99410, loss=0.49  (1774.6  examples/sec;  0.072  sec/batch)
step 99420, loss=0.70  (1871.4  examples/sec;  0.068  sec/batch)
step 99430, loss=0.50  (1855.7  examples/sec;  0.069  sec/batch)
step 99440, loss=0.79  (1896.2  examples/sec;  0.068  sec/batch)
step 99450, loss=0.70  (1924.9  examples/sec;  0.066  sec/batch)
step 99460, loss=0.69  (1895.0  examples/sec;  0.068  sec/batch)
step 99470, loss=0.78  (1935.6  examples/sec;  0.066  sec/batch)
step 99480, loss=0.43  (1915.5  examples/sec;  0.067  sec/batch)
step 99490, loss=0.57  (1991.3  examples/sec;  0.064  sec/batch)
step 99500, loss=0.57  (1885.4  examples/sec;  0.068  sec/batch)
step 99510, loss=0.57  (1833.5  examples/sec;  0.070  sec/batch)
step 99520, loss=0.55  (1962.7  examples/sec;  0.065  sec/batch)
step 99530, loss=0.87  (1920.8  examples/sec;  0.067  sec/batch)
step 99540, loss=0.55  (1822.7  examples/sec;  0.070  sec/batch)
step 99550, loss=0.53  (1930.9  examples/sec;  0.066  sec/batch)
step 99560, loss=0.63  (1904.6  examples/sec;  0.067  sec/batch)
step 99570, loss=0.51  (1987.6  examples/sec;  0.064  sec/batch)
step 99580, loss=0.64  (1858.3  examples/sec;  0.069  sec/batch)
step 99590, loss=0.49  (1865.9  examples/sec;  0.069  sec/batch)
step 99600, loss=0.50  (1906.7  examples/sec;  0.067  sec/batch)
step 99610, loss=0.61  (1853.0  examples/sec;  0.069  sec/batch)
step 99620, loss=0.44  (1853.9  examples/sec;  0.069  sec/batch)
step 99630, loss=0.64  (1944.0  examples/sec;  0.066  sec/batch)
step 99640, loss=0.43  (1874.1  examples/sec;  0.068  sec/batch)
step 99650, loss=0.51  (1941.8  examples/sec;  0.066  sec/batch)
step 99660, loss=0.45  (1932.0  examples/sec;  0.066  sec/batch)
step 99670, loss=0.74  (1894.3  examples/sec;  0.068  sec/batch)
step 99680, loss=0.69  (1916.3  examples/sec;  0.067  sec/batch)
step 99690, loss=0.59  (1982.7  examples/sec;  0.065  sec/batch)
step 99700, loss=0.48  (1897.3  examples/sec;  0.067  sec/batch)
step 99710, loss=0.58  (1843.5  examples/sec;  0.069  sec/batch)
step 99720, loss=0.58  (1879.6  examples/sec;  0.068  sec/batch)
step 99730, loss=0.69  (1945.8  examples/sec;  0.066  sec/batch)
step 99740, loss=0.58  (1943.1  examples/sec;  0.066  sec/batch)
step 99750, loss=0.65  (1916.4  examples/sec;  0.067  sec/batch)
step 99760, loss=0.75  (1842.6  examples/sec;  0.069  sec/batch)
step 99770, loss=0.49  (1873.0  examples/sec;  0.068  sec/batch)
step 99780, loss=0.53  (1908.6  examples/sec;  0.067  sec/batch)
step 99790, loss=0.45  (1876.0  examples/sec;  0.068  sec/batch)
step 99800, loss=0.51  (1848.8  examples/sec;  0.069  sec/batch)
step 99810, loss=0.49  (2002.2  examples/sec;  0.064  sec/batch)
step 99820, loss=0.57  (1844.2  examples/sec;  0.069  sec/batch)
step 99830, loss=0.48  (1914.1  examples/sec;  0.067  sec/batch)
step 99840, loss=0.53  (1921.7  examples/sec;  0.067  sec/batch)
step 99850, loss=0.60  (1834.3  examples/sec;  0.070  sec/batch)
step 99860, loss=0.58  (1834.3  examples/sec;  0.070  sec/batch)
step 99870, loss=0.50  (2053.8  examples/sec;  0.062  sec/batch)
step 99880, loss=0.73  (1902.0  examples/sec;  0.067  sec/batch)
step 99890, loss=0.65  (1929.2  examples/sec;  0.066  sec/batch)
step 99900, loss=0.70  (1886.3  examples/sec;  0.068  sec/batch)
step 99910, loss=0.55  (1914.5  examples/sec;  0.067  sec/batch)
step 99920, loss=0.77  (1941.3  examples/sec;  0.066  sec/batch)
step 99930, loss=0.54  (1929.5  examples/sec;  0.066  sec/batch)
step 99940, loss=0.61  (1895.4  examples/sec;  0.068  sec/batch)
step 99950, loss=0.50  (1949.0  examples/sec;  0.066  sec/batch)
step 99960, loss=0.54  (1972.2  examples/sec;  0.065  sec/batch)
step 99970, loss=0.60  (1885.1  examples/sec;  0.068  sec/batch)
step 99980, loss=0.58  (1921.5  examples/sec;  0.067  sec/batch)
step 99990, loss=0.50  (1979.7  examples/sec;  0.065  sec/batch)
'''
评测模型在测试集上的准确率,测试集10000个样本
(1)使用固定的batch_size,然后一个batch一个batch输入测试数据
(2)计算一共多少个batch才能将全部样本测试完毕
(3)每一个step使用session的run方法获取images_test, labels_test的batch
(4)在执行top_k_top计算模型在这个batch的top 1上预测正确的样本数,最后汇总所有正确的预测结果
'''
num_examples = 10000
import math
num_iter = int(math.ceil(num_examples / batch_size))
true_count = 0
total_sample_count = num_iter * batch_size
step = 0
while step < num_iter :
    image_batch, label_batch = sess.run([images_test, labels_test])
    predictions = sess.run([top_k_op], feed_dict={image_holder: image_batch, label_holder: label_batch})
    true_count += np.sum(predictions)
    step += 1
print true_count
print total_sample_count
8051
9984
'''
最后将准确率的评测结果计算并打印出来
'''
precision = float(true_count) / float(total_sample_count)
print true_count
print total_sample_count
print  ("分类正确率是: %.3f" % precision)
8051
9984
分类正确率是: 0.806