看代码，wenyu332你使用Torch预训练bert，那有没有TensorFlow预训练bert的参考呢？或者有没有将模型bert_weight.bin转成bert_model.ckpt的方法呢？

[lizhzh8]

在网上，搜到过将TensorFlow模型转成Torch的案例。反过来，将Torch模型转成TensorFlow模型的没有找到过。谢谢了！ 我想试试TensorFlow预训练bert。谢谢了

[wenyu332]

这个你可以在google上面找找有没有大神有转换工具，我觉得应该有开源的，但我没有尝试过。。。

[wenyu332]

找到了 可以这样转换 import numpy as np import tensorflow as tf import torch v_list=[]

for k,v in torch.load('pytorch_model.bin',map_location='cpu').items(): v_list.append(np.array(v))

def change(ckpt_path, new_ckpt_path): index = 0 with tf.Session() as sess: for var_name, _ in tf.contrib.framework.list_variables(ckpt_path): print(var_name) var = tf.contrib.framework.load_variable(ckpt_path, var_name) var = tf.Variable(v_list[index]) index+=1

    saver = tf.train.Saver()
    sess.run(tf.global_variables_initializer())
    saver.save(sess, new_ckpt_path)

ckpt_path = './chinese_L-12_H-768_A-12/bert_model.ckpt' new_ckpt_path = './bert_model.ckpt' change(ckpt_path, new_ckpt_path)

具体的方法来源是：https://github.com/thunlp/OpenCLaP/issues/3

[lizhzh8]

找到了 可以这样转换 import numpy as np import tensorflow as tf import torch v_list=[]

for k,v in torch.load('pytorch_model.bin',map_location='cpu').items(): v_list.append(np.array(v))

def change(ckpt_path, new_ckpt_path): index = 0 with tf.Session() as sess: for var_name, _ in tf.contrib.framework.list_variables(ckpt_path): print(var_name) var = tf.contrib.framework.load_variable(ckpt_path, var_name) var = tf.Variable(v_list[index]) index+=1

    saver = tf.train.Saver()
    sess.run(tf.global_variables_initializer())
    saver.save(sess, new_ckpt_path)

ckpt_path = './chinese_L-12_H-768_A-12/bert_model.ckpt' new_ckpt_path = './bert_model.ckpt' change(ckpt_path, new_ckpt_path)

具体的方法来源是：thunlp/OpenCLaP#3

感谢wenyu