.pth文件怎么转换为.ckpt文件

import tensorflow.compat.v1 as tf

import torch

def convert(bin_path, ckptpath):

    with  tf.Session() as sess:

        for var_name, value in torch.load(bin_path, map_location='cpu').items():

            print(var_name)  # 输出权重文件中的变量名

            tf.Variable(initial_value=value, name=var_name)

        saver = tf.train.Saver()

        sess.run(tf.global_variables_initializer())

        saver.save(sess, ckpt_path)

bin_path = '/content//attn/DAMSMencoders/bird/t200.pth'

ckpt_path = '/content/attn/models/new_model.ckpt'

convert(bin_path, ckpt_path)

 

 

原文链接：https://segmentfault.com/a/1190000040566987