参考:使用pytorch_pretrained_bert将tensorflow模型转化为pytorch模型_sunyueqinghit的博客-CSDN博客_bert tensorflow转pytorch
1. 下载bert安装包GitHub - google-research/bert: TensorFlow code and pre-trained models for BERT
2. 下载转换的py文件transformers/convert_bert_original_tf_checkpoint_to_pytorch.py at master · huggingface/transformers · GitHub
3. 将2放入1文件夹下,打开prompt,cd到对应的文件夹的位置,方便读取2中的py文件,输入以下代码(其中文件夹位置需要对应修改),即可完成转换
python convert_bert_original_tf_checkpoint_to_pytorch.py --tf_checkpoint_path C:\Users\18742\Desktop\language_detector\bert_base_dir\multi_cased_L-12_H-768_A-12\bert_model.ckpt --bert_config_file C:\Users\18742\Desktop\language_detector\bert_base_dir\multi_cased_L-12_H-768_A-12\bert_config.json --pytorch_dump_path C:\Users\18742\Desktop\language_detector\bert_base_dir\multi_cased_L-12_H-768_A-12\pytorch_model.bin