Use transformers to convert TF model to pytorch model

scene

Use tensorflow take TF The model is transformed into PyTorch Model

step

Get the following three files ：

• src/transformers/models/bert/convert_bert_original_tf2_checkpoint_to_pytorch.py： This is the tensorflow2.x Bert The model is transformed into PyTorch Available models .
• src/transformers/models/bert/modeling_bert.py：Bert Model usage examples .
• BERT-Base, Multilingual Cased (New, recommended)： be based on Bert Multi - language pre - training model .

It is assumed here that PyTorch 了 .
Start to transform TF2 Model bit PyTorch Model ：

#  Installation dependency 
pip3 install tensorflow transformers
export BERT_BASE_DIR=~/Downloads/nlp_bert/multi_cased_L-12_H-768_A-12
transformers-cli convert --model_type bert \
  --tf_checkpoint $BERT_BASE_DIR/bert_model.ckpt \
  --config $BERT_BASE_DIR/bert_config.json \
  --pytorch_dump_output $BERT_BASE_DIR/pytorch_model.bin

there pytorch_model.bin Namely TF2 Has been trained to transform the model PyTorch Model .

Reference resources ：

• Google Pre training model google-research/bert
• Converting Tensorflow Checkpoints