f18ae210e1 | ||
---|---|---|
tensorflow_code | ||
.gitignore | ||
CONTRIBUTING.md | ||
Comparing TF and PT models.ipynb | ||
LICENSE | ||
README.md | ||
__init__.py | ||
convert_tf_checkpoint_to_pytorch.py | ||
convert_tf_checkpoint_to_pytorch_special_edition.py | ||
extract_features_pytorch.py | ||
input.txt | ||
modeling_pytorch.py | ||
modeling_test_pytorch.py | ||
optimization_pytorch.py | ||
optimization_test_pytorch.py | ||
requirements.txt | ||
run_classifier_pytorch.py | ||
run_squad_pytorch.py | ||
sample_text.txt | ||
tokenization_pytorch.py | ||
tokenization_test_pytorch.py |
README.md
PyTorch implementation of Google AI's BERT
Introduction
This is a PyTorch implementation of the TensorFlow code released by Google AI with the paper BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding.
Converting the TensorFlow pre-trained models to Pytorch
You can convert the pre-trained weights released by GoogleAI by calling the script convert_tf_checkpoint_to_pytorch.py
.
It takes a TensorFlow checkpoint (bert_model.ckpt
) containg the pre-trained weights and converts it to a .bin
file readable for PyTorch.
TensorFlow pre-trained models can be found in the original TensorFlow code. We give an example with the BERT-Base Uncased
model:
export BERT_BASE_DIR=/path/to/bert/uncased_L-12_H-768_A-12
export BERT_PYTORCH_DIR=/path/to/pytorch/bert/uncased_L-12_H-768_A-12
python convert_tf_checkpoint_to_pytorch.py \
--tf_checkpoint_path=$BERT_BASE_DIR/bert_model.ckpt \
--bert_config_file=$BERT_BASE_DIR/bert_config.json \
--pytorch_dump_path=$BERT_PYTORCH_DIR/pytorch_model.bin
Fine-tuning with BERT: running the examples
We showcase the same examples as in the original implementation: fine-tuning on the MRPC classification corpus and the question answering dataset SQUAD.
Before running theses examples you should download the
GLUE data by running
this script
and unpack it to some directory $GLUE_DIR
. Please also download the BERT-Base
checkpoint, unzip it to some directory $BERT_BASE_DIR
, and convert it to its PyTorch version as explained in the previous section.
This example code fine-tunes BERT-Base
on the Microsoft Research Paraphrase
Corpus (MRPC) corpus and runs in less than 10 minutes on a single K-80.
export GLUE_DIR=/path/to/glue
python run_classifier_pytorch.py \
--task_name MRPC \
--do_train \
--do_eval \
--do_lower_case \
--data_dir $GLUE_DIR/MRPC/ \
--vocab_file $BERT_BASE_DIR/vocab.txt \
--bert_config_file $BERT_BASE_DIR/bert_config.json \
--init_checkpoint $BERT_PYTORCH_DIR/pytorch_model.bin \
--max_seq_length 128 \
--train_batch_size 32 \
--learning_rate 2e-5 \
--num_train_epochs 3.0 \
--output_dir /tmp/mrpc_output_pytorch/
The next example fine-tunes BERT-Base
on the SQuAD question answering task.
The data for SQuAD can be downloaded with the following links and should be saved in a $SQUAD_DIR
directory.
export SQUAD_DIR=/path/to/SQUAD
python run_squad_pytorch.py \
--vocab_file=$BERT_BASE_DIR/vocab.txt \
--bert_config_file=$BERT_BASE_DIR/bert_config.json \
--init_checkpoint=$BERT_PYTORCH_DIR/pytorch_model.bin \
--do_train \
--train_file=$SQUAD_DIR/train-v1.1.json \
--do_predict \
--predict_file=$SQUAD_DIR/dev-v1.1.json \
--train_batch_size=12 \
--learning_rate=5e-5 \
--num_train_epochs=2.0 \
--max_seq_length=384 \
--doc_stride=128 \
--output_dir=../debug_squad/
Comparing TensorFlow and PyTorch models
We also include a small Notebook we used to verify that the conversion of the weights to PyTorch are consistent with the original TensorFlow weights. Please follow the instructions in the Notebook to run it.
Note on pre-training
The original TensorFlow code also release two scripts for pre-training BERT: create_pretraining_data.py and run_pretraining.py. As the authors notice, pre-training BERT is particularly expensive and requires TPU to run in a reasonable amout of time (see here).
We have decided not to port these scripts for now and wait for the TPU support on PyTorch (see the recent official announcement).
Requirements
The main dependencies of this code are:
- PyTorch (>= 0.4.0)
- tqdm