Skip to content

Latest commit

 

History

History
 
 

bert

README.md

BERT (Bidirectional Encoder Representations from Transformers)

The academic paper which describes BERT in detail and provides full results on a number of tasks can be found here: https://arxiv.org/abs/1810.04805.

This repository contains TensorFlow 2.x implementation for BERT.

Contents

Pre-trained Models

Our current released checkpoints are exactly the same as TF 1.x official BERT repository, thus inside BertConfig, there is backward_compatible=True. We are going to release new pre-trained checkpoints soon.

Access to Pretrained Checkpoints

We provide checkpoints that are converted from google-research/bert, in order to keep consistent with BERT paper.

Note: We have switched BERT implementation to use Keras functional-style networks in nlp/modeling. The new checkpoints are:

Here are the stable model checkpoints work with v2.0 release.

Note: these checkpoints are not compatible with the current master examples.

We recommend to host checkpoints on Google Cloud storage buckets when you use Cloud GPU/TPU.

Restoring from Checkpoints

tf.train.Checkpoint is used to manage model checkpoints in TF 2. To restore weights from provided pre-trained checkpoints, you can use the following code:

init_checkpoint='the pretrained model checkpoint path.'
model=tf.keras.Model() # Bert pre-trained model as feature extractor.
checkpoint = tf.train.Checkpoint(model=model)
checkpoint.restore(init_checkpoint)

Checkpoints featuring native serialized Keras models (i.e. model.load()/load_weights()) will be available soon.

Set Up

export PYTHONPATH="$PYTHONPATH:/path/to/models"

Install tf-nightly to get latest updates:

pip install tf-nightly-gpu

With TPU, GPU support is not necessary. First, you need to create a tf-nightly TPU with ctpu tool:

ctpu up -name <instance name> --tf-version=”nightly”

Second, you need to install TF 2 tf-nightly on your VM:

pip install tf-nightly

Warning: More details TPU-specific set-up instructions and tutorial should come along with official TF 2.x release for TPU. Note that this repo is not officially supported by Google Cloud TPU team yet until TF 2.1 released.

Process Datasets

Pre-training

There is no change to generate pre-training data. Please use the script ../data/create_pretraining_data.py which is essentially branched from BERT research repo to get processed pre-training data and it adapts to TF2 symbols and python3 compatibility.

Fine-tuning

To prepare the fine-tuning data for final model training, use the ../data/create_finetuning_data.py script. Resulting datasets in tf_record format and training meta data should be later passed to training or evaluation scripts. The task-specific arguments are described in following sections:

  • GLUE

Users can download the GLUE data by running this script and unpack it to some directory $GLUE_DIR.

export GLUE_DIR=~/glue
export BERT_BASE_DIR=gs://cloud-tpu-checkpoints/bert/keras_bert/uncased_L-24_H-1024_A-16

export TASK_NAME=MNLI
export OUTPUT_DIR=gs://some_bucket/datasets
python ../data/create_finetuning_data.py \
 --input_data_dir=${GLUE_DIR}/${TASK_NAME}/ \
 --vocab_file=${BERT_BASE_DIR}/vocab.txt \
 --train_data_output_path=${OUTPUT_DIR}/${TASK_NAME}_train.tf_record \
 --eval_data_output_path=${OUTPUT_DIR}/${TASK_NAME}_eval.tf_record \
 --meta_data_file_path=${OUTPUT_DIR}/${TASK_NAME}_meta_data \
 --fine_tuning_task_type=classification --max_seq_length=128 \
 --classification_task_name=${TASK_NAME}
  • SQUAD

The SQuAD website contains detailed information about the SQuAD datasets and evaluation.

The necessary files can be found here:

export SQUAD_DIR=~/squad
export SQUAD_VERSION=v1.1
export BERT_BASE_DIR=gs://cloud-tpu-checkpoints/bert/keras_bert/uncased_L-24_H-1024_A-16
export OUTPUT_DIR=gs://some_bucket/datasets

python ../data/create_finetuning_data.py \
 --squad_data_file=${SQUAD_DIR}/train-${SQUAD_VERSION}.json \
 --vocab_file=${BERT_BASE_DIR}/vocab.txt \
 --train_data_output_path=${OUTPUT_DIR}/squad_${SQUAD_VERSION}_train.tf_record \
 --meta_data_file_path=${OUTPUT_DIR}/squad_${SQUAD_VERSION}_meta_data \
 --fine_tuning_task_type=squad --max_seq_length=384

Fine-tuning with BERT

Cloud GPUs and TPUs

  • Cloud Storage

The unzipped pre-trained model files can also be found in the Google Cloud Storage folder gs://cloud-tpu-checkpoints/bert/keras_bert. For example:

export BERT_BASE_DIR=gs://cloud-tpu-checkpoints/bert/keras_bert/uncased_L-24_H-1024_A-16
export MODEL_DIR=gs://some_bucket/my_output_dir

Currently, users are able to access to tf-nightly TPUs and the following TPU script should run with tf-nightly.

  • GPU -> TPU

Just add the following flags to run_classifier.py or run_squad.py:

  --distribution_strategy=tpu
  --tpu=grpc://${TPU_IP_ADDRESS}:8470

Sentence and Sentence-pair Classification Tasks

This example code fine-tunes BERT-Large on the Microsoft Research Paraphrase Corpus (MRPC) corpus, which only contains 3,600 examples and can fine-tune in a few minutes on most GPUs.

We use the BERT-Large (uncased_L-24_H-1024_A-16) as an example throughout the workflow. For GPU memory of 16GB or smaller, you may try to use BERT-Base (uncased_L-12_H-768_A-12).

export BERT_BASE_DIR=gs://cloud-tpu-checkpoints/bert/keras_bert/uncased_L-24_H-1024_A-16
export MODEL_DIR=gs://some_bucket/my_output_dir
export GLUE_DIR=gs://some_bucket/datasets
export TASK=MRPC

python run_classifier.py \
  --mode='train_and_eval' \
  --input_meta_data_path=${GLUE_DIR}/${TASK}_meta_data \
  --train_data_path=${GLUE_DIR}/${TASK}_train.tf_record \
  --eval_data_path=${GLUE_DIR}/${TASK}_eval.tf_record \
  --bert_config_file=${BERT_BASE_DIR}/bert_config.json \
  --init_checkpoint=${BERT_BASE_DIR}/bert_model.ckpt \
  --train_batch_size=4 \
  --eval_batch_size=4 \
  --steps_per_loop=1 \
  --learning_rate=2e-5 \
  --num_train_epochs=3 \
  --model_dir=${MODEL_DIR} \
  --distribution_strategy=mirrored

To use TPU, you only need to switch distribution strategy type to tpu with TPU information and use remote storage for model checkpoints.

export BERT_BASE_DIR=gs://cloud-tpu-checkpoints/bert/keras_bert/uncased_L-24_H-1024_A-16
export TPU_IP_ADDRESS='???'
export MODEL_DIR=gs://some_bucket/my_output_dir
export GLUE_DIR=gs://some_bucket/datasets

python run_classifier.py \
  --mode='train_and_eval' \
  --input_meta_data_path=${GLUE_DIR}/${TASK}_meta_data \
  --train_data_path=${GLUE_DIR}/${TASK}_train.tf_record \
  --eval_data_path=${GLUE_DIR}/${TASK}_eval.tf_record \
  --bert_config_file=$BERT_BASE_DIR/bert_config.json \
  --init_checkpoint=$BERT_BASE_DIR/bert_model.ckpt \
  --train_batch_size=32 \
  --eval_batch_size=32 \
  --learning_rate=2e-5 \
  --num_train_epochs=3 \
  --model_dir=${MODEL_DIR} \
  --distribution_strategy=tpu \
  --tpu=grpc://${TPU_IP_ADDRESS}:8470

SQuAD 1.1

The Stanford Question Answering Dataset (SQuAD) is a popular question answering benchmark dataset. See more in SQuAD website.

We use the BERT-Large (uncased_L-24_H-1024_A-16) as an example throughout the workflow. For GPU memory of 16GB or smaller, you may try to use BERT-Base (uncased_L-12_H-768_A-12).

export BERT_BASE_DIR=gs://cloud-tpu-checkpoints/bert/keras_bert/uncased_L-24_H-1024_A-16
export SQUAD_DIR=gs://some_bucket/datasets
export MODEL_DIR=gs://some_bucket/my_output_dir
export SQUAD_VERSION=v1.1

python run_squad.py \
  --input_meta_data_path=${SQUAD_DIR}/squad_${SQUAD_VERSION}_meta_data \
  --train_data_path=${SQUAD_DIR}/squad_${SQUAD_VERSION}_train.tf_record \
  --predict_file=${SQUAD_DIR}/dev-v1.1.json \
  --vocab_file=${BERT_BASE_DIR}/vocab.txt \
  --bert_config_file=$BERT_BASE_DIR/bert_config.json \
  --init_checkpoint=$BERT_BASE_DIR/bert_model.ckpt \
  --train_batch_size=4 \
  --predict_batch_size=4 \
  --learning_rate=8e-5 \
  --num_train_epochs=2 \
  --model_dir=${MODEL_DIR} \
  --distribution_strategy=mirrored

To use TPU, you need switch distribution strategy type to tpu with TPU information.

export BERT_BASE_DIR=gs://cloud-tpu-checkpoints/bert/keras_bert/uncased_L-24_H-1024_A-16
export TPU_IP_ADDRESS='???'
export MODEL_DIR=gs://some_bucket/my_output_dir
export SQUAD_DIR=gs://some_bucket/datasets
export SQUAD_VERSION=v1.1

python run_squad.py \
  --input_meta_data_path=${SQUAD_DIR}/squad_${SQUAD_VERSION}_meta_data \
  --train_data_path=${SQUAD_DIR}/squad_${SQUAD_VERSION}_train.tf_record \
  --predict_file=${SQUAD_DIR}/dev-v1.1.json \
  --vocab_file=${BERT_BASE_DIR}/vocab.txt \
  --bert_config_file=$BERT_BASE_DIR/bert_config.json \
  --init_checkpoint=$BERT_BASE_DIR/bert_model.ckpt \
  --train_batch_size=32 \
  --learning_rate=8e-5 \
  --num_train_epochs=2 \
  --model_dir=${MODEL_DIR} \
  --distribution_strategy=tpu \
  --tpu=grpc://${TPU_IP_ADDRESS}:8470

The dev set predictions will be saved into a file called predictions.json in the model_dir:

python $SQUAD_DIR/evaluate-v1.1.py $SQUAD_DIR/dev-v1.1.json ./squad/predictions.json