PyTorch Pretrained BERT Model and vocab

This repository contains pre-trained models and vocab for:

• PyTorch Pretrained BERT

This will allow you to better manage the BERT model without having to pull the model and vocab from https://s3.amazonaws.com/models.huggingface.co/bert each time you run it.

Download PyTorch Pretrained BERT Model and vocab

before download, you can change line 10 in download_pytorch-pretrained-BERT_model_and_vocab.sh to determine the path then, run:

sh download_pytorch-pretrained-BERT_model_and_vocab.sh

Installation

This repo was tested on Python 2.7 and 3.5+ (examples are tested only on python 3.5+) and PyTorch 0.4.1/1.0.0

With pip

PyTorch pretrained bert can be installed by pip as follows:

pip install pytorch-pretrained-bert

With git

or you can use git to clone the pytorch-pretrained-BERT repository

git clone https://github.com/huggingface/pytorch-pretrained-BERT.git

this allow you to change the code!

Usage

BERT

Here is a quick-start example using BertTokenizer, BertModel and BertForMaskedLM class with Google AI's pre-trained Bert base uncased model. See the doc section below for all the details on these classes.

You can change the pretrained_model_name_or_path = '/157Dataset/data-chen.yirong/pytorch_bert_pretrained_model/bert-base-cased/' to determine the bert loading path.

First let's prepare a tokenized input with BertTokenizer

import torchfrom pytorch_pretrained_bert import BertTokenizer, BertModel, BertForMaskedLM# OPTIONAL: if you want to have more information on what's happening, activate the logger as followsimport logging
logging.basicConfig(level=logging.INFO)# Load pre-trained model tokenizer (vocabulary)pretrained_model_name_or_path = '/157Dataset/data-chen.yirong/pytorch_bert_pretrained_model/bert-base-cased/'tokenizer = BertTokenizer.from_pretrained(pretrained_model_name_or_path)# Tokenized inputtext = "[CLS] Who was Jim Henson ? [SEP] Jim Henson was a puppeteer [SEP]"tokenized_text = tokenizer.tokenize(text)# Mask a token that we will try to predict back with `BertForMaskedLM`masked_index = 8tokenized_text[masked_index] = '[MASK]'assert tokenized_text == ['[CLS]', 'who', 'was', 'jim', 'henson', '?', '[SEP]', 'jim', '[MASK]', 'was', 'a', 'puppet', '##eer', '[SEP]']# Convert token to vocabulary indicesindexed_tokens = tokenizer.convert_tokens_to_ids(tokenized_text)# Define sentence A and B indices associated to 1st and 2nd sentences (see paper)segments_ids = [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1]# Convert inputs to PyTorch tensorstokens_tensor = torch.tensor([indexed_tokens])
segments_tensors = torch.tensor([segments_ids])

Let's see how to use BertModel to get hidden states

# Load pre-trained model (weights)model = BertModel.from_pretrained(pretrained_model_name_or_path)
model.eval()# If you have a GPU, put everything on cudatokens_tensor = tokens_tensor.to('cuda')
segments_tensors = segments_tensors.to('cuda')
model.to('cuda')# Predict hidden states features for each layerwith torch.no_grad():
    encoded_layers, _ = model(tokens_tensor, segments_tensors)# We have a hidden states for each of the 12 layers in model bert-base-uncasedassert len(encoded_layers) == 12

And how to use BertForMaskedLM

# Load pre-trained model (weights)model = BertForMaskedLM.from_pretrained(pretrained_model_name_or_path)
model.eval()# If you have a GPU, put everything on cudatokens_tensor = tokens_tensor.to('cuda')
segments_tensors = segments_tensors.to('cuda')
model.to('cuda')# Predict all tokenswith torch.no_grad():
    predictions = model(tokens_tensor, segments_tensors)# confirm we were able to predict 'henson'predicted_index = torch.argmax(predictions[0, masked_index]).item()
predicted_token = tokenizer.convert_ids_to_tokens([predicted_index])[0]assert predicted_token == 'henson'