Load Biobert pre-trained weights into Bert model with Pytorch bert hugging face run_classifier.py code

[sheetalsh456]

These are the steps I followed to get Biobert working with the existing Bert hugging face pytorch code.

1. I downloaded the pre-trained weights 'biobert_pubmed_pmc.tar.gz' from the Releases page.

2. I ran this command to convert the tf checkpoint to pytorch model

python pytorch-pretrained-BERT/pytorch_pretrained_bert/convert_tf_checkpoint_to_pytorch.py --tf_checkpoint_path="biobert/pubmed_pmc_470k/biobert_model.ckpt.index" --bert_config_file="biobert/pubmed_pmc_470k/bert_config.json" --pytorch_dump_path="biobert/pubmed_pmc_470k/Pytorch/biobert.model"

This created a file 'biobert.model' in the specified path.

3. As mentioned in this link , I compressed 'biobert.model' created above and 'biobert/pubmed_pmc_470k/bert_config.json' together into a biobert_model.tar.gz

4. I then ran the run_classifier.py of hugging face bert with the following command, using the tar.gz created above.

python pytorch-pretrained-BERT/examples/run_classifier.py --data_dir="Data/" --bert_model="biobert_model.tar.gz" --task_name="qqp" --output_dir="OutputModels/Pretrained/" --do_train --do_eval --do_lower_case

I get the error

'UnicodeDecodeError: 'utf-8' codec can't decode byte 0x8b in position 1: invalid start byte' 

in the line

tokenizer = BertTokenizer.from_pretrained(args.bert_model, do_lower_case=args.do_lower_case)

Am I doing something wrong?

I just wanted to run run_classifier.py code provided by hugging face with biobert pretrained weights in the same way that we run bert with it. Is there a way to do this?

[thomwolf]

Have you tried the solutions discussed in the other issues on this topic:

• Problems converting TF BioBERT model to PyTorch #312
• cannot load BERTAdam when restoring from BioBert #239

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[nikhilsid]

Have you tried the solutions discussed in the other issues on this topic:

• Problems converting TF BioBERT model to PyTorch #312
• cannot load BERTAdam when restoring from BioBert #239

Hi @thomwolf ,

I followed the instructions here to convert the checkpoint and then placing the files (pytorch_model.bin, bert_config.json, and vocab.txt) in one folder to compress it. I did not need to ignore any weights as mentioned in any solutions you mentioned above ( #312 or #239 )

I copied the compressed folder to the home folder of 'pytorch-transformers'. Then I ran the following command from there itself to run the example code ('examples/run_glue.py') on my data.

Then, I am trying to run the example code given by running the following command,

python ./examples/run_glue.py \ --model_type bert \ --model_name_or_path biobert.gz \ --task_name=sts-b \ --do_train \ --do_eval \ --do_lower_case \ --data_dir=$DIR \ --max_seq_length 128 \ --per_gpu_eval_batch_size=8 \ --per_gpu_train_batch_size=8

But, I get the same error as mentioned in the main discussion:

UnicodeDecodeError: 'utf-8' codec can't decode byte 0x8b in position 1: invalid start byte

at this location:

File "./examples/run_glue.py", line 424, in main config = config_class.from_pretrained(args.config_name if args.config_name else args.model_name_or_path, num_labels=num_labels, finetuning_task=args.task_name)

Can you please tell what to change.

[stefan-it]

Try to pass the extracted folder of your converted bioBERT model to the --model_name_or_path :)

Here's a short example:

• Download the BioBERT v1.1 (+ PubMed 1M) model (or any other model) from the bioBERT repo
• Extract the downloaded file, e.g. with tar -xzf biobert_v1.1_pubmed.tar.gz
• Convert the bioBERT model TensorFlow checkpoint to a PyTorch and PyTorch-Transformers compatible one: pytorch_transformers bert biobert_v1.1_pubmed/model.ckpt-1000000 biobert_v1.1_pubmed/bert_config.json biobert_v1.1_pubmed/pytorch_model.bin
• Move config mv biobert_v1.1_pubmed/bert_config.json biobert_v1.1_pubmed/config.json

Then pass the folder name to the --model_name_or_path argument. You can run this simple script to check, if everything works:

from pytorch_transformers import BertModel
model = BertModel.from_pretrained('biobert_v1.1_pubmed')

[nt-sumit]

How we load a tensor as a pretrained model in bert

[nipunsadvilkar]

@stefan-it As per new transformers-cli third command would change as follows:

transformers-cli convert --model_type bert \
--tf_checkpoint biobert_v1.1_pubmed/model.ckpt-1000000 \
--config biobert_v1.1_pubmed/bert_config.json \
--pytorch_dump_output biobert_v1.1_pubmed/pytorch_model.bin

[JessicaLopezEspejel]

Hello!
Just to complement the @stefan-it instructions in step number 3, it works for me the following code:

import os
from pytorch_pretrained_bert.convert_tf_checkpoint_to_pytorch import convert_tf_checkpoint_to_pytorch
path_bin = 'my_directory/pytorch_model.bin'
path_bert = 'my_bert_directory/'

if (not os.path.exists(path_bin)):
convert_tf_checkpoint_to_pytorch(
path_bert + "biobert_model.ckpt",
path_bert + "bert_config.json",
path_bert + "pytorch_model.bin"
)

My folder (biobert_v1.0_pmc) was originally 5 files:
3 Tensorflow checkpoint files
A vocab file
A config file

[tokarev-i-v]

• Move config mv biobert_v1.1_pubmed/bert_config.json biobert_v1.1_pubmed/config.json

Tnank you very much! It did help me!