README

Evaluate pre-trained transformer on several biomedical tasks

1. Relation Extraction
   • ChemProt
2. Named Entity Recognition
   • BC5CDR-Chem
   • BC5CDR-Disease
3. Semantic Similarity
   • BIOSSES

1. BlueBERT

The detailed instructions on running the train/eval are present in the repository

Example 1 for NER:

BlueBERT_DIR="checkpoints/NCBI_BERT_pubmed_uncased"
DATASET_DIR="../bert_data/BC5CDR/chem/"
OUTPUT_DIR="checkpoints/NCBI_BERT_pubmed_uncased-BC5CDR-chem/"

python -m bluebert.run_bluebert_ner \
  --do_prepare=true \
  --do_train=true \
  --do_eval=true \
  --do_predict=true \
  --task_name="bc5cdr" \
  --vocab_file=$BlueBERT_DIR/vocab.txt \
  --bert_config_file=$BlueBERT_DIR/bert_config.json \
  --init_checkpoint=$BlueBERT_DIR/bert_model.ckpt \
  --num_train_epochs=30.0 \
  --do_lower_case=true \
  --data_dir=$DATASET_DIR \
  --output_dir=$OUTPUT_DIR

2. BioBERT

See the instructions in this repository

3. BioBERT, ClinicalBERT, SciBERT, BlueBERT

See the instructions in this repository

4. T5

Checkout subfolders bc5cdr for NER and chemprot for Rel-Ext.