Initial commit

README.md

@@ -0,0 +1,88 @@
+# BioM-Transformers: Building Large Biomedical Language Models with BERT, ALBERT and ELECTRA
+
+# Abstract
+
+
+The impact of design choices on the performance
+of biomedical language models recently
+has been a subject for investigation. In
+this paper, we empirically study biomedical
+domain adaptation with large transformer models
+using different design choices. We evaluate
+the performance of our pretrained models
+against other existing biomedical language
+models in the literature. Our results show that
+we achieve state-of-the-art results on several
+biomedical domain tasks despite using similar
+or less computational cost compared to other
+models in the literature. Our findings highlight
+the significant effect of design choices on
+improving the performance of biomedical language
+models.
+
+# Model Description
+- This model is fine-tuned on the SQuAD2.0 dataset and then on the BioASQ8B-Factoid training dataset. We convert the BioASQ8B-Factoid training dataset to SQuAD1.1 format and train and evaluate our model (BioM-ELECTRA-Base-SQuAD2) on this dataset.
+
+- Please note that this version (PyTorch) is different than what we used in our participation in BioASQ9B (TensorFlow with Layer-Wise Decay). We combine all five batches of the BioASQ8B testing dataset as one dev.json file.
+
+- Below is unofficial results of our models against the original ELECTRA base and large :
+
+
+| Model | Exact Match (EM) | F1 Score  | 
+| ---  | --- | --- |
+| ELECTRA-Base-SQuAD2 | 61.89 | 74.39 |
+| BioM-ELECTRA-Base-SQuAD2-BioASQ8B  | 70.31 | 80.90 |
+| ELECTRA-Large-SQuAD2 | 67.36 | 78.90 |
+| BioM-ELECTRA-Large-SQuAD2-BioASQ8B | 74.31 | 84.72 |
+
+
+
+Training script
+
+```python
+python3 run_squad.py --model_type electra --model_name_or_path sultan/BioM-ELECTRA-Base-SQuAD2 \
+--train_file BioASQ8B/train.json \
+--predict_file BioASQ8B/dev.json \
+--do_lower_case \
+--do_train \
+--do_eval \
+--threads 20 \
+--version_2_with_negative \
+--num_train_epochs 3 \
+--learning_rate 3e-5 \
+--max_seq_length 512 \
+--doc_stride 128 \
+--per_gpu_train_batch_size 8 \
+--gradient_accumulation_steps 2  \
+--per_gpu_eval_batch_size 128   \
+--logging_steps 50 \
+--save_steps 5000 \
+--fp16 \
+--fp16_opt_level O1 \
+--overwrite_output_dir \
+--output_dir BioM-ELECTRA-Base-SQuAD-BioASQ \
+--overwrite_cache
+```
+
+# Acknowledgment
+
+We would like to acknowledge the support we have from Tensorflow Research Cloud (TFRC) team to grant us access to TPUv3 units.
+
+# Citation
+
+
+```bibtex
+@inproceedings{alrowili-shanker-2021-biom,
+title = "{B}io{M}-Transformers: Building Large Biomedical Language Models with {BERT}, {ALBERT} and {ELECTRA}",
+author = "Alrowili, Sultan and
+Shanker, Vijay",
+booktitle = "Proceedings of the 20th Workshop on Biomedical Language Processing",
+month = jun,
+year = "2021",
+address = "Online",
+publisher = "Association for Computational Linguistics",
+url = "https://www.aclweb.org/anthology/2021.bionlp-1.24",
+pages = "221--227",
+abstract = "The impact of design choices on the performance of biomedical language models recently has been a subject for investigation. In this paper, we empirically study biomedical domain adaptation with large transformer models using different design choices. We evaluate the performance of our pretrained models against other existing biomedical language models in the literature. Our results show that we achieve state-of-the-art results on several biomedical domain tasks despite using similar or less computational cost compared to other models in the literature. Our findings highlight the significant effect of design choices on improving the performance of biomedical language models.",
+}
+```

config.json

@@ -0,0 +1,28 @@
+{
+  "_name_or_path": "sultan/BioM-ELECTRA-Large-SQuAD2",
+  "architectures": [
+    "ElectraForQuestionAnswering"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "embedding_size": 1024,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "electra",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "summary_activation": "gelu",
+  "summary_last_dropout": 0.1,
+  "summary_type": "first",
+  "summary_use_proj": true,
+  "torch_dtype": "float32",
+  "transformers_version": "4.9.0.dev0",
+  "type_vocab_size": 2,
+  "vocab_size": 28895
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a735bb061833678abb0c0be4c4a6d2db8c75c8b66fd77e138f3d8679d99602f
+size 1329876465