Skip to content

Latest commit

 

History

History
40 lines (17 loc) · 1.15 KB

README.md

File metadata and controls

40 lines (17 loc) · 1.15 KB

VANER: Biomedical Named Entity Recognition by LLM

Note: the vaner is based on LLama-2-7b-hf, you need to download this base model.

Train:

Mix multiple datasets for Instruction-Tuning with DBR

CUDA_VISIBLE_DEVICES=0 python unllama_train_vaner.py unidev_kgmix2 128 mt vaner

Single datasets Instruction-Tuning with DBR

CUDA_VISIBLE_DEVICES=0 python unllama_train_vaner.py ncbi 128 mt vaner

If you lack computational resources for training, you can also download our pre-trained models. We provide two models:

Download Links

bjy_unidev_kgmmix2:

This model has been fine-tuned with a mix of 8 datasets and has undergone DBR.

bjy_unidev_kgmmix2_nodataname:

This model has also been fine-tuned with a mix of 8 datasets, but does not specify the name of the dataset in the construction of the prompt, corresponding to the VANER_adapt method in the paper.

Evaluation:

Change 'model_id' in unllama_eval_vaner.py as the path of LLama-2-7b-hf.

Evaluation script

CUDA_VISIBLE_DEVICES=0 python unllama_eval_vaner.py ncbi vaner_unidev_kgmix2_checkpoint-number