Introduction

this project is to provide a tool to extract data item from patent materials

Usage

Install prerequisite

python3 -m pip install -r requirements.txt

label text in TANL format

python3 main.py --input_dir <path/to/directory/of/patents> [--output_dir <path/to/output/directory>] [--ckpt <path/to/customized/ckpt>]

finetune

to improve the LLM on ability to extract electrolyte related information, we use supervised finetuning to moderate pretrain LLM's behavior.

generate dataset

python3 create_dataset.py --input datasets/origin.json --output datasets/trainset.jsonl

finetuning LLM

python3 finetune.py --pretrained_ckpt <hugging/face/model/id> --sft_ckpt <path/to/ckpt> --dataset <path/to/dataset> --device (cuda|cpu)