Data and Code of our paper: "MARM: Medical Adaptive Reasoning Model".
This repository contains the codebase for SFT and RL based on LLaMA-Factory (https://github.com/hiyouga/LLaMA-Factory) and TRL (https://github.com/huggingface/trl). We use two separate conda environments for each stage.
File "data/train_sft.jsonl" is used for training in the SFT stage, while File "data/train_rl.jsonl" is used for training in the RL stage.
The "run.py" script is used to run inference on the trained model.
The "eval.py "script is used to evaluate model performance, while the "statistic.py" script is used to compute and summarize experimental results.
cd evaluate
python eval.py
python statistic.py