TRIAGE-4B-P12-SFT-RL

SFT+RL checkpoints released as part of TRIAGE, a framework that casts clinical risk prediction over irregularly sampled medical time series (ISMTS) as a reasoning task.

The model was presented in the paper TRIAGE: Dialectical Reasoning for Explainable Risk Prediction on Irregularly Sampled Medical Time Series with LLMs.

TRIAGE trains an LLM to generate dialectical reasoning over competing clinical outcomes by eliciting outcome-specific rationales. This approach mitigates risk polarization and enables a single LLM to yield continuous risk scores grounded in explicit clinical reasoning. This specific model is applied to the P12 dataset.

Code, data, & training pipeline: https://github.com/HyeongWon-Jang/TRIAGE

Model Description

Each split in this repository is RL-trained on top of an SFT warm-start and placed in its own split_N/ subfolder. Per-split RL checkpoints were selected by Validation AUPRC over the RL trajectory.

The SFT warm-start used to initialize this RL run lives on the rl_init branch of this same repo.

Quick start

from transformers import AutoModelForCausalLM, AutoTokenizer

split = "split_1"  # one of split_1 ... split_5
repo  = "Hyeongwon/TRIAGE-4B-P12-SFT-RL"

tokenizer = AutoTokenizer.from_pretrained(repo, subfolder=split)
model     = AutoModelForCausalLM.from_pretrained(repo, subfolder=split, device_map="auto")

The model expects a task-specific input/output template; for the full inference pipeline, see the linked GitHub repository.

Data

• Raw: PhysioNet Mortality Prediction Challenge 2012
• Processed (Raindrop, CC BY 4.0): figshare DOI

Further preprocessing and split-construction details are in the linked GitHub repository.

License

Model checkpoints are released under CC BY-NC 4.0 (non-commercial). Datasets remain under their respective licenses.