hugo/protocolos-clinicos-br-rl-1gen-14b

RL adapter (LoRA, 1 generator) — CPT + RL

GRPO LoRA adapter (r=32, α=64) trained on top of the CPT (1 generator: GPT-4.1-mini) model, using the HealthBench-BR train split as reward.

• —Base model: hugo/protocolos-clinicos-br-cpt-1gen-14b
• —Type: LoRA adapter (PEFT)

Test-split accuracy

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base = AutoModelForCausalLM.from_pretrained("hugo/protocolos-clinicos-br-cpt-1gen-14b", torch_dtype="auto", device_map="auto")
tok  = AutoTokenizer.from_pretrained("hugo/protocolos-clinicos-br-rl-1gen-14b")
model = PeftModel.from_pretrained(base, "hugo/protocolos-clinicos-br-rl-1gen-14b")

Intended use & limitations

Research model for studying domain adaptation of LLMs to Brazilian clinical guidelines. Not a certified medical device. Even at the best accuracy reported in the paper, residual errors may involve consequential details (dosages, contraindications). Use only under qualified professional supervision.

Citation

See the paper and code at the project repository:

Code & paper: https://github.com/hugoabonizio/clinical-protocols-br