---

license: apache-2.0 base_model: Qwen/Qwen3-4B-Instruct-2507 tags:

• biology
• genomics
• dna
• regulatory-element
• enhancer
• chain-of-thought
• qwen3 library_name: transformers pipeline_tag: text-generation

R3LM

Download

pip install huggingface_hub

python -c "
from huggingface_hub import snapshot_download
snapshot_download(
    repo_id='DuanYi/R3LM',
    repo_type='model',
    local_dir='./model',
)
"

Quick inference

Requires LLaMA-Factory with the custom chatml_with_n template registered — see configs/TEMPLATE.md.

git clone https://github.com/DuanYi516/R3LM.git && cd R3LM
pip install -e .
pip install llamafactory vllm

# download weights into ./model/ (command above)

bash scripts/generate_rationales.sh \
  K562 \
  model/R3LM_K562 \
  /path/to/rcc-sharegpt.jsonl \
  outputs/K562/self-generated-rationales.jsonl

Decoding defaults: temperature 0.95, top_p 0.7, max_new_tokens 1024.

Related resources

• Code: github.com/DuanYi516/R3LM
• Dataset: DuanYi/Cre-ReasonBench
• Base model: Qwen/Qwen3-4B-Instruct-2507

Citation

@inproceedings{Duan2026Biological,
  author    = {Yi Duan and Zhao Yang and Jiwei Zhu and Ying Ba and Chuan Cao and Bing Su},
  title     = {Biological Reasoning-Informed Regression for Interpretable Regulatory {DNA} Activity Prediction},
  booktitle = {Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2 (KDD 2026)},
  year      = {2026},
  doi       = {10.1145/3770855.3818836},
}

License

Apache License 2.0 — same as the R3LM codebase.