Qwen3-0.6B-Longevity

Longevity-LLM · Qwen3-0.6B

Longevity-LLM (L-LLM) is a family of compact, domain-adapted language models for interpreting heterogeneous aging biology data. This checkpoint, L-Qwen3-0.6B, is the smallest family member and was produced by full-parameter supervised fine-tuning of Qwen/Qwen3-0.6B on aging-related multi-omics and clinical data.

The family was developed jointly by Insilico Medicine and Liquid AI and accompanies the study "An Open Benchmark and Language Models for AI in Aging Biology" (Zhavoronkov et al., 2026).

Model description

• Base model: Qwen/Qwen3-0.6B (Apache-2.0)
• Model type: Decoder-only transformer, multitask instruction-tuned
• Language: English
• License: CC-BY-ND-4.0
• Context length: 32,768 tokens
• Precision: BF16

Training data

The model was trained on the shared L-LLM corpus spanning aging biology. See LongevityBench for more details

Training procedure

L-Qwen3-0.6B was trained with full-parameter supervised fine-tuning using the Hugging Face TRL library. Prompts were formatted in ChatML with a dynamic-thinking template (user turns suffixed with /think or /no_think to select response mode at inference).

Example usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "insilicomedicine/Qwen3-0.6B-Longevity"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype="bfloat16", device_map="auto")

messages = [
    {"role": "user", "content": "Given this DNA methylation profile, estimate the donor's age. /think"},
]
inputs = tok.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
out = model.generate(inputs, max_new_tokens=512)
print(tok.decode(out[0][inputs.shape[-1]:], skip_special_tokens=True))

Serving with vLLM:

vllm serve insilicomedicine/Qwen3-0.6B-Longevity --dtype bfloat16 --max-model-len 32768

Intended use and limitations

Intended for research on aging biology and omics interpretation. Outputs are model predictions, not clinical advice, and should be validated experimentally. Performance is strongest on the modalities represented in the training corpus.

License

Released under CC-BY-ND-4.0. The base model Qwen/Qwen3-0.6B is distributed under Apache-2.0.

Citation

Please cite our Cell paper if you want to use this model in your research projects.

@article{zhavoronkov_open_2026,
	title = {An open benchmark and language models for {AI} in aging biology},
	volume = {189},
	issn = {0092-8674, 1097-4172},
	url = {https://www.cell.com/cell/abstract/S0092-8674(26)00999-2},
	doi = {10.1016/j.cell.2026.08.026},
	language = {English},
	number = {19},
	urldate = {2026-09-27},
	journal = {Cell},
	publisher = {Elsevier},
	author = {Zhavoronkov, Alex and Naumov, Vladimir and Sidorenko, Denis and Aliper, Alex and Aladinskiy, Vladimir and Hasani, Ramin and Amini, Alexander and Nasto, Katerina and Reymond, Mathieu and Shayakhmetov, Rim and Miftakhutdinov, Zulfat and Gladyshev, Vadim N. and Galkin, Fedor},
	month = sep,
	year = {2026},
	keywords = {aging, aging clock, benchmark, fine-tuning, foundation model, geroprotector, large language model, longevity, multi-omics, target discovery},
	pages = {5980--5994.e8},
}

@misc{longevity_llm_qwen3_0_6b_2026,
    author       = {{Insilico Medicine} and {Liquid AI}},
    title        = {Longevity-LLM: L-Qwen3-0.6B},
    year         = 2026,
    url          = { https://huggingface.co/insilicomedicine/Qwen3-0.6B-Longevity },
    doi          = { 10.57967/hf/9872 },
    publisher    = { Hugging Face }
}