Sepideh2027/PathogenAgentAI-BioGPT-LoRA

Year: 2024–2025
License: MIT
Author: Sepideh Moafi

PathogenAgentAI-BioGPT LoRA Adapter

Model Description

A parameter-efficient LoRA adapter for BioGPT, developed as part of the PathogenAgentAI research software project. Fine-tuned on ClinVar-derived instruction data for biomedical language-model applications.

Model Details

• Model type: LoRA adapter (PEFT-compatible)
• Base model: microsoft/biogpt
• Framework: PEFT 0.19.1
• Adapter size: ~12.6 MB
• Precision: BF16
• Format: Safetensors
• License: MIT

LoRA Configuration

• Rank (r): 16
• LoRA alpha: 32
• LoRA dropout: 0.05
• Target modules: q_proj, k_proj, v_proj, o_proj
• Task type: CAUSAL_LM

Intended Use

• Parameter-efficient adaptation of BioGPT for biomedical text generation
• Research on biomedical language models
• Experimentation with ClinVar-derived variant information
• Reproducible research and model adaptation experiments

Not intended for clinical diagnosis, treatment decisions, or direct clinical use.

How to Get Started

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("microsoft/biogpt")
model = PeftModel.from_pretrained(base_model, "Sepideh2027/PathogenAgentAI-BioGPT-LoRA")
Training Details

· Training data: ~20,000 ClinVar-derived instruction-style examples (16k/2k/2k train/validation/test split)
· Training procedure: LoRA, BF16 mixed precision
· Hardware: NVIDIA T4 GPU

Limitations

· Inherits all limitations of the base BioGPT model.
· Not validated for clinical use.
· Generated outputs should be independently verified against authoritative biomedical sources.

Related Resources

· Fine-tuned model: Sepideh2027/biogpt-clinvar-finetuned
· Instruction-tuning dataset: Sepideh2027/Agent

Citation

@software{Moafi2025pathogenagent,
  author = {Moafi, Sepideh},
  title = {PathogenAgentAI-BioGPT LoRA Adapter},
  year = {2025},
  url = {https://huggingface.co/Sepideh2027/PathogenAgentAI-BioGPT-LoRA}
}

BioGPT-ClinVar: Parameter-Efficient Fine-Tuning of a Biomedical Foundation Model. Research Square, 2026. DOI: 10.21203/rs.3.rs-10196893/v1

Contact

Sepideh Moafi
GitHub: AIResearcher20
Hugging Face: Sepideh2027