Model Card for anuragphukan5/PubMed-Llama-3.1-8B-Instruct

A LoRA‐adapter fine-tuned version of LLaMA-3.1-8B-Instruct on the full PubMedQA semantic abstract QA dataset (≈250 000 examples), quantized to 4-bit for speed and memory efficiency.

---

Model Details

Model Description

This model is an 8-billion-parameter causal language model based on Meta’s LLaMA-3.1-8B-Instruct, fine-tuned via LoRA adapters (rank 8, α = 16, dropout = 0.05) on a merged PubMedQA dataset containing expert-annotated, artificially generated, and filtered PubMed abstract QA pairs. Quantization to 4-bit (NF4, double quant) with BF16 compute was used to accelerate training and inference on A100 GPUs.

• Developed by: Anurag Phukan
• Shared by: anuragphukan5 on Hugging Face
• Model type: Causal Language Model (decoder-only)
• Language: English (biomedical)
• License: Creative Commons Attribution 4.0 International (CC BY 4.0)
• Finetuned from: meta-llama/Llama-3.1-8B-Instruct

Model Sources

• Repository: https://huggingface.co/anuragphukan5/PubMed-Llama-3.1-8B-Instruct
• Dataset Card: https://huggingface.co/datasets/anuragphukan5/PubMedQA_SemanticAbstractQA

---

Uses

Direct Use

• Biomedical Q&A: Generate concise semantic abstracts in response to biomedical research questions, given a PubMed abstract context.
• Literature summarization: Extract focused summaries of medical abstracts.

Downstream Use

• Clinical decision support (research stage): May be integrated into systems to assist in quick literature lookup—not for direct patient care.
• Medical education: Generate explanatory answers for students.

Out-of-Scope Use

• Generating non-biomedical or general-purpose content.
• Clinical or treatment advice without professional oversight.
• Any use where factual accuracy cannot be verified against authoritative sources.

---

Bias, Risks, and Limitations

• Biases: May reflect publication and author biases present in PubMed abstracts.
• Hallucinations: As a generative model, it can produce plausible-sounding but incorrect answers.
• Not medical advice: Outputs should be validated by qualified experts.

Recommendations

• Always verify generated answers against the original abstracts.
• Do not rely on this model for clinical decision-making.
• Combine with retrieval and fact-checking pipelines for higher reliability.

---

How to Get Started

from transformers import pipeline

pipe = pipeline(
    "text-generation",
    model="anuragphukan5/PubMed-Llama-3.1-8B-Instruct",
    tokenizer="meta-llama/Llama-3.1-8B-Instruct",
    device=0
)

prompt = (
    "### Question:\nWhat are the molecular mechanisms of drug resistance in M. tuberculosis?\n\n"
    "### Context:\n<PubMed abstract text here>\n\n"
    "### Answer:\n"
)
print(pipe(prompt, max_new_tokens=128)[0]["generated_text"])