mistral-7b-pubmedqa-qlora

Mistral-7B PubMedQA QLoRA Adapter

Fine-tuned LoRA adapter for Mistral-7B-Instruct-v0.2 on biomedical question answering.

Model Details

• Base model: mistralai/Mistral-7B-Instruct-v0.2
• Fine-tuning method: QLoRA (4-bit NF4 quantization, bitsandbytes)
• LoRA config: r=16, alpha=32, target modules: q/k/v/o projections
• Trainable parameters: 13.6M (0.19% of 7.2B total)
• Training data: PubMedQA (975 examples, 2 epochs, 244 steps)
• Training loss: 1.57 → 0.23

Performance

• ROUGE-L improvement: 49.3% over base Mistral-7B on held-out biomedical Q&A

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from peft import PeftModel
import torch

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_quant_type="nf4"
)

base_model = AutoModelForCausalLM.from_pretrained(
    "mistralai/Mistral-7B-Instruct-v0.2",
    quantization_config=bnb_config,
    device_map="auto"
)
model = PeftModel.from_pretrained(base_model, "niranjan6994/mistral-7b-pubmedqa-qlora")
tokenizer = AutoTokenizer.from_pretrained("niranjan6994/mistral-7b-pubmedqa-qlora")

Dataset

PubMedQA — biomedical research question answering dataset with yes/no/maybe decisions and long-form answers grounded in PubMed abstracts.