This is a LoRA (Low-Rank Adaptation) adapter for the meta-llama/Llama-3.1-8B-Instruct model, fine-tuned for medical multiple-choice question answering with structured JSON output generation.

Model Details

Base Model

Model: meta-llama/Llama-3.1-8B-Instruct
Architecture: Llama 3.1 instruction-tuned language model
Parameters: 8B parameters
Precision: BFloat16

LoRA Configuration

Rank (r): 64
Alpha: 128
Dropout: 0.1
Target Modules: q_proj, up_proj, gate_proj, v_proj, o_proj, k_proj, down_proj
Task Type: Causal Language Modeling

Training Details

Dataset

Source: asanchez75/medical_textbooks_mcmq
Domain: Medical multiple-choice questions
Language: Primarily French medical content
Format: JSON-structured input/output pairs

Training Configuration

Epochs: 3
Learning Rate: 2e-5
Batch Size: 4 (per device)
Gradient Accumulation: 4 steps
Effective Batch Size: 16
Sequence Length: 2048 tokens
Hardware: NVIDIA A100 SXM4 40GB

Performance

Final Test Loss: 0.493
Training Time: ~19 minutes
Memory Usage: 23.4GB peak (A100 40GB)

Usage

Loading the Adapter

python

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base model
base_model = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Llama-3.1-8B-Instruct",
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "YOUR_HF_USERNAME/REPO_NAME")

Inference Example

python

import json

# Input format
input_data = {
    "question_id": 12345,
    "question": "Quel est le traitement de première intention de l'hypertension artérielle?",
    "option_a": "Inhibiteurs de l'ECA",
    "option_b": "Bêta-bloquants", 
    "option_c": "Diurétiques thiazidiques",
    "option_d": "Antagonistes calciques",
    "option_e": "Sartans"
}

# Format prompt
system_prompt = "You are a helpful medical assistant. Given a multiple-choice question in JSON format, provide the correct answer options and a detailed explanation in JSON format."

messages = [
    {"role": "system", "content": system_prompt},
    {"role": "user", "content": json.dumps(input_data, ensure_ascii=False)}
]

# Generate response
formatted_prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(formatted_prompt, return_tensors="pt")

with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.0)
    
response = tokenizer.decode(outputs[0], skip_special_tokens=True)

Expected Output Format

json

{
    "correct_options": "A, C", 
    "explanation": "Les inhibiteurs de l'ECA et les diurétiques thiazidiques sont recommandés en première intention..."
}

Model Architecture

This adapter targets the following modules in the Llama-3.1-8B-Instruct model:

Query projection (q_proj)
Key projection (k_proj)
Value projection (v_proj)
Output projection (o_proj)
Gate projection (gate_proj)
Up projection (up_proj)
Down projection (down_proj)

Limitations and Biases

Domain Specific: Optimized for French medical content
MCQ Format: Designed for structured multiple-choice questions
Medical Focus: Performance may vary on non-medical content
Language: Primarily trained on French medical terminology
Base Model: Built on instruction-tuned Llama 3.1 architecture

Citation

If you use this model, please cite:

bibtex

@misc{llama31-8b-mcq-lora,
    title={Llama-3.1-8B-Instruct LoRA Adapter for Medical MCQ JSON Generation},
    author={Your Name},
    year={2025},
    publisher={Hugging Face},
    url={https://huggingface.co/YOUR_USERNAME/REPO_NAME}
}

License

This adapter is released under the Apache-2.0 license, consistent with the base Llama-3.1-8B-Instruct model.

asanchez75/Llama3.1-8b-mcq-lora

Llama-3.1-8B-Instruct LoRA Adapter for Medical MCQ JSON Generation