Llama-3-8B-Hernia-Analyst-600-Patients-8k

This is a specialized, fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct, designed to function as an expert "AI Research Assistant" for analyzing patient narratives related to Abdominal Wall Hernia (AWH).

This model represents a significant upgrade over previous versions, as it was fine-tuned on a larger dataset of 600 synthetic patients and trained using the full 8192 token context window. This enables it to analyze longer, more complex patient narratives without truncation, resulting in a more accurate and comprehensive analysis.

The model's primary function is to take unstructured, free-text patient stories as input and transform them into a structured, multi-level JSON output. This output adheres to a specific Quality of Life (QoL) framework derived from clinical research, notably the work published in Hernia (2022) 26:795–808.

Model Description

The core objective of this model is to automate and standardize the process of qualitative analysis for patient-reported outcomes. It has been trained to identify and structure information across five key domains:

• Body Image
• Mental Health
• Symptoms and Function
• Interpersonal Relationships
• Employment

The model produces a detailed JSON object that includes an executive summary, a ranked list of the most prominent QoL domains, and a deep-dive analysis for each domain, identifying relevant subthemes and clinical concepts mentioned by the patient.

Intended Use

This model is intended for research and prototyping purposes only. Its primary use case is to process long-form patient narratives (e.g., from detailed interview transcripts or comprehensive questionnaires) and generate a structured, machine-readable analysis. This can be used for large-scale research, data visualization, or to assist clinicians in rapidly understanding the key QoL issues for a patient.

Disclaimer: This is not a medical device. The output should not be used for clinical diagnosis, treatment decisions, or any direct patient care without verification and interpretation by a qualified healthcare professional.

How to Use

The model expects prompts formatted in the Llama 3 Instruct template. The following Python code demonstrates how to load the model and run inference on a new patient narrative, making it a powerful tool for offline analysis.

# This installs specific, stable versions of the libraries known to work well together
# in the Colab environment.

!pip uninstall -y sentence-transformers
!pip install torch==2.3.1+cu121 torchvision==0.18.1+cu121 torchaudio==2.3.1 --index-url https://download.pytorch.org/whl/cu121
!pip install -q "transformers==4.43.2" "datasets==2.18.0" "accelerate==0.29.3" "peft==0.10.0" "bitsandbytes==0.43.1" "trl==0.8.6" "protobuf==3.20.3"
!pip install -q einops scipy sentencepiece tensorboard

# # After installation, we need to restart the runtime one time for the changes to take effect.
# # This is a standard procedure in Colab.
import os
os.kill(os.getpid(), 9)

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
import json

# --- 1. CONFIGURATION ---
# The unique ID of your powerful, 8k-context model on the Hugging Face Hub
model_name = "Laxmikant17/Llama-3-8B-Hernia-Analyst-600-Patients-8k"

# --- 2. LOAD MODEL AND TOKENIZER ---
print(f"Loading fine-tuned model: {model_name}")

# Use 4-bit quantization for efficient inference on consumer GPUs (like in Colab)
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16
)

# Load the model from the Hub with quantization
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    quantization_config=bnb_config,
    device_map="auto" # Automatically use the GPU if available
)
tokenizer = AutoTokenizer.from_pretrained(model_name)
model.eval() # Set the model to evaluation mode

print("✅ Model loaded successfully!")

# --- 3. PREPARE YOUR INPUT ---
# This can be a very long and detailed patient narrative
test_narrative = """
The pain is the worst part. It's a constant, burning sensation that gets worse when I stand for more than ten minutes. I can't even lift my grocery bags without feeling a sharp pull. I also feel deformed. I avoid looking at myself without a shirt on. I just want to feel normal again. It's been really tough mentally. I feel a sense of dread every morning when I wake up, just knowing the discomfort is waiting for me. I've become irritable and I'm not pleasant to be around, which is straining my relationship with my family.
"""

# Format the input using the exact Llama 3 Instruct template the model was trained on
instruction = "Your sole function is to be a structured data generator. Analyze the patient narrative and produce a single, valid JSON object as your only output. Adhere strictly to the required format and terminology from the provided knowledge base."
prompt = f"<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\n{instruction}\n\n**Patient Narrative:**\n{test_narrative}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

# --- 4. GENERATE ANALYSIS ---
print("\n🚀 Generating analysis...")
with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=4096, # Give the model plenty of space for its JSON output
        do_sample=False,
        pad_token_id=tokenizer.eos_token_id
    )

# Robustly extract and print the JSON from the model's response
decoded_output = tokenizer.decode(outputs, skip_special_tokens=True)
try:
    assistant_marker = 'assistant\n\n'
    assistant_response_start = decoded_output.find(assistant_marker)
    response_part = decoded_output[assistant_response_start + len(assistant_marker):].strip()
    json_start = response_part.find('{')
    json_end = response_part.rfind('}') + 1
    json_string = response_part[json_start:json_end]

    print("\n--- ✅ MODEL-GENERATED ANALYSIS ---")
    parsed_json = json.loads(json_string)
    print(json.dumps(parsed_json, indent=2))
except Exception as e:
    print(f"\n--- 🚨 ERROR: Could not parse the model's response. ---")
    print(f"Error: {e}")
    print("\nFull raw output for debugging:")
    print(decoded_output)