Website - https://www.alphaai.biz

Model Name: Medical-Guide-COT-llama3.2-1B

Developed by: Alpha AI

License: apache-2.0

Finetuned from model: meta-llama/Llama-3.2-1B-Instruct

Formats available: Float16 (safetensors + GGUF-f16), GGUF quantized (q4_k_m, q5_k_m, q8_0)

Overview

Medical-Guide-COT-llama3.2-1B is a lightweight yet powerful medical reasoning model designed to produce explicit Chain of Thought (CoT) reasoning with <think>...</think> tags for transparency and clarity. Built for interpretability and performance, this model excels in structured medical question answering.

• Finetuning Objective: Supervised fine-tuning (SFT) on medical QA datasets with enforced reasoning chains.
• Instruction format: Adheres to Llama 3.2 Instruct prompting standards.
• Deployment flexibility: Offers multiple GGUF quantized variants for local, edge, or efficient inference environments.

Training Data

• Public sources: PubMedQA, MedMCQA, USMLE-type questions (filtered)

• Proprietary augmentation: Alpha AI's curated "Clinical-Cases-CoT" dataset with physician-authored reasoning chains

• Sample size: 42,000 examples (approx. 60% public / 40% private)

• Token structure:

  <think>
  Step-by-step clinical reasoning...
  </think>
  Final answer.
  

Model Specifications

Attribute	Value
Base Model	meta-llama/Llama-3.2-1B-Instruct
Model Type	Causal Language Model
Finetuned By	Alpha AI
Precision	Float16, GGUF q4_k_m / q5_k_m / q8_0
Context Length	8,192 tokens
Language	English

Intended Use

• Medical Education: Transparent QA for students (USMLE/PLAB prep)
• Prototype Decision Support: Clear reasoning steps before answers
• Research on COT Safety: Evaluation of model interpretability and hallucination control

Example Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "alpha-ai/Medical-Guide-COT-llama3.2-1B"
model = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained(model_id)

prompt = """### Question:
A 65-year-old male presents with sudden chest pain radiating to the back. Most likely diagnosis?
### Answer:
"""
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7, top_p=0.9)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Expected Output Format:

<think>
Sudden tearing chest pain suggests aortic dissection.
Hypertension is a key risk factor. Location of pain supports Stanford Type A.
</think>
Acute aortic dissection (Stanford Type A)

Limitations & Usage Warnings

• Not a clinical diagnostic tool. Use only for research or educational purposes.
• Bias & Hallucination Risk. Outputs must be validated by qualified professionals.
• Sensitive Content. Model not trained on PHI but care should be taken with input prompts.

License

Distributed under the Apache-2.0 license.

Acknowledgments

Thanks to Meta AI for Llama-3.2, the creators of open medical QA datasets, and the Alpha AI medical advisory board for domain alignment and data verification.

Website: https://www.alphaai.biz