TU Munich / Kather Lab · Healthcare
MedAlpaca
An instruction-tuned medical language model based on LLaMA, fine-tuned on curated medical question-answer datasets for clinical dialogue.
Overview
MedAlpaca adapts Meta's LLaMA model for medical applications through instruction fine-tuning on a curated collection of medical question-answer pairs, flashcard data, and medical dialogue datasets. It is designed to engage in medical conversations, answer clinical questions, and provide medical explanations. Available in 7B and 13B parameter variants, MedAlpaca demonstrates that smaller models can achieve meaningful medical reasoning through targeted fine-tuning.
Parameters
7B / 13B variants
Base Model
LLaMA
Training Data
Medical Q&A, flashcards, USMLE datasets
Context Window
2048 tokens
License
GPL-3.0
Capabilities
Medical question answering
Clinical dialogue generation
Medical explanation in plain language
Medical knowledge assessment
Use Cases
Building medical chatbot assistants for patient triage
Creating educational tools for medical students
Generating patient-friendly explanations of medical conditions
Providing preliminary medical information before professional consultation
Pros
- +Open-source and accessible for medical AI research
- +Multiple size variants for different deployment scenarios
- +Effective instruction following in medical context
- +Fine-tuned on diverse medical Q&A sources
Cons
- -GPL license restricts commercial use
- -Smaller context window than modern alternatives
- -May produce incorrect medical information requiring verification
- -Not validated against clinical benchmarks for patient safety
Pricing
Free and open-source for research. Self-hosting required. The 7B variant runs on consumer GPUs with 16GB+ VRAM.