Shanghai Jiao Tong University · Healthcare

PMC-LLaMA

A medical language model trained on 4.8 million biomedical papers from PubMed Central for medical knowledge comprehension and reasoning.

Overview

PMC-LLaMA is built by further pre-training the LLaMA model on 4.8 million biomedical academic papers from PubMed Central. The model is then instruction-tuned on medical question-answer pairs to enhance its ability to follow medical queries. It demonstrates strong performance on medical benchmarks and represents an effort to inject comprehensive biomedical literature knowledge into an open-source language model foundation.

Parameters

7B / 13B variants

Base Model

LLaMA

Training Data

4.8M PubMed Central papers

Context Window

2048 tokens

License

Apache 2.0

Capabilities

Medical knowledge comprehension

Biomedical question answering

Medical literature analysis

Clinical reasoning support

Medical text generation

Use Cases

Answering medical knowledge questions for research and education

Analyzing and summarizing biomedical research papers

Supporting evidence-based clinical decision processes

Generating biomedical content for review and synthesis

Pros

  • +Trained on one of the largest biomedical corpora available
  • +Strong medical knowledge comprehension from full-text papers
  • +Open-source with permissive licensing
  • +Both pre-trained and instruction-tuned variants available

Cons

  • -Based on older LLaMA architecture with limited context
  • -May lag behind newer models on general reasoning tasks
  • -Biomedical focus means limited non-medical capabilities
  • -Requires careful evaluation before any clinical application

Pricing

Free and open-source. Available on Hugging Face for download. Self-hosting costs similar to other 7B/13B models.

Related Models