PMC-LLaMA
Medical LLaMA Trained on 4.8M PubMed Central Papers
Powered by Dr7.ai Medical AI Platform
PMC-LLaMA is an open medical language model from Shanghai Jiao Tong University and Shanghai AI Laboratory. It continues LLaMA pretraining on 4.8 million biomedical papers and 30,000 medical textbooks, then instruction-tunes the result for medical question answering and dialogue.
Try PMC-LLaMA-Style Medical Q&A
Ask USMLE-style or clinical questions â the demo shows the literature-grounded medical reasoning PMC-LLaMA was built for.
Unlock Full PMC-LLaMA Potential
Get unlimited medical Q&A, longer context and API access for exam preparation and research workflows.
What is PMC-LLaMA?
PMC-LLaMA is an open-source medical large language model developed by researchers at Shanghai Jiao Tong University and the Shanghai AI Laboratory (Wu et al., 2023, "PMC-LLaMA: Towards Building Open-source Language Models for Medicine"). It is available in 7B and 13B sizes.
Training happens in two stages. First, LLaMA is further pretrained on 4.8 million PubMed Central biomedical papers plus 30,000 medical textbooks (the MedC-K corpus), producing "MedLLaMA". Second, the model is instruction-tuned on MedC-I â medical QA, rationales and conversations â to become PMC-LLaMA.
The 13B model reported 56.4% on MedQA (USMLE), 56.0% on MedMCQA and 77.9% on PubMedQA, ahead of earlier open medical models and competitive with ChatGPT at the time on these benchmarks. It is a research model: it is not a medical device and its answers require expert verification.
USMLE Optimization
Released 2023 by Shanghai Jiao Tong University & Shanghai AI Laboratory (Wu et al.)
Open weights on Hugging Face (axiong/PMC_LLaMA_13B); LLaMA licence terms apply
Key Features
Knowledge injection from the biomedical literature
Core Capabilities
USMLE-style medical question answering
Continued pretraining on 4.8M PubMed Central papers
30,000 medical textbooks in the pretraining corpus
Instruction tuning on medical QA, rationales and dialogue
Available in 7B and 13B parameter sizes
Open weights for research use
Strong PubMedQA and MedMCQA results for an open model
Fits on a single high-memory GPU
Performance Benchmarks
Scores reported in the PMC-LLaMA paper for the 13B model (not measured by Dr7.ai)
MedQA (USMLE)
56.4%US Medical Licensing Examination-style questions
MedMCQA
56.0%Indian medical entrance exam questions
PubMedQA
77.9%Biomedical literature question answering
Pretraining corpus
4.8M papersPlus 30K textbooks (MedC-K), followed by MedC-I instruction data
Use Cases
Medical AI Research
An open baseline for studying domain-adaptive pretraining in medicine.
Clinical Knowledge Support
Literature-grounded answers to clinical questions for professional review.
Medical Education
USMLE-style practice questions with explanations.
How to Use PMC-LLaMA
Open 7B / 13B weights
Get the Weights
Load the 7B or 13B checkpoints from Hugging Face with transformers. LLaMA licence terms apply.
Integration Options
Use it as a chat or QA model, or fine-tune further on your own data.
- âĸ Medical question answering
- âĸ Exam-style question generation
- âĸ Further fine-tuning on institutional data
- âĸ Baseline for medical LLM research
Deployment Considerations
Plan for privacy, validation and human oversight before putting PMC-LLaMA anywhere near patient data.
- âĸ De-identify data and comply with HIPAA / GDPR
- âĸ Validate on your own institution's data before use
- âĸ Keep a licensed clinician in the loop for every output
- âĸ Document limitations and intended use for auditors
Important Considerations
Clinical Validation Required
PMC-LLaMA is a research model. Its outputs must be reviewed by qualified healthcare professionals before any clinical use.
Regulatory Compliance
The model is not a cleared or approved medical device. Ensure compliance with local healthcare regulations and data-protection rules (e.g. HIPAA, GDPR) before deployment.
PMC-LLaMA vs Other Medical Language Models
Where an open 13B medical model fits
PMC-LLaMA
Open medical LLaMA (SJTU / Shanghai AI Lab)
- âOpen weights in practical 7B / 13B sizes
- âPretrained on millions of real biomedical papers
- âInstruction-tuned for medical QA and dialogue
- âStrong benchmark scores among open models at release
- âRuns on a single high-memory GPU
- âTransparent two-stage training recipe
Research and education that need an open, literature-grounded medical LLM
Other Medical AI Models
How it compares
- ÃGeneral LLMs (GPT-4, Gemini) are stronger at open-ended reasoning but are closed and costly
- ÃLarger open models (Meditron-70B, MedGemma-27B) need far more GPU memory
- ÃText-only models cannot look at images
- ÃImage-only classifiers cannot explain their findings in language
- ÃCommercial products are validated but not inspectable
- ÃMost alternatives publish fewer reproducible benchmarks
Pick by modality (text vs. image), licence, and the hardware you actually have
Frequently Asked Questions
Common questions about PMC-LLaMA
Who developed PMC-LLaMA?
Researchers at Shanghai Jiao Tong University and the Shanghai AI Laboratory (Wu et al., 2023). It is not a Stanford or Google model.
What was PMC-LLaMA trained on?
4.8 million PubMed Central papers and 30,000 medical textbooks for continued pretraining, then medical QA, rationale and dialogue data for instruction tuning.
How does PMC-LLaMA score on USMLE questions?
The 13B model reports 56.4% on MedQA (USMLE), 56.0% on MedMCQA and 77.9% on PubMedQA.
Can I run PMC-LLaMA locally?
Yes. The 13B model fits on a single high-memory GPU, and the 7B model on consumer hardware with quantisation.
Does this demo run PMC-LLaMA itself?
The demo uses Dr7.ai's medical assistant to show literature-grounded medical Q&A. For the exact model, use the open weights.