| title | emoji | sdk | sdk_version | app_file |
|---|---|---|---|---|
Llama 2 7b Med Model |
⚕️ |
gradio |
3.48.0 |
app.py |
App URL: https://rjvishwa-rajat-llama-2-7b-med-model.hf.space/
Model: https://huggingface.co/rjvishwa/Llama-2-7b-Med-Model/tree/main
| Name | Source | n | n included in training |
|---|---|---|---|
| Medical Flashcards | medalpaca/medical_meadow_medical_flashcards | 33955 | 33955 |
| Wikidoc | medalpaca/medical_meadow_wikidoc | 67704 | 10000 |
| Wikidoc Patient Information | medalpaca/medical_meadow_wikidoc_patient_information | 5942 | 5942 |
| Stackexchange academia | medalpaca/medical_meadow_stack_exchange | 40865 | 40865 |
| Stackexchange biology | medalpaca/medical_meadow_stack_exchange | 27887 | 27887 |
| Stackexchange fitness | medalpaca/medical_meadow_stack_exchange | 9833 | 9833 |
| Stackexchange health | medalpaca/medical_meadow_stack_exchange | 7721 | 7721 |
| Stackexchange bioinformatics | medalpaca/medical_meadow_stack_exchange | 5407 | 5407 |
| USMLE Self Assessment Step 1 | medalpaca/medical_meadow_usmle_self | 119 | 92 (test only) |
| USMLE Self Assessment Step 2 | medalpaca/medical_meadow_usmle_self | 120 | 110 (test only) |
| USMLE Self Assessment Step 3 | medalpaca/medical_meadow_usmle_self | 135 | 122 (test only) |
| MEDIQA | original, preprocessed | 2208 | 2208 |
| CORD-19 | original, preprocessed | 1056660 | 50000 |
| MMMLU | original, preprocessed | 3787 | 3787 |
| Pubmed Health Advice | original, preprocessed | 10178 | 10178 |
| Pubmed Causal | original, preprocessed | 2446 | 2446 |
| ChatDoctor | original | 215000 | 10000 |
| OpenAssistant | original | 9209 | 9209 |
The Llama Med Model represents an advanced suite of large language models meticulously fine-tuned for the specific purpose of medical question-answering and dialogue applications. Our primary goal is to provide a comprehensive selection of open-source language models, thereby facilitating the streamlined development of medical chatbot solutions.
The Meta Llama2 model, boasting an impressive 7 billion parameters, has been fine-tuned on a meticulously curated medical dataset. This training process was conducted on a robust infrastructure comprising two GPU nodes, each equipped with 15 GB of VRAM. Remarkably, this intricate training endeavor was successfully completed in a mere 8 hours.