For finetuning, I was supposed to use MIMIC-IV and MIMIC-V databases, but I couldn't get their access in due time. So instead, I've picked up some datasets from Kaggle on health diseases and their treatments (nothing too sophisticated) and finetuned on those datasets. It was roughly a million lines, and don't ask me what the dataset was because I don't remember at this point. Go through Kaggle and look for something similar, you'll see.
Finetuning was done in Google Collab, in the free tier setup, I'm not rich. Also, it was LoRA finetuning with Unsloth.
Most important warning: This model might be corrupted or a failure, nothing risky but probably won't work.
Uploaded finetuned model
- Developed by: Amritemperor123
- License: apache-2.0
- Finetuned from model : ruslanmv/Medical-Llama3-8B
This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.