Base model: Llama1-7B, all 3 medical datasets

Full training

Acc: 0.47