Open
Description
I’m very interested in your work. However, I have one question I would like to ask. I noticed that Alpaca-style prompts were used during fine-tuning, but the same format was not applied during inference. Is this approach reasonable? Could this discrepancy potentially lead to suboptimal fine-tuning performance?
Metadata
Metadata
Assignees
Labels
No labels