Can this method improve data-sufficient fine-tuning?
guotong1988 opened this issue · comments
百折不回头 commented
I am new to LM-BFF.
Thank you very much!
Tianyu Gao commented
Hi guotong, can you specify your question? What do you mean by improving "data-efficient" fine-tuning?
百折不回头 commented
My fault. I mean the fine-tune data count is large enough.
Tianyu Gao commented
I see. We haven't tried fine-tuning this on the full data, but if you see Figure 3, it shows the trends for LM-BFF and fine-tuning with more training data, and they tend to converge in the end.