Performance in T5.1.1

[toriving]

Hi, I think this is a very cool study.
I have a question about this study. Have you tried the T5.1.1?
When I experimented, it was confirmed that the performance of T5.1.1 was lower than that of T5.
Not only that, there were cases where the performance was significantly lower than full fine-tuning.
Do you have any comments regarding this?

[ylsung]

Thanks for reporting this. I didn't try T5.1.1 before. I will look into this when I am not that swamped.... Probably some time at the end of November.

[wang-yiwei]

Hello Sung 😄
Do you have any plan to integrate your tuning algorithm into hugginface's peft lib?

[ylsung]

I would love to, and I do have a simpler implementation right now. But currently, I don't have a concrete timeline for when I can contribute to the peft lib, as I have some new projects at hand now.

[qincuiping]

I would love to, and I do have a simpler implementation right now. But currently, I don't have a concrete timeline for when I can contribute to the peft lib, as I have some new projects at hand now.

We want to try other models (like llama) on your tuning algorithm, so hope you could take some time out to integrating your tuning algorithm into hugginface's peft lib. 😄