Spaces:
Running
on
CPU Upgrade
None of the UL2 based models have been evaluated.
I like to see Flan-T5 compared to the rest.
Hi @felixz , are they in the request queue?
@clefourrier
So I don't see google/flan-ul2 in the request queue but when i tried to submit it I got error saying "Model has already been submitted".
I guess flan-ul2 is the model from Google wich is best model claimed in their paper.
An older model from last year google/flan-t5-xxl would be nice to see and it has been in queue for a while.
Thank you for reporting, investigating today
@felixz
After further investigation, it would seem we are not supporting them at the moment.
We will do our best to add support fast :)
I guess I understand why.. without some more public research and continued innovation by Google there is no momentum to revive the T5 architecture. The initial research did point to T5 outperforming GPT on instructional data. Still Google closed up all their research since and we don't exactly know where that went. PaLM2 could be T5 architecture but those details were never reviled by Google.