Spaces:
Running
Blank outputs for Flan UL2
The cached examples work fine, but anything else (e.g. What is 1 + 1?, Write a haiku about McDonalds) generates nothing for the Flan UL2 output part. Any ideas as to why this is happening? If I'm not mistaken I think it worked a few days ago just fine.
Thanks for the report
@ryanramos
It should be back now!
Sorry for the late reply; thanks! Can I ask what was the issue? I noticed that the code is using the Hugging Face inference API instead of running the models locally; and that when the Space worked my own API calls also worked, and when the Space was down my API calls didn't work either. What was the fix? And what makes Flan UL2 load nearly instantaneously through the inference API and not other 20B parameter models like the original UL2?
I think the API endpoints were down for this specific architecture but not sure cc @olivierdehaene