Spaces:

hf-accelerate

/

model-memory-usage

like 805

Running on CPU Upgrade

Thanks! Probably due to the outtage. Looks to be working now for me, can you confirm?

Yes, seems to be working for me now as well. Thanks!

Hmmm wait - my first attempt worked, but now seeing issues when trying on a second model

not working for me

Can you guys provide me with reproducers please? What models are we using :)

Looks like it's hanging, investigating.

Okay restarted again... Let's try one more time? Might be a weird spaces infra issue.

Ok seems to be working now! Tested several times with several models and looks good. Thanks!

Seems like it doesn't work again..

Thanks for the report @kiucho . I'm rebuilding one final time to see if this pr fixed it https://github.com/huggingface/accelerate/pull/2435, and I'll keep an eye on this thread for the next few days. If we hit a queue again/freeze I'll report this to the team

It seems like it is hitting the queue again :/

For me it just perma-loads with a queue message 😞

Thanks all, asking infra to help out with this as it's a critical issue!

The issue was due to a bug of huggingface_hub==0.20.3. https://github.com/huggingface/huggingface_hub/pull/1962
As huggingface_hub==0.19.4 doesn't have the bug, I pinned the version for now. https://huggingface.co/spaces/hf-accelerate/model-memory-usage/discussions/22

Thank you, it's back in business! This is a super useful utility!

Closing as we're working again thanks to @hysts sharp deduction skills :)