Ariel

icon indicating a website link https://replicate.com/

icon company Replicate icon location Searching for interesting and reproducible machine learning projects that can be added to Replicate API and cloud demos.

Results 7 comments of Ariel

CUDA out of memory? (3080ti 12GB)

I'm running x4 scaling on a V100 (40GB) without _**xformers**_ and can run on 256x256 inputs but fail on 512x512 inputs with this error: ``` RuntimeError: CUDA out of memory....

docker and ideally cog support

Hi, See my pull request #55 There is also a link to a working demo on Replicate

Slow model loading

Yes, I'm not talking about the downloading. Only the loading into memory takes ~5 minutes each. This is huge right? Any way to speed this up?

Controling the number of inversion steps

Thanks for the elaboration! It seem though that **ddim_steps** is not related to inversion only to sampling after the inversion https://github.com/MichalGeyer/plug-and-play/blob/ce9e4e23c24b4241cb9ecae9ea0b090c70871870/run_features_extraction.py#L229 Is it using the fact that model.num_timesteps is set...

Add Replicate demo and API

Hi @DengPingFan, Sorry for the very late reply. It seems to work now on Replicate.com. Can you check again and tell me if you'r still getting this error?

Add Replicate demo and API

Yes I see the crash after few trials. Thanks @radi-cho I think using **no_grad** will help but probably wont solve the memory leak right? @DengPingFan can you help on this?...

Fix typo

Please accept this pull request