Aditya Borate
Aditya Borate
Hi! I'm planning to work on this for Hacktoberfest 2024 and hope to have it fixed.
I've updated the algorithm and added more test cases as per the feedback received. Please review the latest changes. Thank you!
Hi! This is my Hugging Face username: `iota-10`
Hey @Passenger12138 @DN6, I spent some time looking into this and verifying the reproduction steps. It looks like there are two fundamental conflicts here that make this specific combination of...
Thanks for your insights @sayakpaul. However, the documentation code should use `channels_last_3d` instead of `channels_last` to support the video architecture, shouldn't it?
I was able to reproduce the above UserWarning on a non-CUDA setup. In addition to `Kandinsky5TimeEmbeddings`, I noticed the `Kandinsky5Modulation` class also uses the same decorator pattern with a hardcoded...
> Hi! I See the problem. If you have fixes you want to propose please create a pull request. However, does Flex Attention work fine on non Cuda systems? I...
Thanks for pointing that out @vladmandic. To fix this, we should upcast the input to fp32, compute the embeddings, and downcast the result of the math back to `weight.dtype` before...
@yiyixuxu @sayakpaul A gentle ping to review
> Thank you! Could you also provide your testing script? The verification script is already provided in the PR description above. If you want to test minimally, we can just...