Hi, I was playing around with this repo while helping to integrate it with HF Diffusers https://github.com/huggingface/diffusers/pull/3492, and found that if we used normal attention QKVAttentionLegacy we were not able to generate images of good quality, although using flash attention QKVFlashAttention gave the expected results.
The reason for this turned out to be a slightly different ordering of the heads and keys in the flash attention and normal attention implementations. Fixing which allowed us to generate good images using normal attention, without flash attention ( which can be a pain to install ).
Sending this PR with the fix, would be good to integrate this since flash attention is hard to install and can't be installed in all devices.
The quantization might not be that important, its just that Flash attention works with fp16 only I think, so I did an explicit conversion in the normal attention to get exactly same results
TLDR : You can now generate good images from the pretrained checkpoints without using Flash attention
Hi, I was playing around with this repo while helping to integrate it with HF Diffusers https://github.com/huggingface/diffusers/pull/3492, and found that if we used normal attention
QKVAttentionLegacywe were not able to generate images of good quality, although using flash attentionQKVFlashAttentiongave the expected results.The reason for this turned out to be a slightly different ordering of the heads and keys in the flash attention and normal attention implementations. Fixing which allowed us to generate good images using normal attention, without flash attention ( which can be a pain to install ).
Sending this PR with the fix, would be good to integrate this since flash attention is hard to install and can't be installed in all devices.
The quantization might not be that important, its just that Flash attention works with fp16 only I think, so I did an explicit conversion in the normal attention to get exactly same results
TLDR : You can now generate good images from the pretrained checkpoints without using Flash attention