Eugene Brodsky comments

Repositories
Issues
Comments

Results 132 comments of


                                            Eugene Brodsky

Redo custom attention processor to support other attention types

The PyTorch blog says Flash Attention is supported from sm80 compute capability onwards: https://pytorch.org/blog/accelerated-pytorch-2/, so perhaps we should default to `xformers` for anything lower than that. (As @hipsterusername said, with...

[bug]: invokeai docker v5.3.1 main-rocm still using cpu instead of rocm

We've recently updated ROCm images; closing this - please open another issue if still experiencing problems.