Doesn't work with CPUs
Hi y'all -- I've been doing a deep dive into this awesome capability and AFAICT it's fairly clear that the CPU path, particularly on Macs, has never worked. Please do correct me if I'm wrong on this, but there appears to be a handfu of bugs that result in cuda being chosen regardless of your selection which keep pulling the execution paths into the cuda route which doesn't work on Macs
For example model_load.py line 258: load_model_from_config() has wrong number of parameters resulting in default being cuda.
I have absolutely no issue with this -- and would like to help in some way. The question is:
- do we want to support CPUs at all? OR
- do we accept it's
cudaonly for now and remove the extra complexity in the code base that does these checks formap_locationanddevice(it's unclear to me what the difference is either they seem to refer to the same thing).
I'm fine with either: (1) is more work and (2) will simplify various parts of the code base.
Thoughts most welcome!
there was very little to no mac support when i was actively developing, things may have changed... happy to get your input :)
It’s a big task I think because of the dependencies ensuring they all support Apple Silicon. Macs, in this part of the tech world, are unfortunately very much third class citizens.
On Thu, 18 May 2023 at 15:25, deforum @.***> wrote:
there was very little to no mac support when i was actively developing, things may have changed... happy to get your input :)
— Reply to this email directly, view it on GitHub https://github.com/deforum-art/deforum-stable-diffusion/issues/189#issuecomment-1553146511, or unsubscribe https://github.com/notifications/unsubscribe-auth/AABD62J4WOBRGB522SFOI3DXGYWOXANCNFSM6AAAAAAXPPAE6A . You are receiving this because you authored the thread.Message ID: @.***>
i have started working on a new version that will support mac