Shouldn't Cuda have priority over DML ? #15

Closed
opened 2023-02-10 12:08:02 +00:00 by accountts · 1 comment

In utils/device.py

if has_dml():
name = 'dml'
elif torch.cuda.is_available():
name = 'cuda'

In utils/device.py if has_dml(): name = 'dml' elif torch.cuda.is_available(): name = 'cuda'
Owner

No.

  • don't install both torch variants.
  • don't use DirectML if you can use CUDA. I stress this enough in my documentation.
  • to the end user that can follow directions and install one (1) variant for the right device, it does not matter
  • it only matters to basket cases like me who, in testing, was using both variants with mixed GPUs and explicitly wanted to switch devices by simply overriding has_dml, rather than dealing with the nightmare of dealing with env vars to nix exposing CUDA capability.
No. * don't install both torch variants. * don't use DirectML if you can use CUDA. I stress this enough in my documentation. * to the end user that can follow directions and install one (1) variant for the right device, it does not matter * it only matters to basket cases like me who, in testing, was using both variants with mixed GPUs and explicitly wanted to switch devices by simply overriding `has_dml`, rather than dealing with the nightmare of dealing with env vars to nix exposing CUDA capability.
mrq closed this issue 2023-02-10 13:48:08 +00:00
Sign in to join this conversation.
No Label
No Milestone
No project
No Assignees
2 Participants
Notifications
Due Date
The due date is invalid or out of range. Please use the format 'yyyy-mm-dd'.

No due date set.

Dependencies

No dependencies set.

Reference: mrq/tortoise-tts#15
No description provided.