Add conditional replacement of `@torch.inference_mode` for inference on AMD DirectML GPUs#3295

Merged

deruyter92 merged 9 commits into

mainfrom

jaap/amd_direct_ml_inference

May 12, 2026

Commits on Apr 28, 2026

Add conditional no_grad decorator for AMD DirectML GPUs.
deruyter92
committed
Update comment
deruyter92
committed
update naming: `_no_grad_decorator` -> `_inference_mode_decorator`
deruyter92
committed
Add pytest for directml conditional no_grad inference mode
deruyter92
committed
update pyproject.toml and uv.lock: add importlib to `dev` dependencies
deruyter92
committed
Update deeplabcut/pose_estimation_pytorch/runners/inference.py

deruyter92
and
C-Achard
authored
restore pyproject.toml and uv.lock (no changes needed)
deruyter92
committed
add error-hint (using contextmanager) for directml-related runtime error
deruyter92
committed
add note in TechHardware.md for DirectML inference troubleshooting
deruyter92
committed