biondizzle ffc1a5c6a8 Fix workspace_shapes: remove wrong assertion, compute output dim from K ...

The framework may pass K in different forms (packed or unpacked).
Use max(K*2, hidden_dim) to handle both cases.

2026-05-19 04:28:04 +00:00

..

kernels/linear/nvfp4

Replace autograd.Function with torch.library.custom_op for Dynamo compat

2026-05-19 01:54:48 +00:00

patches

Fix workspace_shapes: remove wrong assertion, compute output dim from K

2026-05-19 04:28:04 +00:00

cutedsl_quant_method.py

Replace autograd.Function with torch.library.custom_op for Dynamo compat

2026-05-19 01:54:48 +00:00

nvfp4_cutedsl.py

Replace autograd.Function with torch.library.custom_op for Dynamo compat

2026-05-19 01:54:48 +00:00