biondizzle e0f385ac45 Fix workspace_shapes: output dim is hidden_dim, not K*2 ...

K comes from hidden_states.size(-1) which is the full BF16 dimension
(7168), not the packed weight dimension. K*2=14336 is wrong.
The MoE output is always hidden_dim (7168).

2026-05-19 04:42:22 +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: output dim is hidden_dim, not K*2

2026-05-19 04:42:22 +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