biondizzle e404e18efb Also replace layers/mhc.py CustomOp dispatch ...

The original layers/mhc.py forward_cuda calls
torch.ops.vllm.mhc_pre_tilelang which triggers TileLang JIT.
Replace with our torch implementations in forward_cuda.
This is what the CustomOp dispatch routes through.

2026-05-19 05:31:05 +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

Also replace layers/mhc.py CustomOp dispatch

2026-05-19 05:31:05 +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