biondizzle 4c2effa2be Fix attention patch: source from v0.21.0 stable, not local clone ...

The local vllm clone has different imports (breakable_cudagraph) that
don't exist in the Docker image. Now sourced from v0.21.0 tag.

2026-05-19 06:44:59 +00:00

..

kernels/linear/nvfp4

Fix activation global scale: don't double-invert input_global_scale_inv

2026-05-19 06:03:08 +00:00

patches

Fix attention patch: source from v0.21.0 stable, not local clone

2026-05-19 06:44:59 +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