vllm/v1/attention/backends/mla/cutlass_mla.py · 89138b21cc246ae944c741d5c399c148e2b770ab · OpenDAS / vllm_cscc

[Bugfix] Zero-init MLA attention output buffers to prevent NaN from CUDA graph padding (#37442) · 89138b21

Elvir Crnčević authored Mar 19, 2026


Signed-off-by: Elvir Crncevic <elvircrn@gmail.com>
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Co-authored-by: Matthew Bonanni <mbonanni@redhat.com>
(cherry picked from commit ef2c4f77)

89138b21

cutlass_mla.py 9.18 KB

Replace cutlass_mla.py