vllm/v1/attention/backends/mla/flashinfer_mla.py · ef2c4f778df5aa07a44e663330e2dfdc16927d2a · OpenDAS / vllm_cscc

[Bugfix] Zero-init MLA attention output buffers to prevent NaN from CUDA graph padding (#37442) · ef2c4f77

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>

ef2c4f77

flashinfer_mla.py 8.48 KB

Replace flashinfer_mla.py