biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Attention] Use FA4 for MLA prefill (#34732)

Browse Source 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
This commit is contained in:
Matthew Bonanni
2026-03-12 12:10:17 -04:00
committed by GitHub
parent 85199f9681
commit f444c05c32
9 changed files with 413 additions and 78 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/config/attention.py
View File

@@ -30,14 +30,14 @@ class AttentionConfig:
use_cudnn_prefill: bool = False
"""Whether to use cudnn prefill."""
use_trtllm_ragged_deepseek_prefill: bool = True
use_trtllm_ragged_deepseek_prefill: bool = False
"""Whether to use TRTLLM ragged deepseek prefill."""
use_trtllm_attention: bool | None = None
"""If set to True/False, use or don't use the TRTLLM attention backend
in flashinfer. If None, auto-detect the attention backend in flashinfer."""
disable_flashinfer_prefill: bool = False
disable_flashinfer_prefill: bool = True
"""Whether to disable flashinfer prefill."""
disable_flashinfer_q_quantization: bool = False