A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 5, 19:45 UTC
[Bugfix][MLA] Change default SM100 MLA prefill backend back to TRT-LLM (#38562)
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Matthew Bonanni committed
Mar 30, 2026 at 12:51 UTC
2c734ed0e06a48808522fe8f59f6b4ffe0cf0397
1 file changed
+1
-1
vllm/config/attention.py
+1
-1
@@ -30,7 +30,7 @@ class AttentionConfig:
30
use_cudnn_prefill: bool = False
31
"""Whether to use cudnn prefill."""
32
33
- use_trtllm_ragged_deepseek_prefill: bool = False
33
+ use_trtllm_ragged_deepseek_prefill: bool = True
34
"""Whether to use TRTLLM ragged deepseek prefill."""
35
36
use_trtllm_attention: bool | None = None