From 9c81f35b1ae6c70681661e11c461ddbb7e417aff Mon Sep 17 00:00:00 2001 From: Matthew Bonanni Date: Mon, 6 Apr 2026 17:51:46 -0400 Subject: [PATCH] [Attention][MLA] Re-enable FA4 as default MLA prefill backend (#38819) --- vllm/config/attention.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/vllm/config/attention.py b/vllm/config/attention.py index 014bb9b22..1da647a6d 100644 --- a/vllm/config/attention.py +++ b/vllm/config/attention.py @@ -30,7 +30,7 @@ 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