From 5314665badcb93f798e117aacad8ce02f148cd73 Mon Sep 17 00:00:00 2001 From: Giancarlo Delfin <32987265+TheEpicDolphin@users.noreply.github.com> Date: Thu, 25 Jun 2026 21:29:25 -0500 Subject: [PATCH] [Model Runner V2][DFlash] Enable dflash attention backend selection (#46770) Signed-off-by: Giancarlo Delfin --- vllm/v1/worker/gpu/spec_decode/dflash/utils.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/vllm/v1/worker/gpu/spec_decode/dflash/utils.py b/vllm/v1/worker/gpu/spec_decode/dflash/utils.py index 01f6923a76a..c4f98e715b9 100644 --- a/vllm/v1/worker/gpu/spec_decode/dflash/utils.py +++ b/vllm/v1/worker/gpu/spec_decode/dflash/utils.py @@ -26,7 +26,9 @@ def load_dflash_model(target_model: nn.Module, vllm_config: VllmConfig) -> nn.Mo draft_vllm_config = replace( vllm_config, attention_config=replace( - vllm_config.attention_config, use_non_causal=not causal + vllm_config.attention_config, + use_non_causal=not causal, + backend=speculative_config.attention_backend, ), ) with set_model_tag("dflash_head"):