[Model Runner V2][DFlash] Enable dflash attention backend selection (#46770)
Signed-off-by: Giancarlo Delfin <gdelfin@inferact.ai>
Giancarlo Delfin committed
Jun 25, 2026 at 21:29 UTC
5314665badcb93f798e117aacad8ce02f148cd73
1 file changed
+3
-1
vllm/v1/worker/gpu/spec_decode/dflash/utils.py
+3
-1
@@ -26,7 +26,9 @@ def load_dflash_model(target_model: nn.Module, vllm_config: VllmConfig) -> nn.Mo
26
draft_vllm_config = replace(
27
vllm_config,
28
attention_config=replace(
29
- vllm_config.attention_config, use_non_causal=not causal
29
+ vllm_config.attention_config,
30
+ use_non_causal=not causal,
31
+ backend=speculative_config.attention_backend,
32
),
33
)
34
with set_model_tag("dflash_head"):