Add FlexAttention to V1 (#16078)

Signed-off-by: drisspg <drisspguessous@gmail.com>
2025-06-07 00:58:55 -04:00
parent c4296b1a27
commit cf02f9b283
5 changed files with 575 additions and 0 deletions
--- a/vllm/engine/arg_utils.py
+++ b/vllm/engine/arg_utils.py
@@ -1409,6 +1409,7 @@ class EngineArgs:
            "FLASHINFER_VLLM_V1",
            "ROCM_AITER_MLA",
            "TORCH_SDPA_VLLM_V1",
+            "FLEX_ATTENTION",
        ]
        if (envs.is_set("VLLM_ATTENTION_BACKEND")
                and envs.VLLM_ATTENTION_BACKEND not in V1_BACKENDS):