[misc] Add FlashMLA as a new option of VLLM_ATTENTION_BACKEND env (#14267)

2025-03-06 05:28:50 +08:00
parent ac60dc7fe1
commit 1b7624bf5c
1 changed files with 1 additions and 0 deletions
--- a/vllm/envs.py
+++ b/vllm/envs.py
@@ -321,6 +321,7 @@ environment_variables: dict[str, Callable[[], Any]] = {
    # - "XFORMERS": use XFormers
    # - "ROCM_FLASH": use ROCmFlashAttention
    # - "FLASHINFER": use flashinfer
+    # - "FLASHMLA": use FlashMLA
    "VLLM_ATTENTION_BACKEND":
    lambda: os.getenv("VLLM_ATTENTION_BACKEND", None),