[UX] Rename CUTLASS_MLA_VLLM_V1 to CUTLASS_MLA (#21966)

Signed-off-by: mgoin <mgoin64@gmail.com>

vllm/engine/arg_utils.py

@@ -1417,7 +1417,7 @@ class EngineArgs:
             "PALLAS_VLLM_V1",
             "TRITON_ATTN_VLLM_V1",
             "TRITON_MLA",
-            "CUTLASS_MLA_VLLM_V1",
+            "CUTLASS_MLA",
             "FLASHMLA",
             "FLASHINFER",
             "FLASHINFER_VLLM_V1",

vllm/platforms/cuda.py

@@ -162,7 +162,7 @@ class CudaPlatformBase(Platform):
                 if cls.is_device_capability(100):
                     # Blackwell => Force CutlassMLA.
                     use_cutlass_mla = True
-                    envs.VLLM_ATTENTION_BACKEND = "CUTLASS_MLA_VLLM_V1"
+                    envs.VLLM_ATTENTION_BACKEND = "CUTLASS_MLA"
                 else:
                     # Not Blackwell
                     use_flashmla = True
@@ -170,7 +170,7 @@ class CudaPlatformBase(Platform):
                 # Forced case
                 use_flashmla = (envs.VLLM_ATTENTION_BACKEND == "FLASHMLA")
                 use_cutlass_mla = (
-                    envs.VLLM_ATTENTION_BACKEND == "CUTLASS_MLA_VLLM_V1")
+                    envs.VLLM_ATTENTION_BACKEND == "CUTLASS_MLA")
 
             from vllm.attention.ops.flashmla import is_flashmla_supported
             if use_flashmla and is_flashmla_supported()[0] \
@@ -182,7 +182,7 @@ class CudaPlatformBase(Platform):
             if use_cutlass_mla and cache_config.block_size != 128:
                 cache_config.block_size = 128
                 logger.info("Forcing kv cache block size to 128 for "
-                            "CUTLASS_MLA_VLLM_V1 backend.")
+                            "CUTLASS_MLA backend.")
 
         compilation_config = vllm_config.compilation_config
         if (envs.VLLM_ALL2ALL_BACKEND == "deepep_high_throughput"
@@ -213,7 +213,7 @@ class CudaPlatformBase(Platform):
         if use_mla:
             # TODO(lucas): refactor to be more concise
             #  we should probably consider factoring out V1 here
-            if selected_backend == _Backend.CUTLASS_MLA_VLLM_V1:
+            if selected_backend == _Backend.CUTLASS_MLA:
                 if use_v1:
                     logger.info_once("Using Cutlass MLA backend on V1 engine.")
                     return ("vllm.v1.attention.backends.mla."

vllm/platforms/interface.py

@@ -53,7 +53,7 @@ class _Backend(enum.Enum):
     TRITON_MLA_VLLM_V1 = enum.auto()
     FLASHMLA_VLLM_V1 = enum.auto()
     FLASHMLA = enum.auto()  # Supported by V1
-    CUTLASS_MLA_VLLM_V1 = enum.auto()
+    CUTLASS_MLA = enum.auto()
     PALLAS = enum.auto()
     PALLAS_VLLM_V1 = enum.auto()
     IPEX = enum.auto()

vllm/v1/attention/backends/mla/cutlass_mla.py

@@ -21,7 +21,7 @@ class CutlassMLABackend(MLACommonBackend):
 
     @staticmethod
     def get_name() -> str:
-        return "CUTLASS_MLA_VLLM_V1"
+        return "CUTLASS_MLA"
 
     @staticmethod
     def get_impl_cls() -> type["CutlassMLAImpl"]: