[Fix] Add SM check to flashinfer MOE backend (#29144)

Signed-off-by: jiahanc <173873397+jiahanc@users.noreply.github.com>
Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: mgoin <mgoin64@gmail.com>

vllm/model_executor/layers/quantization/utils/flashinfer_utils.py

@@ -282,6 +282,16 @@ def get_flashinfer_moe_backend() -> FlashinferMoeBackend:
 
     flashinfer_moe_backend = envs.VLLM_FLASHINFER_MOE_BACKEND
     if flashinfer_moe_backend in backend_map:
+        if (
+            flashinfer_moe_backend == "latency"
+            and not current_platform.is_device_capability(100)
+        ):
+            logger.info_once(
+                "Flashinfer TRTLLM MOE backend is only supported on "
+                "SM100 and later, using CUTLASS backend instead",
+                scope="local",
+            )
+            return FlashinferMoeBackend.CUTLASS
         return backend_map[flashinfer_moe_backend]
     elif current_platform.is_device_capability(90):
         return FlashinferMoeBackend.CUTLASS