update platforms init

vllm/attention/selector.py

@@ -203,8 +203,7 @@ def which_attn_to_use(
         selected_backend = (_Backend.ROCM_FLASH if selected_backend
                             == _Backend.FLASH_ATTN else selected_backend)
         if selected_backend == _Backend.ROCM_FLASH:
-            # if current_platform.get_device_capability()[0] != 9:
-            if torch.cuda.get_device_capability()[0] != 9:
+            if current_platform.get_device_capability()[0] != 9:
                 # not Instinct series GPUs.
                 logger.info("flash_attn is not supported on NAVI GPUs.")
         else:

vllm/platforms/__init__.py

+import torch
 from .interface import Platform, PlatformEnum, UnspecifiedPlatform
 
 current_platform: Platform
@@ -32,13 +33,15 @@ except Exception:
 is_rocm = False
 
 try:
-    import amdsmi
-    amdsmi.amdsmi_init()
-    try:
-        if len(amdsmi.amdsmi_get_processor_handles()) > 0:
+    if torch.version.hip is not None:
         is_rocm = True
-    finally:
-        amdsmi.amdsmi_shut_down()
+    # import amdsmi
+    # amdsmi.amdsmi_init()
+    # try:
+    #     if len(amdsmi.amdsmi_get_processor_handles()) > 0:
+    #         is_rocm = True
+    # finally:
+    #     amdsmi.amdsmi_shut_down()
 except Exception:
     pass