[Bugfix][Hardware][AMD] Fix device parameter and exception handling

vllm/compilation/fusion.py

@@ -38,20 +38,20 @@
 FP4_DTYPE = torch.uint8
 
 
-def empty_bf16(*args, **kwargs):
-    return torch.empty(*args, **kwargs, dtype=torch.bfloat16, device="cuda")
+def empty_bf16(*args, device=current_platform.device_type, **kwargs):
+    return torch.empty(*args, **kwargs, dtype=torch.bfloat16, device=device)
 
 
-def empty_fp32(*args, **kwargs):
-    return torch.empty(*args, **kwargs, dtype=torch.float32, device="cuda")
+def empty_fp32(*args, device=current_platform.device_type, **kwargs):
+    return torch.empty(*args, **kwargs, dtype=torch.float32, device=device)
 
 
-def empty_i32(*args, **kwargs):
-    return torch.empty(*args, **kwargs, dtype=torch.int32, device="cuda")
+def empty_i32(*args, device=current_platform.device_type, **kwargs):
+    return torch.empty(*args, **kwargs, dtype=torch.int32, device=device)
 
 
-def empty_i64(*args, **kwargs):
-    return torch.empty(*args, **kwargs, dtype=torch.int64, device="cuda")
+def empty_i64(*args, device=current_platform.device_type, **kwargs):
+    return torch.empty(*args, **kwargs, dtype=torch.int64, device=device)
 
 
 RMS_OP = torch.ops._C.rms_norm.default

vllm/model_executor/layers/quantization/kernels/scaled_mm/aiter.py

@@ -32,7 +32,7 @@ def is_supported(
 
         try:
             import aiter  # noqa: F401 # deliberately attempt to import aiter
-        except Exception:
+        except (ImportError, ModuleNotFoundError):
             return (
                 False,
                 "AiterScaledMMLinearKernel requires `aiter` which is not "