diff --git a/vllm/kernels/helion/utils.py b/vllm/kernels/helion/utils.py
index 65e327a82..600e459f6 100644
--- a/vllm/kernels/helion/utils.py
+++ b/vllm/kernels/helion/utils.py
@@ -2,14 +2,21 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 """Utility functions for Helion kernel management."""
 
-import torch
+import logging
+
+from vllm.platforms import current_platform
+
+logger = logging.getLogger(__name__)
 
 
 def get_gpu_name(device_id: int | None = None) -> str:
     if device_id is None:
-        device_id = torch.cuda.current_device()
-    props = torch.cuda.get_device_properties(device_id)
-    return props.name
+        logger.warning(
+            "get_gpu_name() called without device_id, defaulting to 0. "
+            "This may return the wrong device name in multi-node setups."
+        )
+        device_id = 0
+    return current_platform.get_device_name(device_id)
 
 
 def canonicalize_gpu_name(name: str) -> str:
@@ -18,6 +25,7 @@ def canonicalize_gpu_name(name: str) -> str:
 
     Converts to lowercase and replaces spaces and hyphens with underscores.
     e.g., "NVIDIA A100-SXM4-80GB" -> "nvidia_a100_sxm4_80gb"
+          "AMD_Instinct_MI300X"   -> "amd_instinct_mi300x"
 
     Raises ValueError if name is empty.
     """