[Feat] Add NPU Backend support for vLLM-Omni (#89)

Signed-off-by: gcanlin <canlinguosdu@gmail.com>

vllm_omni/diffusion/distributed/utils.py

@@ -5,7 +5,11 @@ import os
 
 import torch
 
+from vllm_omni.utils.platform_utils import detect_device_type
+
 
 def get_local_device() -> torch.device:
-    """Return the torch device for the current rank."""
-    return torch.device(f"cuda:{os.environ.get('LOCAL_RANK', 0)}")
+    """Return the torch device for the current rank based on detected device type."""
+    device_type = detect_device_type()
+    local_rank = os.environ.get("LOCAL_RANK", 0)
+    return torch.device(f"{device_type}:{local_rank}")