Add AMD GPU compatibility fix for tensor device mismatches

__init__.py

@@ -1,3 +1,6 @@
 from .nodes import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS
 
-__all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS"]
+__all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS"]
+
+# Add AMD GPU compatibility support
+from . import amd_fix

amd_fix.py

@@ -0,0 +1,49 @@
+# amd_fix.py - AMD GPU compatibility fix for tensor device mismatches
+import torch
+
+# Store the original concat function
+original_concat = torch.concat
+
+# Create a safe replacement that ensures tensors are on the same device
+def safe_concat(tensors, *args, **kwargs):
+    if not isinstance(tensors, (list, tuple)) or len(tensors) <= 1:
+        return original_concat(tensors, *args, **kwargs)
+
+    # Find the first CUDA device in the list
+    target_device = None
+    for t in tensors:
+        if isinstance(t, torch.Tensor) and t.device.type == 'cuda':
+            target_device = t.device
+            break
+
+    # If no CUDA device found, use the first tensor's device
+    if target_device is None and isinstance(tensors[0], torch.Tensor):
+        target_device = tensors[0].device
+
+    # Only move tensors if needed
+    if target_device is not None:
+        fixed_tensors = []
+        for t in tensors:
+            if isinstance(t, torch.Tensor) and t.device != target_device:
+                fixed_tensors.append(t.to(target_device))
+            else:
+                fixed_tensors.append(t)
+        return original_concat(fixed_tensors, *args, **kwargs)
+    else:
+        return original_concat(tensors, *args, **kwargs)
+
+# Apply the patch
+def apply_fix():
+    try:
+        # Only apply this when running on AMD GPU
+        if torch.cuda.is_available() and "amd" in torch.cuda.get_device_name(0).lower():
+            # Apply our patch
+            torch.concat = safe_concat
+            print("Applied device-safe concatenation patch for AMD GPUs")
+        return True
+    except Exception as e:
+        print(f"Failed to apply AMD fix: {str(e)}")
+        return False
+
+# Apply fix when imported
+apply_fix()