Remove vulkan from available backends when --backend nvidia is used with GGUF models

When user explicitly passes --backend nvidia with a GGUF model,
vulkan is now removed from the available backends list since
llama-cpp-python will use CUDA instead of Vulkan.

coderai

@@ -4809,14 +4809,15 @@ def main():
     # Detect available backends
     available = detect_available_backends()
     
-    # Check if any model is GGUF and backend is nvidia/cuda - in that case, vulkan uses CUDA
+    # If user explicitly requests nvidia/cuda backend with a GGUF model, 
+    # remove vulkan from available since we'll use CUDA instead
     if model_names:
         first_model = model_names[0]
         is_gguf_model = first_model.endswith('.gguf') or 'gguf' in first_model.lower()
         if is_gguf_model and args.backend in ('nvidia', 'cuda'):
-            # When using nvidia/cuda backend with GGUF, vulkan actually uses CUDA
-            if 'vulkan' in available and available['vulkan']:
-                print("\nNote: GGUF model with nvidia backend - vulkan backend will use CUDA")
+            # When using nvidia/cuda backend with GGUF, vulkan uses CUDA, so remove it
+            if 'vulkan' in available:
+                del available['vulkan']
     
     print("\nAvailable backends:")
     for name, available_flag in available.items():