fix: use 'cuda:0' by default if torch_device is 'cuda'

ktransformers/server/backend/interfaces/ktransformers.py

@@ -130,6 +130,7 @@ class KTransformersInterface(TransformersInterface):
         logger.debug(f"input_ids: {input_ids.shape}")
 
         device = self.device_map.get("blk.0.self_attn", {}).get("generate_device", "cuda:0")
+        device = "cuda:0" if device == "cuda" else device
 
         if is_new:
             self.ever_generated_ids.clear()