Remove sequential CPU offload from model loading and simplify return value in safe_model_load function

model_loader.py

@@ -39,7 +39,7 @@ def safe_model_load():
             print(f"XFormers not available: {e}")
             
         pipe.enable_attention_slicing()
-        pipe.enable_sequential_cpu_offload()
+        # pipe.enable_sequential_cpu_offload()
         pipe.to("cuda")
         
         # For memory-sensitive environments
@@ -47,13 +47,8 @@ def safe_model_load():
             torch.multiprocessing.set_sharing_strategy('file_system')
         except Exception as e:
             print(f"Exception raised (torch.multiprocessing): {e}")
-        # Moondream
-        model = vl(api_key=md_api_key)
         
-        return {
-            "pipeline": pipe,
-            "captioner": model
-        }
+        return pipe
         
     except Exception as e:
         print(f"Model loading failed: {e}")