Update handler.py

handler.py

@@ -29,7 +29,9 @@ class EndpointHandler:
             token=hf_token
         ).to("cuda")
 
-        self.pipe.enable_xformers_memory_efficient_attention()
+        # IMPORTANT FIX — remove xformers, use PyTorch attention instead
+        if hasattr(self.pipe, "enable_model_cpu_offload"):
+            self.pipe.enable_model_cpu_offload()
 
         print("🚀 Pipeline loaded successfully!")
 
@@ -39,7 +41,6 @@ class EndpointHandler:
         prompt = data.get("prompt", "")
         negative_prompt = data.get("negative_prompt", "blurry, bad quality, distorted, extra limbs")
 
-        # Always generate 10 images
         num_images = 10
 
         # Decode uploaded image