Update app.py

app.py

@@ -375,37 +375,31 @@ def on_change_prompt(img: Image.Image | None, prompt: str | None, bg_prompt: str
     return gr.update(interactive=bool(img and prompt))
 
 
-@spaces.GPU(duration=20)
+@spaces.GPU()
 def process_prompt(img: Image.Image, prompt: str, bg_prompt: str | None = None, 
                   aspect_ratio: str = "1:1", position: str = "bottom-center", 
-                  scale_percent: float = 100, text_params: dict | None = None) -> tuple[Image.Image, Image.Image]:
+                  scale_percent: float = 100, text_params: dict | None = None):
     try:
-        # 입력 이미지 최적화
-        img = process_image(img)
+        # GPU 설정
+        if torch.cuda.is_available():
+            device = torch.device('cuda')
+            # 모델들을 GPU로 이동
+            gd_model.to(device)
+            segmenter.to(device)
+            pipe.to(device)
+        else:
+            device = torch.device('cpu')
+        
+        # 나머지 처리 로직...
         
-        with torch.cuda.amp.autocast():  # 자동 혼합 정밀도 사용
-            results, _ = _process(img, prompt, bg_prompt, aspect_ratio)
-            
-            if bg_prompt:
-                combined = combine_with_background(
-                    foreground=results[2],
-                    background=results[1],
-                    position=position,
-                    scale_percent=scale_percent
-                )
-                
-                if text_params and text_params.get('text'):
-                    combined = add_text_to_image(combined, text_params)
-                    
-                return combined, results[2]
-            
-            return results[1], results[2]
-            
-    except Exception as e:
-        print(f"Error in process_prompt: {str(e)}")
-        raise gr.Error(str(e))
     finally:
-        clear_memory()
+        # GPU 메모리 정리
+        if torch.cuda.is_available():
+            try:
+                with torch.cuda.device('cuda'):
+                    torch.cuda.empty_cache()
+            except Exception as e:
+                print(f"GPU cleanup warning: {e}")
 
 
 def process_bbox(img: Image.Image, box_input: str) -> tuple[Image.Image, Image.Image]:
@@ -810,6 +804,7 @@ if __name__ == "__main__":
             queue=True
         )
 
+# Gradio 실행 설정 수정
 demo.launch(
     server_name="0.0.0.0",
     server_port=7860,
@@ -818,7 +813,5 @@ demo.launch(
     enable_queue=True,
     cache_examples=False,
     show_error=True,
-    show_tips=False,
-    max_size=1,  # 큐 크기 제한
-    memory_limit="48Gi"  # 메모리 제한 설정
+    show_tips=False
 )