Update app.py

app.py

@@ -22,6 +22,14 @@ from diffusers import FluxPipeline
 # 상단에 import 추가
 from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM
 
+import gc
+
+def clear_memory():
+    """메모리 정리 함수"""
+    gc.collect()
+    torch.cuda.empty_cache()
+    
+
 
 model_name = "Helsinki-NLP/opus-mt-ko-en"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
@@ -323,10 +331,12 @@ def on_change_prompt(img: Image.Image | None, prompt: str | None, bg_prompt: str
     return gr.update(interactive=bool(img and prompt))
 
 
+        
 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) -> tuple[Image.Image, Image.Image]:
     try:
+        clear_memory()  # 처리 전 메모리 정리        
         if img is None or prompt.strip() == "":
             raise gr.Error("Please provide both image and prompt")
         
@@ -362,6 +372,9 @@ def process_prompt(img: Image.Image, prompt: str, bg_prompt: str | None = None,
         print(f"Error in process_prompt: {str(e)}")
         raise gr.Error(str(e))
 
+    finally:
+        clear_memory()  # 처리 후 메모리 정리
+        
 def process_bbox(img: Image.Image, box_input: str) -> tuple[Image.Image, Image.Image]:
     try:
         if img is None or box_input.strip() == "":
@@ -667,5 +680,12 @@ with gr.Blocks(theme=gr.themes.Soft(), css=css) as demo:
             </div>
         </div>
     """)
-demo.queue(max_size=30, api_open=False)
-demo.launch()
+demo.queue(max_size=20)  # 큐 크기 제한
+demo.launch(
+    server_name="0.0.0.0",
+    server_port=7860,
+    share=False,
+    enable_queue=True,
+    max_threads=4,  # 스레드 수 제한
+    allowed_paths=["examples"]
+)