Update app.py

app.py

@@ -275,6 +275,7 @@ def infer(genre_txt_content, lyrics_txt_content, num_segments, max_new_tokens):
         os.makedirs(output_dir, exist_ok=True)
         empty_output_folder(output_dir)
 
+        # 수정된 command - 지원되지 않는 인수 제거
         command = [
             "python", "infer.py",
             "--stage1_model", model_path,
@@ -286,9 +287,7 @@ def infer(genre_txt_content, lyrics_txt_content, num_segments, max_new_tokens):
             "--output_dir", output_dir,
             "--cuda_idx", "0",
             "--max_new_tokens", str(actual_max_tokens),
-            "--use_flash_attention", "True",
-            "--use_bettertransformer", "True",
-            "--use_compile", "True"
+            "--disable_offload_model"  # GPU 메모리 최적화를 위해 추가
         ]
 
         env = os.environ.copy()
@@ -299,10 +298,16 @@ def infer(genre_txt_content, lyrics_txt_content, num_segments, max_new_tokens):
                 "PATH": f"/usr/local/cuda/bin:{env.get('PATH', '')}",
                 "LD_LIBRARY_PATH": f"/usr/local/cuda/lib64:{env.get('LD_LIBRARY_PATH', '')}",
                 "PYTORCH_CUDA_ALLOC_CONF": "max_split_size_mb:512",
-                "CUDA_LAUNCH_BLOCKING": "0",
-                "TORCH_DISTRIBUTED_DEBUG": "DETAIL"
+                "CUDA_LAUNCH_BLOCKING": "0"
             })
 
+        # transformers 캐시 마이그레이션 처리
+        try:
+            from transformers.utils import move_cache
+            move_cache()
+        except Exception as e:
+            logging.warning(f"Cache migration warning (non-critical): {e}")
+
         process = subprocess.run(
             command,
             env=env,