.

app.py

@@ -64,11 +64,46 @@ class LocalModelManager:
         self.models = {}
         self.tokenizers = {}
         self.pipelines = {}
+        
+    def preload_models(self, model_paths, tasks=None):
+        """アプリケーション起動時にモデルを事前ロード"""
+        if tasks is None:
+            tasks = {}  # デフォルトは空の辞書
+            
+        logger.info("Preloading models at application startup...")
+        for model_path in model_paths:
+            task = tasks.get(model_path, "text-generation")
+            try:
+                logger.info(f"Preloading model: {model_path} for task: {task}")
+                self.tokenizers[model_path] = AutoTokenizer.from_pretrained(model_path)
+                
+                if task == "text-generation":
+                    self.pipelines[model_path] = pipeline(
+                        "text-generation",
+                        model=model_path,
+                        tokenizer=self.tokenizers[model_path],
+                        torch_dtype=torch.bfloat16,
+                        trust_remote_code=True,
+                        device_map="auto"
+                    )
+                else:  # classification
+                    self.pipelines[model_path] = pipeline(
+                        "text-classification",
+                        model=model_path,
+                        tokenizer=self.tokenizers[model_path],
+                        torch_dtype=torch.bfloat16,
+                        trust_remote_code=True,
+                        device_map="auto"
+                    )
+                logger.info(f"Model preloaded successfully: {model_path}")
+            except Exception as e:
+                logger.error(f"Error preloading model {model_path}: {str(e)}")
+                # 続行するが、エラーをログに記録
 
     async def load_model(self, model_path: str, task: str = "text-generation"):
-        """モデルの遅延ロード"""
-        if model_path not in self.models:
-            logger.info(f"Loading model: {model_path}")
+        """モデルが既にロードされているか確認し、なければロード"""
+        if model_path not in self.pipelines:
+            logger.info(f"Loading model on demand: {model_path}")
             try:
                 self.tokenizers[model_path] = AutoTokenizer.from_pretrained(model_path)
                 
@@ -140,6 +175,7 @@ class ModelManager:
         self.api_clients = {}
         self.local_manager = LocalModelManager()
         self._initialize_clients()
+        self._preload_local_models()
 
     def _initialize_clients(self):
         """Inference APIクライアントの初期化"""
@@ -150,6 +186,26 @@ class ModelManager:
                     token=True  # これによりHFトークンを使用
                 )
 
+    def _preload_local_models(self):
+        """ローカルモデルの事前ロード"""
+        models_to_preload = []
+        tasks = {}
+        
+        # テキスト生成モデルの追加
+        for model in TEXT_GENERATION_MODELS:
+            if model.type == ModelType.LOCAL and model.model_path:
+                models_to_preload.append(model.model_path)
+                tasks[model.model_path] = "text-generation"
+        
+        # 分類モデルの追加
+        for model in CLASSIFICATION_MODELS:
+            if model.type == ModelType.LOCAL and model.model_path:
+                models_to_preload.append(model.model_path)
+                tasks[model.model_path] = "text-classification"
+        
+        # 事前ロード実行
+        self.local_manager.preload_models(models_to_preload, tasks)
+
     async def run_text_generation(self, text: str, selected_types: List[str]) -> List[str]:
         """テキスト生成モデルの実行"""
         results = []