.

app.py

@@ -117,6 +117,10 @@ def generate_text_local(model_path, text):
         logger.info(f"Running local text generation with {model_path}")
         pipeline = pipelines[model_path]
         
+        # モデルを明示的に GPU に移動
+        if hasattr(pipeline.model, "to"):
+            pipeline.model.to("cuda")
+        
         # デバイス情報をログに記録
         device_info = next(pipeline.model.parameters()).device
         logger.info(f"Model {model_path} is running on device: {device_info}")
@@ -127,6 +131,11 @@ def generate_text_local(model_path, text):
             do_sample=False,
             num_return_sequences=1
         )
+
+        # モデルを CPU に戻す
+        if hasattr(pipeline.model, "to"):
+            pipeline.model.to("cpu")
+
         return outputs[0]["generated_text"]
     except Exception as e:
         logger.error(f"Error in local text generation with {model_path}: {str(e)}")
@@ -153,11 +162,20 @@ def classify_text_local(model_path, text):
         logger.info(f"Running local classification with {model_path}")
         pipeline = pipelines[model_path]
         
+        # モデルを明示的に GPU に移動
+        if hasattr(pipeline.model, "to"):
+            pipeline.model.to("cuda")
+
         # デバイス情報をログに記録
         device_info = next(pipeline.model.parameters()).device
         logger.info(f"Model {model_path} is running on device: {device_info}")
         
         result = pipeline(text)
+
+        # モデルを CPU に戻す
+        if hasattr(pipeline.model, "to"):
+            pipeline.model.to("cpu")
+
         return str(result)
     except Exception as e:
         logger.error(f"Error in local classification with {model_path}: {str(e)}")