减少切换模型时的内存/显存占用

modules/models.py

@@ -391,6 +391,7 @@ class LLaMA_Client(BaseLLMModel):
 
 class ModelManager:
     def __init__(self, **kwargs) -> None:
+        self.model = None
         self.get_model(**kwargs)
 
     def get_model(
@@ -409,7 +410,7 @@ class ModelManager:
         dont_change_lora_selector = False
         if model_type != ModelType.OpenAI:
             config.local_embedding = True
-        self.model = None
+        del self.model
         model = None
         try:
             if model_type == ModelType.OpenAI:

modules/presets.py

@@ -80,6 +80,9 @@ MODELS = [
     "llama-65b-hf",
 ]  # 可选的模型
 
+os.makedirs("models", exist_ok=True)
+os.makedirs("lora", exist_ok=True)
+os.makedirs("history", exist_ok=True)
 for dir_name in os.listdir("models"):
     if os.path.isdir(os.path.join("models", dir_name)):
         if dir_name not in MODELS: