Update app.py

app.py

@@ -26,12 +26,16 @@ def load_model(model_path):
     )
     return model, tokenizer
 
-current_model, current_tokenizer = load_model("deepseek-ai/DeepSeek-R1-Distill-Llama-8B")
+# 初始化預設模型
+current_model_name = "DeepSeek-R1-Distill-Llama-8B"
+current_model, current_tokenizer = load_model(MODEL_NAMES[current_model_name])
 
 def chat(message, history, model_name):
-    global current_model, current_tokenizer
+    global current_model, current_tokenizer, current_model_name
 
-    if model_name != current_model:
+    # 檢查是否需要更換模型
+    if model_name != current_model_name:
+        current_model_name = model_name
         current_model, current_tokenizer = load_model(MODEL_NAMES[model_name])
 
     device = "cuda" if torch.cuda.is_available() else "cpu"
@@ -43,18 +47,20 @@ def chat(message, history, model_name):
 
 with gr.Blocks() as app:
     gr.Markdown("## Chatbot with DeepSeek Models")
+
     model_selector = gr.Dropdown(
-            choices=list(MODEL_NAMES.keys()),
-            value="DeepSeek-R1-Distill-Llama-8B",
-            label="Select Model",
-        )
+        choices=list(MODEL_NAMES.keys()),
+        value=current_model_name,
+        label="Select Model",
+    )
+
     chat_interface = gr.ChatInterface(
-        chat,
+        fn=lambda message, history: chat(message, history, model_selector.value),
         type="messages",
         flagging_mode="manual",
         save_history=True,
     )
 
+    model_selector.change(fn=lambda model_name: None, inputs=[model_selector], outputs=[])
 
-
-app.launch()
+app.launch()