Update app.py

app.py

@@ -69,7 +69,7 @@ def load_pipeline(model_name):
     return pipe
 
 
-def retrieve_context(query, max_results=6, max_chars=1000):
+def retrieve_context(query, max_results=6, max_chars=600):
     """
     Retrieve search snippets from DuckDuckGo (runs in background).
     Returns a list of result strings.
@@ -248,7 +248,7 @@ def update_default_prompt(enable_search):
 # ------------------------------
 # Gradio UI
 # ------------------------------
-with gr.Blocks(title="Yee-R1-Demo") as demo:
+with gr.Blocks(title="Yee R1 Demo") as demo:
     gr.Markdown("## Yee-R1 Demo")
     gr.Markdown("小熠（Yee）AI 数据安全专家")
     with gr.Row():
@@ -257,8 +257,8 @@ with gr.Blocks(title="Yee-R1-Demo") as demo:
             search_chk = gr.Checkbox(label="Enable Web Search", value=True)
             sys_prompt = gr.Textbox(label="System Prompt", lines=3, value=update_default_prompt(search_chk.value))
             gr.Markdown("### Generation Parameters")
-            max_tok = gr.Slider(64, 16384, value=4096, step=32, label="Max Tokens")
-            temp = gr.Slider(0.1, 2.0, value=0.6, step=0.1, label="Temperature")
+            max_tok = gr.Slider(64, 16384, value=2048, step=32, label="Max Tokens")
+            temp = gr.Slider(0.1, 2.0, value=0.7, step=0.1, label="Temperature")
             k = gr.Slider(1, 100, value=40, step=1, label="Top-K")
             p = gr.Slider(0.1, 1.0, value=0.9, step=0.05, label="Top-P")
             rp = gr.Slider(1.0, 2.0, value=1.2, step=0.1, label="Repetition Penalty")
@@ -269,7 +269,7 @@ with gr.Blocks(title="Yee-R1-Demo") as demo:
             clr = gr.Button("Clear Chat")
             cnl = gr.Button("Cancel Generation")
         with gr.Column(scale=7):
-            chat = gr.Chatbot(type="messages", show_copy_all_button=True)
+            chat = gr.Chatbot(type="messages")
             txt = gr.Textbox(placeholder="Type your message and press Enter...")
             dbg = gr.Markdown()
 
@@ -280,4 +280,4 @@ with gr.Blocks(title="Yee-R1-Demo") as demo:
                inputs=[txt, chat, sys_prompt, search_chk, mr, mc,
                        model_dd, max_tok, temp, k, p, rp, st],
                outputs=[chat, dbg])
-    demo.launch()
+    demo.launch()