chore: update something

app.py

@@ -384,7 +384,7 @@ def generate(
     chat_history: list[tuple[str, str]],
     allow_used_tools: bool = True,
     system_prompt: str = "",
-    max_new_tokens: int = 2048,
+    max_new_tokens: int = 1536,
     temperature: float = 0.4,
     top_p: float = 0.95,
     top_k: int = 50,
@@ -468,12 +468,11 @@ def generate(
                     gr.Info("Searching for information on the Google.")
                     document_references = search_with_google(keyword)
 
+        apply_tools = (
+            True if allow_used_tools is True and previous_response is None else False
+        )
         input_ids = build_input_ids(
-            apply_tools=(
-                True
-                if allow_used_tools is True and previous_response is None
-                else False
-            ),
+            apply_tools=apply_tools,
             references=document_references,
         )
         streamer = TextIteratorStreamer(
@@ -505,12 +504,16 @@ def generate(
             if state["mark"] is None:
                 state["mark"] = time.time()
             outputs.append(text)
-            if state["mark"] + waiting_tools_timeout < time.time():
+            if (
+                apply_tools is False
+                or state["mark"] + waiting_tools_timeout < time.time()
+            ):
                 state["respond"] = True
                 yield "".join(outputs)
 
         if (
-            state["respond"] is False
+            apply_tools is True
+            and state["respond"] is False
             and state["mark"] + waiting_tools_timeout > time.time()
         ):
             gr.Info("Searching for information on the internet.")