update

app.py

@@ -35,13 +35,22 @@ TODO: 使用说明
 with gr.Blocks() as demo:
     # Knowledge Distillation through Self Chatting
     gr.HTML("""<h1 align="center">Distilling the Knowledge through Self Chatting</h1>""")
-    system = gr.Dropdown(
-        choices=system_list,
-        value=system_list[0],
-        allow_custom_value=True,
-        interactive=True,
-        label="System message"
-    )
+    with gr.Row():
+        system = gr.Dropdown(
+            choices=system_list,
+            value=system_list[0],
+            allow_custom_value=True,
+            interactive=True,
+            label="System message",
+            scale=4,
+        )
+        gr.Dropdown(
+            ["Qwen2-0.5B-Instruct", "llama3.1", "gemini"],
+            value="Qwen2-0.5B-Instruct",
+            label="model",
+            interactive=True,
+            scale=1,
+        )
     chatbot = gr.Chatbot(avatar_images=("assets/man.png", "assets/bot.png"))
     with gr.Row():
         with gr.Column(scale=4):
@@ -52,15 +61,6 @@ with gr.Blocks() as demo:
                 undo_btn = gr.Button("↩️ Undo")
                 clear_btn = gr.Button("🗑️  Clear")  # 🧹 Clear History (清除历史)
                 stop_btn = gr.Button("停止生成", variant="primary")
-        with gr.Column(scale=1):
-            # generate_query_btn = gr.Button("Generate First Query")
-            gr.Dropdown(
-                ["moss", "chatglm-2", "chatpdf"],
-                value="moss",
-                label="model",
-                interactive=True,
-                # info="Will add more animals later!"
-            )
 
         slider_max_tokens = gr.Slider(minimum=1, maximum=config.MAX_SEQUENCE_LENGTH,
                                       value=config.DEFAULT_MAX_TOKENS, step=1, label="Max tokens")
@@ -82,12 +82,10 @@ with gr.Blocks() as demo:
 
     generate_btn.click(generate, [chatbot, history], outputs=[generated_text, chatbot, history],
                        show_progress="full")
-    retry_btn.click(undo_generate, [chatbot, history], outputs=[generated_text, chatbot, history],
-                    show_progress="full")
+    retry_btn.click(undo_generate, [chatbot, history], outputs=[generated_text, chatbot, history])
     retry_btn.click(generate, [chatbot, history], outputs=[generated_text, chatbot, history],
                     show_progress="full")
-    undo_btn.click(undo_generate, [chatbot, history], outputs=[generated_text, chatbot, history],
-                   show_progress="full")
+    undo_btn.click(undo_generate, [chatbot, history], outputs=[generated_text, chatbot, history])
 
     slider_max_tokens.change(set_max_tokens, inputs=[slider_max_tokens])
     slider_top_p.change(set_top_p, inputs=[slider_top_p])

app_util.py

@@ -1,3 +1,4 @@
+import json
 import gradio as gr
 from utils.logging_util import logger
 from models.cpp_qwen2 import bot
@@ -72,12 +73,17 @@ def generate(chatbot, history):
         yield out
 
 
-def regenerate():
-    """
-    删除上一轮，重新生成。
-    :return:
-    """
-    pass
+def undo_generate(chatbot, history):
+    if history[-1]["role"] == "user":
+        history = history[:-1]
+        chatbot = chatbot[:-1]
+    elif history[-1]["role"] == "assistant":
+        history = history[:-1]
+        chatbot[-1] = (chatbot[-1][0], None)
+    else:
+        pass
+    logger.info(f"after undo, {json.dumps(chatbot, ensure_ascii=False)}, {json.dumps(history, ensure_ascii=False)}")
+    return "", chatbot, history
 
 
 def reset_user_input():
@@ -98,15 +104,3 @@ def set_top_p(top_p):
 
 def set_temperature(temperature):
     bot.generation_kwargs["temperature"] = temperature
-
-
-def undo_generate(chatbot, history):
-    if history[-1]["role"] == "user":
-        history = history[:-1]
-        chatbot = chatbot[:-1]
-    elif history[-1]["role"] == "assistant":
-        history = history[:-1]
-        chatbot[-1] = (chatbot[-1][0], None)
-    else:
-        pass
-    return "", chatbot, history

config.py

@@ -2,7 +2,7 @@
 
 MAX_SEQUENCE_LENGTH = 2048  # max_seq_len
 
-DEFAULT_MAX_TOKENS = 512
+DEFAULT_MAX_TOKENS = 128
 # DEFAULT_MAX_NEW_TOKENS = None
 DEFAULT_TOP_K = 100
 DEFAULT_TOP_P = 0.95

models/cpp_qwen2.py

@@ -104,6 +104,7 @@ class Qwen2Simulator(Simulator):
         return output_text
 
     def _stream_generate(self, inputs):
+        logger.info(f"generation_kwargs {self.generation_kwargs}")
         output = self.llm(
             inputs,
             stream=True,