[web demo] Add thinking process for demo

examples/web_demo/requirements.txt

@@ -1,4 +1,4 @@
-gradio==5.11.0
+gradio==5.18.0
 sentencepiece==0.1.99
 transformers==4.41.2
 transformers_stream_generator==0.0.5

examples/web_demo/web_demo_api.py

@@ -12,7 +12,7 @@ def boolean_string(string):
 
 
 parser = argparse.ArgumentParser()
-parser.add_argument("-u", "--url", type=str, default="http://local:8000/v1", help="base url")
+parser.add_argument("-u", "--url", type=str, default="http://localhost:8000/v1", help="base url")
 parser.add_argument("-m", "--model", type=str, default="xft", help="model name")
 parser.add_argument("-t", "--token", type=str, default="EMPTY", help="api key")
 parser.add_argument("-i", "--ip", type=str, default="0.0.0.0", help="gradio server ip")
@@ -25,7 +25,7 @@ def clean_input():
 
 
 def reset():
-    return [], []
+    return []
 
 
 class ChatDemo:
@@ -43,7 +43,7 @@ def launch(self, server_name="0.0.0.0", server_port=7860, share=False):
         with gr.Blocks() as demo:
             self.html_func()
 
-            chatbot = gr.Chatbot()
+            chatbot = gr.Chatbot(type="messages")
             with gr.Row():
                 with gr.Column(scale=2):
                     user_input = gr.Textbox(show_label=False, placeholder="Input...", lines=1, container=False)
@@ -52,53 +52,61 @@ def launch(self, server_name="0.0.0.0", server_port=7860, share=False):
                 with gr.Column(scale=1):
                     emptyBtn = gr.Button("Clear History")
 
-            history = gr.State([])
             submitBtn.click(
                 self.predict,
-                [user_input, chatbot, history],
-                [chatbot, history],
+                [user_input, chatbot],
+                [chatbot],
                 show_progress=True,
             )
             submitBtn.click(clean_input, [], [user_input])
-            emptyBtn.click(reset, [], [chatbot, history], show_progress=True)
+            emptyBtn.click(reset, [], [chatbot], show_progress=True)
 
         demo.queue().launch(server_name=server_name, server_port=server_port, share=share, inbrowser=True)
 
-    def post_process_generation(self, chunk, chatbot, query, history):
-        response = chatbot[-1][1] + chunk
-        new_history = history + [(query, response)]
-        chatbot[-1] = (query, response)
-        return chatbot, new_history
-
-    def create_chat_input(self, query, history):
+    def post_process_generation(self, chunk, chatbot):
+        if chunk == "<think>":
+            chunk=""
+            chatbot[-1]["metadata"]= {"title": "💭思考过程"}
+        elif chunk == "</think>":
+            if "metadata" in chatbot[-1]:
+                if len(chatbot[-1]["content"].strip()) == 0:
+                    if chatbot[-1]["metadata"] is not None:
+                        del chatbot[-1]["metadata"]
+                else:
+                    chatbot.append({"role": "assistant", "content": ""})
+            chunk=""
+        chatbot[-1]["content"] += chunk
+        return chatbot
+
+    def create_chat_input(self, chatbot):
         msgs = []
-        if history is None:
-            history = []
-        for user_msg, model_msg in history:
-            msgs.append({"role": "user", "content": user_msg})
-            msgs.append({"role": "assistant", "content": model_msg})
-        msgs.append({"role": "user", "content": query})
+        for msg in chatbot:
+            if "metadata" not in msg or msg["metadata"] is None:
+                msgs.append({"role": msg["role"], "content": msg["content"]})
+        if len(msgs) > 8:
+            msgs = msgs[-8:]
         return msgs
 
-    def predict(self, query, chatbot, history):
-        chatbot.append((query, ""))
+    def predict(self, query, chatbot):
+        chatbot.append({"role": "user", "content": query})
+        chatbot.append({"role": "assistant", "content": ""})
 
         completion = self.client.chat.completions.create(
             model=self.model,
-            messages=self.create_chat_input(query, history),
-            max_tokens=2048,
+            messages=self.create_chat_input(chatbot),
+            max_tokens=8192,
             stream=True,
-            temperature=1.0,
+            temperature=0.6,
             extra_body={"top_k": 20, "top_p": 0.8, "repetition_penalty": 1.1},
         )
 
         for chunk in completion:
             if chunk.choices[0].delta.content is not None:
-                yield self.post_process_generation(chunk.choices[0].delta.content, chatbot, query, history)
+                yield self.post_process_generation(chunk.choices[0].delta.content, chatbot)
 
 
 if __name__ == "__main__":
     args = parser.parse_args()
     demo = ChatDemo(args.url, args.model, args.token)
 
-    demo.launch(args.ip, args.port, args.share)
+    demo.launch(args.ip, args.port, args.share)