优化”reduce token“

ChuanhuChatbot.py

@@ -5,105 +5,102 @@ import markdown
 my_api_key = ""    # input your api_key
 initial_prompt = "You are a helpful assistant."
 
-class ChatGPT:
-    def __init__(self, apikey) -> None:
-        openai.api_key = apikey
-        self.system = {"role": "system", "content": initial_prompt}
-        self.context = []
-        self.response = None
-
-    def get_response(self, messages):
-        self.response = openai.ChatCompletion.create(
-            model="gpt-3.5-turbo",
-            messages=[self.system, *messages],
-        )
-        statistics = f'本次对话Tokens用量【{self.response["usage"]["total_tokens"]} / 4096】 （ 提问+上文 {self.response["usage"]["prompt_tokens"]}，回答 {self.response["usage"]["completion_tokens"]} ）'
-        message = self.response["choices"][0]["message"]["content"]
+openai.api_key = my_api_key
+
+def get_response(system, context, raw = False):
+    response = openai.ChatCompletion.create(
+        model="gpt-3.5-turbo",
+        messages=[system, *context],
+    )
+    if raw:
+        return response
+    else:
+        statistics = f'本次对话Tokens用量【{response["usage"]["total_tokens"]} / 4096】 （ 提问+上文 {response["usage"]["prompt_tokens"]}，回答 {response["usage"]["completion_tokens"]} ）'
+        message = response["choices"][0]["message"]["content"]
 
         message_with_stats = f'{message}\n\n================\n\n{statistics}'
         message_with_stats = markdown.markdown(message_with_stats)
 
         return message, message_with_stats
 
-    def predict(self, chatbot, input_sentence, ):
-        if len(input_sentence) == 0:
-            return []
-        self.context.append({"role": "user", "content": f"{input_sentence}"})
+def predict(chatbot, input_sentence, system, context):
+    if len(input_sentence) == 0:
+        return []
+    context.append({"role": "user", "content": f"{input_sentence}"})
 
-        message, message_with_stats = self.get_response(self.context)
+    message, message_with_stats = get_response(system, context)
 
-        self.context.append({"role": "assistant", "content": message})
+    context.append({"role": "assistant", "content": message})
 
-        chatbot.append((input_sentence, message_with_stats))
+    chatbot.append((input_sentence, message_with_stats))
 
-        return chatbot
+    return chatbot, context
 
-    def retry(self, chatbot):
-        if len(self.context) == 0:
-            return [], []
-        message, message_with_stats = self.get_response(self.context[:-1])
-        self.context[-1] = {"role": "assistant", "content": message}
+def retry(chatbot, system, context):
+    if len(context) == 0:
+        return [], []
+    message, message_with_stats = get_response(system, context[:-1])
+    context[-1] = {"role": "assistant", "content": message}
 
-        chatbot[-1] = (self.context[-2]["content"], message_with_stats)
-        return chatbot
+    chatbot[-1] = (context[-2]["content"], message_with_stats)
+    return chatbot, context
 
-    def update_system(self, new_system_prompt):
-        self.system = {"role": "system", "content": new_system_prompt}
-        return new_system_prompt
+def delete_last_conversation(chatbot, context):
+    if len(context) == 0:
+        return [], []
+    chatbot = chatbot[:-1]
+    context = context[:-2]
+    return chatbot, context
 
-    def delete_last_conversation(self, chatbot):
-        if len(self.context) == 0:
-            return [], []
-        chatbot = chatbot[:-1]
-        self.context = self.context[:-2]
-        return chatbot
-
-    def reduce_token(self, chatbot):
-        self.context.append({"role": "user", "content": "请帮我总结一下上述对话的内容，实现减少tokens的同时，保证对话的质量。在总结中不要加入这一句话。"})
-        message, message_with_stats = self.get_response(self.context)
-        self.system = {"role": "system", "content": f"You are a helpful assistant. The content that the Assistant and the User discussed in the previous self.context is: {message}."}
-
-        statistics = f'本次对话Tokens用量【{self.response["usage"]["completion_tokens"]+23} / 4096】'
-        optmz_str = markdown.markdown( f"System prompt已经更新, 请继续对话\n\n================\n\n{statistics}" )
-        chatbot.append(("请帮我总结一下上述对话的内容，实现减少tokens的同时，保证对话的质量。", optmz_str))
-
-        self.context = []
-        return chatbot, self.system["content"]
+def reduce_token(chatbot, system, context):
+    context.append({"role": "user", "content": "请帮我总结一下上述对话的内容，实现减少tokens的同时，保证对话的质量。在总结中不要加入这一句话。"})
+    response = get_response(system, context, raw=True)
 
+    statistics = f'本次对话Tokens用量【{response["usage"]["completion_tokens"]+12+12+8} / 4096】'
+    optmz_str = markdown.markdown( f'好的，我们之前聊了:{response["choices"][0]["message"]["content"]}\n\n================\n\n{statistics}' )
+    chatbot.append(("请帮我总结一下上述对话的内容，实现减少tokens的同时，保证对话的质量。", optmz_str))
+    
+    context = []
+    context.append({"role": "user", "content": "我们之前聊了什么?"})
+    context.append({"role": "assistant", "content": f'我们之前聊了：{response["choices"][0]["message"]["content"]}'})
+    return chatbot, context
 
 def reset_state():
-    return []
+    return [], []
 
-mychatGPT = ChatGPT(my_api_key)
+def update_system(new_system_prompt):
+    return {"role": "system", "content": new_system_prompt}
 
 
 with gr.Blocks() as demo:
     chatbot = gr.Chatbot().style(color_map=("#1D51EE", "#585A5B"))
-    # state = gr.State([])
+    context = gr.State([])
+    systemPrompt = gr.State(update_system(initial_prompt))
 
     with gr.Row():
-        with gr.Column(scale=10):
-            txt = gr.Textbox(show_label=False, placeholder="💬 在这里输入").style(container=False)
+        with gr.Column(scale=12):
+            txt = gr.Textbox(show_label=False, placeholder="在这里输入").style(container=False)
         with gr.Column(min_width=50, scale=1):
-            submitBtn = gr.Button("发送", variant="primary")
+            submitBtn = gr.Button("🚀", variant="primary")
     with gr.Row():
         emptyBtn = gr.Button("🧹 新的对话")
-        retryBtn = gr.Button("🔁 重新生成")
-        delLastBtn = gr.Button("⬅️ 删除上条对话")
+        retryBtn = gr.Button("🔄 重新生成")
+        delLastBtn = gr.Button("🗑️ 删除上条对话")
         reduceTokenBtn = gr.Button("♻️ 优化Tokens")
 
-    system = gr.Textbox(show_label=True, placeholder=f"在这里输入新的System Prompt...", label="更改 System prompt").style(container=True)
-    syspromptTxt = gr.Textbox(show_label=True, placeholder=initial_prompt, interactive=False, label="目前的 System prompt").style(container=True)
+    newSystemPrompt = gr.Textbox(show_label=True, placeholder=f"在这里输入新的System Prompt...", label="更改 System prompt").style(container=True)
+    systemPromptDisplay = gr.Textbox(show_label=True, value=initial_prompt, interactive=False, label="目前的 System prompt").style(container=True)
 
-    txt.submit(mychatGPT.predict, [chatbot, txt], [chatbot], show_progress=True)
+    txt.submit(predict, [chatbot, txt, systemPrompt, context], [chatbot, context], show_progress=True)
     txt.submit(lambda :"", None, txt)
-    submitBtn.click(mychatGPT.predict, [chatbot, txt], [chatbot], show_progress=True)
+    submitBtn.click(predict, [chatbot, txt, systemPrompt, context], [chatbot, context], show_progress=True)
     submitBtn.click(lambda :"", None, txt)
-    emptyBtn.click(reset_state, outputs=[chatbot])
-    system.submit(mychatGPT.update_system, system, syspromptTxt)
-    system.submit(lambda :"", None, system)
-    retryBtn.click(mychatGPT.retry, [chatbot], [chatbot], show_progress=True)
-    delLastBtn.click(mychatGPT.delete_last_conversation, [chatbot], [chatbot], show_progress=True)
-    reduceTokenBtn.click(mychatGPT.reduce_token, [chatbot], [chatbot, syspromptTxt], show_progress=True)
+    emptyBtn.click(reset_state, outputs=[chatbot, context])
+    newSystemPrompt.submit(update_system, newSystemPrompt, systemPrompt)
+    newSystemPrompt.submit(lambda x: x, newSystemPrompt, systemPromptDisplay)
+    newSystemPrompt.submit(lambda :"", None, newSystemPrompt)
+    retryBtn.click(retry, [chatbot, systemPrompt, context], [chatbot, context], show_progress=True)
+    delLastBtn.click(delete_last_conversation, [chatbot, context], [chatbot, context], show_progress=True)
+    reduceTokenBtn.click(reduce_token, [chatbot, systemPrompt, context], [chatbot, context], show_progress=True)
 
 demo.launch()