重大更新：支持像官方网页那样实时传输了；改进的保存/加载机制

ChuanhuChatbot.py

@@ -1,14 +1,15 @@
 import json
 import gradio as gr
-import openai
 import os
 import sys
 import traceback
-# import markdown
+import requests
 
-my_api_key = ""    # 在这里输入你的 API 密钥
+my_api_key = "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"    # 在这里输入你的 API 密钥
 initial_prompt = "You are a helpful assistant."
 
+API_URL = "https://api.openai.com/v1/chat/completions"
+
 if my_api_key == "":
     my_api_key = os.environ.get('my_api_key')
 
@@ -16,17 +17,11 @@ if my_api_key == "empty":
     print("Please give a api key!")
     sys.exit(1)
 
-if my_api_key == "":
-    initial_keytxt = None
-elif len(str(my_api_key)) == 51:
-    initial_keytxt = "默认api-key（未验证）：" + str(my_api_key[:4] + "..." + my_api_key[-4:])
-else:
-    initial_keytxt = "默认api-key无效，请重新输入"
 
 def parse_text(text):
     lines = text.split("\n")
     count = 0
-    for i,line in enumerate(lines):
+    for i, line in enumerate(lines):
         if "```" in line:
             count += 1
             items = line.split('`')
@@ -46,190 +41,208 @@ def parse_text(text):
                 lines[i] = '<br/>'+line
     return "".join(lines)
 
-def get_response(system, context, myKey, raw = False):
-    openai.api_key = myKey
-    response = openai.ChatCompletion.create(
-        model="gpt-3.5-turbo",
-        messages=[system, *context],
-    )
-    openai.api_key = ""
-    if raw:
-        return response
+def predict(inputs, top_p, temperature, openai_api_key, chatbot=[], history=[], system_prompt=initial_prompt, retry=False, summary=False):  # repetition_penalty, top_k
+
+    headers = {
+        "Content-Type": "application/json",
+        "Authorization": f"Bearer {openai_api_key}"
+    }
+
+    chat_counter = len(history) // 2
+
+    print(f"chat_counter - {chat_counter}")
+
+    messages = [compose_system(system_prompt)]
+    if chat_counter:
+        for data in chatbot:
+            temp1 = {}
+            temp1["role"] = "user"
+            temp1["content"] = data[0]
+            temp2 = {}
+            temp2["role"] = "assistant"
+            temp2["content"] = data[1]
+            if temp1["content"] != "":
+                messages.append(temp1)
+                messages.append(temp2)
+            else:
+                messages[-1]['content'] = temp2['content']
+    if retry and chat_counter:
+        messages.pop()
+    elif summary and chat_counter:
+        messages.append(compose_user(
+            "请帮我总结一下上述对话的内容，实现减少字数的同时，保证对话的质量。在总结中不要加入这一句话。"))
+        history = ["我们刚刚聊了什么？"]
     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, parse_text(message_with_stats)
-
-def predict(chatbot, input_sentence, system, context, myKey):
-    if len(input_sentence) == 0:
-        return []
-    context.append({"role": "user", "content": f"{input_sentence}"})
-
-    try:
-        message, message_with_stats = get_response(system, context, myKey)
-    except openai.error.AuthenticationError:
-        chatbot.append((input_sentence, "请求失败，请检查API-key是否正确。"))
-        return chatbot, context
-    except openai.error.Timeout:
-        chatbot.append((input_sentence, "请求超时，请检查网络连接。"))
-        return chatbot, context
-    except openai.error.APIConnectionError:
-        chatbot.append((input_sentence, "连接失败，请检查网络连接。"))
-        return chatbot, context
-    except openai.error.RateLimitError:
-        chatbot.append((input_sentence, "请求过于频繁，请5s后再试。"))
-        return chatbot, context
-    except:
-        chatbot.append((input_sentence, "发生了未知错误Orz"))
-        return chatbot, context
-
-    context.append({"role": "assistant", "content": message})
-
-    chatbot.append((input_sentence, message_with_stats))
-
-    return chatbot, context
-
-def retry(chatbot, system, context, myKey):
-    if len(context) == 0:
-        return [], []
-    
-    try:
-        message, message_with_stats = get_response(system, context[:-1], myKey)
-    except openai.error.AuthenticationError:
-        chatbot.append(("重试请求", "请求失败，请检查API-key是否正确。"))
-        return chatbot, context
-    except openai.error.Timeout:
-        chatbot.append(("重试请求", "请求超时，请检查网络连接。"))
-        return chatbot, context
-    except openai.error.APIConnectionError:
-        chatbot.append(("重试请求", "连接失败，请检查网络连接。"))
-        return chatbot, context
-    except openai.error.RateLimitError:
-        chatbot.append(("重���请求", "请求过于频繁，请5s后再试。"))
-        return chatbot, context
-    except:
-        chatbot.append(("重试请求", "发生了未知错误Orz"))
-        return chatbot, context
-    
-    context[-1] = {"role": "assistant", "content": message}
-
-    chatbot[-1] = (context[-2]["content"], message_with_stats)
-    return chatbot, context
-
-def delete_last_conversation(chatbot, context):
-    if len(context) == 0:
-        return [], []
-    chatbot = chatbot[:-1]
-    context = context[:-2]
-    return chatbot, context
-
-def reduce_token(chatbot, system, context, myKey):
-    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, myKey, raw=True)
-
-    statistics = f'本次对话Tokens用量【{response["usage"]["completion_tokens"]+12+12+8} / 4096】'
-    optmz_str = parse_text( 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 save_chat_history(filepath, system, context):
+        temp3 = {}
+        temp3["role"] = "user"
+        temp3["content"] = inputs
+        messages.append(temp3)
+        chat_counter += 1
+    # messages
+    payload = {
+        "model": "gpt-3.5-turbo",
+        "messages": messages,  # [{"role": "user", "content": f"{inputs}"}],
+        "temperature": temperature,  # 1.0,
+        "top_p": top_p,  # 1.0,
+        "n": 1,
+        "stream": True,
+        "presence_penalty": 0,
+        "frequency_penalty": 0,
+    }
+
+    if not summary:
+        history.append(inputs)
+    print(f"payload is - {payload}")
+    # make a POST request to the API endpoint using the requests.post method, passing in stream=True
+    response = requests.post(API_URL, headers=headers,
+                             json=payload, stream=True)
+    #response = requests.post(API_URL, headers=headers, json=payload, stream=True)
+
+    token_counter = 0
+    partial_words = ""
+
+    counter = 0
+    chatbot.append((history[-1], ""))
+    for chunk in response.iter_lines():
+        if counter == 0:
+            counter += 1
+            continue
+        counter += 1
+        # check whether each line is non-empty
+        if chunk:
+            # decode each line as response data is in bytes
+            if len(json.loads(chunk.decode()[6:])['choices'][0]["delta"]) == 0:
+                break
+            #print(json.loads(chunk.decode()[6:])['choices'][0]["delta"]    ["content"])
+            partial_words = partial_words + \
+                json.loads(chunk.decode()[6:])[
+                    'choices'][0]["delta"]["content"]
+            if token_counter == 0:
+                history.append(" " + partial_words)
+            else:
+                history[-1] = partial_words
+            chatbot[-1] = (history[-2], history[-1])
+        #   chat = [(history[i], history[i + 1]) for i in range(0, len(history)     - 1, 2) ]  # convert to tuples of list
+            token_counter += 1
+            # resembles {chatbot: chat,     state: history}
+            yield chatbot, history
+
+
+
+def delete_last_conversation(chatbot, history):
+    if chat_counter > 0:
+        chat_counter -= 1
+        chatbot.pop()
+        history.pop()
+        history.pop()
+    return chatbot, history
+
+def save_chat_history(filepath, system, history, chatbot):
     if filepath == "":
         return
-    history = {"system": system, "context": context}
-    with open(f"{filepath}.json", "w") as f:
-        json.dump(history, f)
+    if not filepath.endswith(".json"):
+        filepath += ".json"
+    json_s = {"system": system, "history": history, "chatbot": chatbot}
+    with open(filepath, "w") as f:
+        json.dump(json_s, f)
 
-def load_chat_history(fileobj):
-    with open(fileobj.name, "r") as f:
-        history = json.load(f)
-    context = history["context"]
-    chathistory = []
-    for i in range(0, len(context), 2):
-        chathistory.append((parse_text(context[i]["content"]), parse_text(context[i+1]["content"])))
-    return chathistory , history["system"], context, history["system"]["content"]
 
-def get_history_names():
-    with open("history.json", "r") as f:
-        history = json.load(f)
-    return list(history.keys())
+def load_chat_history(filename):
+    with open(filename, "r") as f:
+        json_s = json.load(f)
+    return filename, json_s["system"], json_s["history"], json_s["chatbot"]
+
+
+def get_history_names(plain=False):
+    # find all json files in the current directory and return their names
+    files = [f for f in os.listdir() if f.endswith(".json")]
+    if plain:
+        return files
+    else:
+        return gr.Dropdown.update(choices=files)
 
 
 def reset_state():
     return [], []
 
-def update_system(new_system_prompt):
-    return {"role": "system", "content": new_system_prompt}
-
-def set_apikey(new_api_key, myKey):
-    old_api_key = myKey
-    
-    try:
-        get_response(update_system(initial_prompt), [{"role": "user", "content": "test"}], new_api_key)
-    except openai.error.AuthenticationError:
-        return "无效的api-key", myKey
-    except openai.error.Timeout:
-        return "请求超时，请检查网络设置", myKey
-    except openai.error.APIConnectionError:
-        return "网络错误", myKey
-    except:
-        return "发生了未知错误Orz", myKey
-    
-    encryption_str = "验证成功，api-key已做遮挡处理：" + new_api_key[:4] + "..." + new_api_key[-4:]
-    return encryption_str, new_api_key
+
+def compose_system(system_prompt):
+    return {"role": "system", "content": system_prompt}
+
+
+def compose_user(user_input):
+    return {"role": "user", "content": user_input}
+
+
+def reset_textbox():
+    return gr.update(value='')
 
 
 with gr.Blocks() as demo:
-    keyTxt = gr.Textbox(show_label=True, placeholder=f"在这里输入你的OpenAI API-key...", value=initial_keytxt, label="API Key").style(container=True)
-    chatbot = gr.Chatbot().style(color_map=("#1D51EE", "#585A5B"))
-    context = gr.State([])
-    systemPrompt = gr.State(update_system(initial_prompt))
-    myKey = gr.State(my_api_key)
+    keyTxt = gr.Textbox(show_label=True, placeholder=f"在这里输入你的OpenAI API-key...",
+                        value=my_api_key, label="API Key", type="password").style(container=True)
+    chatbot = gr.Chatbot()  # .style(color_map=("#1D51EE", "#585A5B"))
+    history = gr.State([])
+    TRUECOMSTANT = gr.State(True)
+    FALSECONSTANT = gr.State(False)
     topic = gr.State("未命名对话历史记录")
 
     with gr.Row():
         with gr.Column(scale=12):
-            txt = gr.Textbox(show_label=False, placeholder="在这里输入").style(container=False)
+            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("🚀", 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")
-        reduceTokenBtn = gr.Button("♻️ 优化Tokens")
-    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)
-    with gr.Accordion(label="保存/加载对话历史记录(在文本框中输入文件名，点击“保存对话”按钮，历史记录文件会被存储到本地)", open=False):
+        reduceTokenBtn = gr.Button("♻️ 总结对话")
+    systemPromptTxt = gr.Textbox(show_label=True, placeholder=f"在这里输入System Prompt...",
+                                 label="System prompt", value=initial_prompt).style(container=True)
+    with gr.Accordion(label="保存/加载对话历史记录(在文本框中输入文件名，点击“保存对话”按钮，历史记录文件会被存储到Python文件旁边)", open=False):
         with gr.Column():
             with gr.Row():
                 with gr.Column(scale=6):
-                    saveFileName = gr.Textbox(show_label=True, placeholder=f"在这里输入保存的文件名...", label="保存对话", value="对话历史记录").style(container=True)
+                    saveFileName = gr.Textbox(
+                        show_label=True, placeholder=f"在这里输入保存的文件名...", label="设置保存文件名", value="对话历史记录").style(container=True)
                 with gr.Column(scale=1):
                     saveBtn = gr.Button("💾 保存对话")
-                    uploadBtn = gr.UploadButton("📂 读取对话", file_count="single", file_types=["json"])
-
-    txt.submit(predict, [chatbot, txt, systemPrompt, context, myKey], [chatbot, context], show_progress=True)
-    txt.submit(lambda :"", None, txt)
-    submitBtn.click(predict, [chatbot, txt, systemPrompt, context, myKey], [chatbot, context], show_progress=True)
-    submitBtn.click(lambda :"", None, txt)
-    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, myKey], [chatbot, context], show_progress=True)
-    delLastBtn.click(delete_last_conversation, [chatbot, context], [chatbot, context], show_progress=True)
-    reduceTokenBtn.click(reduce_token, [chatbot, systemPrompt, context, myKey], [chatbot, context], show_progress=True)
-    keyTxt.submit(set_apikey, [keyTxt, myKey], [keyTxt, myKey], show_progress=True)
-    uploadBtn.upload(load_chat_history, uploadBtn, [chatbot, systemPrompt, context, systemPromptDisplay], show_progress=True)
-    saveBtn.click(save_chat_history, [saveFileName, systemPrompt, context], None, show_progress=True)
-
-
-demo.launch()
+            with gr.Row():
+                with gr.Column(scale=6):
+                    uploadDropdown = gr.Dropdown(label="从列表中加载对话", choices=get_history_names(plain=True), multiselect=False)
+                with gr.Column(scale=1):
+                    refreshBtn = gr.Button("🔄 刷新")
+                    uploadBtn = gr.Button("📂 读取对话")
+    #inputs, top_p, temperature, top_k, repetition_penalty
+    with gr.Accordion("参数", open=False):
+        top_p = gr.Slider(minimum=-0, maximum=1.0, value=1.0, step=0.05,
+                          interactive=True, label="Top-p (nucleus sampling)",)
+        temperature = gr.Slider(minimum=-0, maximum=5.0, value=1.0,
+                                step=0.1, interactive=True, label="Temperature",)
+        #top_k = gr.Slider( minimum=1, maximum=50, value=4, step=1, interactive=True, label="Top-k",)
+        #repetition_penalty = gr.Slider( minimum=0.1, maximum=3.0, value=1.03, step=0.01, interactive=True, label="Repetition Penalty", )
+
+    txt.submit(predict, [txt, top_p, temperature, keyTxt,
+               chatbot, history, systemPromptTxt], [chatbot, history])
+    txt.submit(reset_textbox, [], [txt])
+    submitBtn.click(predict, [txt, top_p, temperature, keyTxt, chatbot,
+                    history, systemPromptTxt], [chatbot, history], show_progress=True)
+    submitBtn.click(reset_textbox, [], [txt])
+    emptyBtn.click(reset_state, outputs=[chatbot, history])
+    retryBtn.click(predict, [txt, top_p, temperature, keyTxt, chatbot, history,
+                   systemPromptTxt, TRUECOMSTANT], [chatbot, history], show_progress=True)
+    delLastBtn.click(delete_last_conversation, [chatbot, history], [
+                     chatbot, history], show_progress=True)
+    reduceTokenBtn.click(predict, [txt, top_p, temperature, keyTxt, chatbot, history,
+                         systemPromptTxt, FALSECONSTANT, TRUECOMSTANT], [chatbot, history], show_progress=True)
+    saveBtn.click(save_chat_history, [
+                  saveFileName, systemPromptTxt, history, chatbot], None, show_progress=True)
+    saveBtn.click(get_history_names, None, [uploadDropdown])
+    refreshBtn.click(get_history_names, None, [uploadDropdown])
+    # uploadBtn.upload(load_chat_history, uploadBtn, [
+    #                  saveFileName, systemPromptTxt, history, chatbot], show_progress=True)
+    uploadBtn.click(load_chat_history, [uploadDropdown],  [saveFileName, systemPromptTxt, history, chatbot], show_progress=True)
+
+
+demo.queue().launch(debug=True)