《生成式 AI》课程第3講 CODE TASK 任务2:角色扮演的机器人

课程

《生成式 AI》课程第3講：訓練不了人工智慧嗎？你可以訓練你自己-CSDN博客

我们希望你设计一个机器人服务，你可以用LM玩角色扮演游戏。
与LM进行多轮对话
提示:告诉聊天机器人扮演任意角色。
后续输入:与聊天机器人交互。

Part 2: Role-Play

In this task, you are asked to prompt your chatbot into playing a roleplaying game. You should assign it a character, then prompt it into that character.

You need to:

Come up with a character you want the chatbot to act and the prompt to make the chatbot into that character. Fill the character in character_for_chatbot, and fill the prompt in prompt_for_roleplay.
Hit the run button. (The run button will turn into this state when sucessfully executed.) It will pop up an interface that looks like this: (It May look a little bit different if you use dark mode.)
Interact with the chatbot for 2 rounds. Type what you want to say in the block "Input", then hit the button "Send". (You can use the "Temperature" slide to control the creativeness of the output.)
If you want to change your prompt or the character, hit the run button again to stop the cell. Then go back to step 1.
After you get the desired result, hit the button "Export" to save your result. There will be a file named part2.json appears in the file list. Remember to download it to your own computer before it disappears.

Note:

If you hit the "Export" button again, the previous result will be covered, so make sure to download it first.
You should keep in mind that even with the exact same prompt, the output might still
differ.

以下的代码需要魔法，并且有OpenAI的账号才能使用

# TODO: Fill in the below two lines: character_for_chatbot and prompt_for_roleplay
# The first one is the character you want your chatbot to play
# The second one is the prompt to make the chatbot be a certain character
character_for_chatbot = "FILL IN YOUR CHARACTER"
prompt_for_roleplay = "FILL IN YOUR PROMPT"

# function to clear the conversation
def reset() -> List:
    return []

# function to call the model to generate
def interact_roleplay(chatbot: List[Tuple[str, str]], user_input: str, temp=1.0) -> List[Tuple[str, str]]:
    '''
    * Arguments

      - user_input: the user input of each round of conversation

      - temp: the temperature parameter of this model. Temperature is used to control the output of the chatbot.
              The higher the temperature is, the more creative response you will get.

    '''
    try:
        messages = []
        for input_text, response_text in chatbot:
            messages.append({'role': 'user', 'content': input_text})
            messages.append({'role': 'assistant', 'content': response_text})

        messages.append({'role': 'user', 'content': user_input})

        response = client.chat.completions.create(
            model="gpt-3.5-turbo",
            messages = messages,
            temperature = temp,
            max_tokens=200,
        )
        chatbot.append((user_input, response.choices[0].message.content))

    except Exception as e:
        print(f"Error occurred: {e}")
        chatbot.append((user_input, f"Sorry, an error occurred: {e}"))
    return chatbot

# function to export the whole conversation log
def export_roleplay(chatbot: List[Tuple[str, str]], description: str) -> None:
    '''
    * Arguments

      - chatbot: the model itself, the conversation is stored in list of tuples

      - description: the description of this task

    '''
    target = {"chatbot": chatbot, "description": description}
    with open("part2.json", "w") as file:
        json.dump(target, file)

first_dialogue = interact_roleplay([], prompt_for_roleplay)

# this part constructs the Gradio UI interface
with gr.Blocks() as demo:
    gr.Markdown(f"# Part2: Role Play\nThe chatbot wants to play a role game with you, try interacting with it!!")
    chatbot = gr.Chatbot(value = first_dialogue)
    description_textbox = gr.Textbox(label=f"The character the bot is playing", interactive = False, value=f"{character_for_chatbot}")
    input_textbox = gr.Textbox(label="Input", value = "")
    with gr.Column():
        gr.Markdown("#  Temperature\n Temperature is used to control the output of the chatbot. The higher the temperature is, the more creative response you will get.")
        temperature_slider = gr.Slider(0.0, 2.0, 1.0, step = 0.1, label="Temperature")
    with gr.Row():
        sent_button = gr.Button(value="Send")
        reset_button = gr.Button(value="Reset")
    with gr.Column():
        gr.Markdown("#  Save your Result.\n After you get a satisfied result. Click the export button to recode it.")
        export_button = gr.Button(value="Export")
    sent_button.click(interact_roleplay, inputs=[chatbot, input_textbox, temperature_slider], outputs=[chatbot])
    reset_button.click(reset, outputs=[chatbot])
    export_button.click(export_roleplay, inputs=[chatbot, description_textbox])

# loads the conversation log json file
with open("part2.json", "r") as f:
    context = json.load(f)

# traverse through and load the conversation log properly
chatbot = context['chatbot']
role = context['description']
dialogue = ""
for i, (user, bot) in enumerate(chatbot):
    if i != 0:
        dialogue += f"User: {user}\n"
    dialogue += f"Bot: {bot}\n"

# this part constructs the Gradio UI interface
with gr.Blocks() as demo:
    gr.Markdown(f"# Part2: Role Play\nThe chatbot wants to play a role game with you, try interacting with it!!")
    chatbot = gr.Chatbot(value = context['chatbot'])
    description_textbox = gr.Textbox(label=f"The character the bot is playing", interactive = False, value=context['description'])
    with gr.Column():
        gr.Markdown("# Copy this part to the grading system.")
        gr.Textbox(label = "role", value = role, show_copy_button = True)
        gr.Textbox(label = "dialogue", value = dialogue, show_copy_button = True)

demo.launch(debug = True)

以下是用 API2D 调用 openAI 'model': 'gpt-4o-mini' url = "https://openai.api2d.net/v1/chat/completions"

import requests
import gradio as gr
import json


def get_response(input_text, chat_history):
    """
    根据用户输入和已有的对话历史获取语言模型的回复。
    :param input_text: 当前用户输入的文本内容。
    :param chat_history: 之前的对话历史，是一个包含二元组的列表，每个二元组分别是 (用户消息, 模型回复)。
    :return: 返回状态码、包含模型相关信息及回复内容和token数量信息的格式化字符串，若JSON解析出错则返回相应错误提示。
    """
    url = "https://openai.api2d.net/v1/chat/completions"
    headers = {
        'Content-Type': 'application/json',
        'Authorization': 'Bearer fk0000'  # <-- 把 fkxxxxx 替换成你自己的 Forward Key，注意前面的 Bearer 要保留，并且和 Key 中间有一个空格。
    }
    messages = []

    # 将之前的对话历史添加到消息列表中，格式需符合API要求，依次添加用户消息和模型回复消息
    for user_msg, bot_msg in chat_history:
        messages.append({'role': 'user', 'content': user_msg})
        messages.append({'role': 'assistant', 'content': bot_msg})
    # 先添加当前用户输入的消息
    messages.append({'role': 'user', 'content': input_text})
    data = {
        'model': 'gpt-4o-mini',  # 'gpt-3.5-turbo',
        'messages': messages
    }
    response = requests.post(url, headers=headers, json=data)
    status_code = response.status_code
    try:
        json_data = response.json()
        # 提取模型名称
        model_name = json_data.get('model', '未知模型')
        # 提取助手回复的内容
        assistant_content = json_data.get('choices', [])[0].get('message', {}).get('content', '无回复内容')
        # 提取各类token数量
        prompt_tokens = json_data.get('usage', {}).get('prompt_tokens', 0)
        completion_tokens = json_data.get('usage', {}).get('completion_tokens', 0)
        total_tokens = json_data.get('usage', {}).get('total_tokens', 0)


        # 将本次的用户输入和模型回复添加到对话历史中
        chat_history.append((input_text, assistant_content))
        print(chat_history)

        return chat_history, f"模型: {model_name}\n回复内容: {assistant_content}\n提示词token数: {prompt_tokens}\n回复内容token数: {completion_tokens}\n总token数: {total_tokens}"


    except json.JSONDecodeError:
        return status_code, []
        # 如果解析JSON出错，返回空列表，避免给Chatbot传递不符合格式的数据

# 创建Gradio界面
with gr.Blocks() as demo:
    chatbot = gr.Chatbot()
    user_input = gr.Textbox(lines=2, placeholder="请输入你想发送的内容")
    state = gr.State([])  # 创建一个状态变量，用于存储对话历史，初始化为空列表

    # 通过按钮点击事件触发获取回复和更新对话历史等操作
    send_button = gr.Button("发送")
    send_button.click(
        fn=get_response,
        inputs=[user_input, state],
        outputs=[chatbot, gr.Textbox(label="解析后的响应内容")]
    )

    demo.launch(debug=True)