admin管理员组

文章数量:1631717

视频讲解

在过去我通过chatgpt调用api时只知道进行孤立的调用,即这一次调用时,chatgpt并没有拿到上一次调用的上下文,这无疑损失很大。通过探索,我知道了如何通过修改messages这个字典类型的list来告知chatgpt我和它的聊天历史。

关键代码如下

def generate_chat_completion(question):

    data = {
        'model': 'gpt-3.5-turbo',
        'messages': [
            {"role": "system", "content": "You are a helpful calendar assistant"},
            {"role": "user", "content": "六月的下一个月有多少天?"},
            {"role": "assistant", "content": "七月有31天。"},
            {"role": "user", "content": "八月有多少天?"},
            {"role": "assistant", "content": "八月也有31天。"},
            {"role": "user", "content": "二月有多少天?"},
            {"role": "assistant", "content": "这要分平年和闰年"},
            {"role": "user", "content": "按照先后顺序,重复一下我原先问你的所有问题和你给我的相应回答。"}
        ]
    }

    headers = {
        'Authorization': f"Bearer {KEY}",
        'Content-Type': 'application/json'
    }

    response = requests.post(
        f'{PROXY_URL}/v1/chat/completions', data=json.dumps(data), headers=headers)

    if response.status_code == 200:
        response_json = response.json()
        choices = response_json.get('choices')
        if choices:
            result = response_json['choices'][0]['message']['content']
    else:
        print("chatgpt调用出现问题,状态码是:", response.status_code)
        answer = ""

    return answer, data["messages"]

让我们近距离观察一下传入请求中的data的message属性的值

{"role": "system", "content": "You are a helpful calendar assistant"},
{"role": "user", "content": "六月的下一个月有多少天?"},
{"role": "assistant", "content": "七月有31天。"},
{"role": "user", "content": "八月有多少天?"},
{"role": "assistant", "content": "八月也有31天。"},
{"role": "user", "content": "二月有多少天?"},
{"role": "assistant", "content": "这要分平年和闰年"},
{"role": "user", "content": "按照先后顺序,重复一下我原先问你的所有问题和你给我的相应回答。"}

该值是一个list,一共包含1+6+1个字典,如果不加上历史记录,通常就只是一头一尾两个字典,而中间出现了3组user-assistant字典,就代表着之前的三轮对话。

我调用上述方法的代码如下

def main():
    # 输入问题和上下文
    question = "按照先后顺序,重复一下我原先问你的所有问题,再进行总结。"

    # 调用方法进行对话生成
    response, message_list = generate_chat_completion(
        question)

    # 打印生成的回答
    current_folder = os.getcwd()
    write_file = os.path.join(
        current_folder, "codex_api", "chatgpt_answer7.txt")
    os.makedirs(os.path.dirname(write_file), exist_ok=True)
    with open(write_file, "w", encoding="utf-8") as f:
        for message in message_list:
            f.write(message["role"]+"\n"+message["content"])
            f.write("\n")
            f.write("\n"+'-'*100+"\n")
        f.write(response)

保存下来的文件内容如下

当然,让我来回顾一下之前的问题和回答:

1. 六月的下一个月有多少天?
   回答:七月有31天。

2. 八月有多少天?
   回答:八月也有31天。

3. 二月有多少天?
   回答:二月的天数取决于是平年还是闰年。在平年中,二月有28天;而在闰年中,二月有29天。

请问还有其他问题需要我回答吗

本文标签: 聊天记录潜能解锁chatGPTAPI