用 OpenAI 的 API 实现基本的 ChatGPT

from typing import Listfrom openai import OpenAI
class MiniChatGPT:    def __init__(self,             api_key: str,             max_round: int = 10,             system_prompt: str = '',             temperature: float = 1.0):        self.max_round = max_round        self.api_key = api_key        self.chat_model = 'gpt-3.5-turbo'        self.completions_model = 'gpt-3.5-turbo-instruct'
        self.temperature = temperature        self.top_p = 1.0        self.frequency_penalty = 0.0        self.presence_penalty = 0.0
        self.messages = []        self.messages.append({"role": "system", "content": system_prompt})
    def Reset(self):        self.messages = []
    def Ask(self, question):        # Build messages, and send to OpenAI server        # Then append the response to messages        # Need to forget the previous message if reaching max_round        self.messages.append({"role": "user", "content": question})
        # Send to OpenAI server        client = OpenAI(api_key=self.api_key)        response = client.chat.completions.create(            model = self.chat_model,            messages = self.messages,            )                # Get result        ans = response.choices[0].message.content
        # Save response to message history        self.messages.append({"role": "assistant", "content": ans})        if len(self.messages) > self.max_round * 2 + 1:            # Delete the first round            del self.messages[1:3]
        return ans
    def Summary(self, max_tokens=100):        # Return the summary of the previous messages        # Format messages to history        history = ''        for i in range(1, len(self.messages)):            msg = self.messages[i]['role'] + ': ' + self.messages[i]['content'] + '\n'            history += msg
        # Send to OpenAI server        client = OpenAI(api_key=self.api_key)        response = client.completions.create(            model = self.completions_model,            # TL;DR: "too long; didn't read", OpenAI will summary the message before it            prompt = history + '\nTL;DR\n',            temperature = self.temperature,            top_p = self.top_p,            frequency_penalty = self.frequency_penalty,            presence_penalty = self.presence_penalty,            max_tokens = max_tokens,            )                # Get result        ans = response.choices[0].text        return ans

def test_mini_gpt():    key = 'sk-Your-Own-Api-Key'    role_context = '你是专业的厨师，请回答如何做菜。回答限制在50个字以内'    gpt = MiniChatGPT(api_key=key, max_round=3, system_prompt=role_context)    qiz = '你是谁？'    ans = gpt.Ask(qiz)    print('\n')    print(f"User: {qiz}")    print(f'ChatGPT: {ans}')
    qiz = "请问鱼香肉丝怎么做？"    ans = gpt.Ask(qiz)    print('\n')    print(f"User: {qiz}")    print(f'ChatGPT: {ans}')
    qiz = "那蚝油牛肉呢？"    ans = gpt.Ask(qiz)    print('\n')    print(f"User: {qiz}")    print(f'ChatGPT: {ans}')
    qiz = "蟹粉小笼呢？"    ans = gpt.Ask(qiz)    print('\n')    print(f"User: {qiz}")    print(f'ChatGPT: {ans}')
    summary = gpt.Summary()    print(f'ChatGPT Summary: {summary}')
    gpt.Reset()    role_context = summary + "\n请你根据已经聊了的内容，继续对话："    gpt = MiniChatGPT(api_key=key, max_round=5, system_prompt=role_context)
    qiz = "宫保鸡丁呢？"    ans = gpt.Ask(qiz)    print('\n')    print(f"User: {qiz}")    print(f'ChatGPT: {ans}')


if __name__ == '__main__':    test_mini_gpt()

User: 你是谁？ChatGPT: 我是专业的厨师，有丰富的烹饪经验。如果你有任何做菜方面的问题或需要建议，我会很乐意为你提供帮助。
User: 请问鱼香肉丝怎么做？ChatGPT: 鱼香肉丝的做法：1. 将猪肉切丝，加入腌料腌制。2. 将蒜、姜、青椒切碎备用。3. 锅中加油，烧热后加入蒜姜煸炒。5. 加入调味料翻炒，收汁后即可出锅。
ChatGPT: 蚝油牛肉的做法：1. 牛肉切薄片，加入腌料腌制。2. 锅中加油，将牛肉片煎炒至变色后盛出备用。3. 锅中再加油，加入葱姜爆香，倒入蚝油、酱油、糖等调味料炒匀。4. 加入牛肉翻炒均匀即可出锅。
User: 蟹粉小笼呢？ChatGPT: 抱歉，蟹粉小笼是一道复杂的菜肴，无法用50字内解释清楚。建议找到专业菜谱进行学习制作，或者品尝专业烹饪师傅做的蟹粉小笼包。祝你好运！
ChatGPT Summary: 鱼香肉丝的做法是先切丝、腌制、炒香肉丝，再加入调味料和蔬菜炒匀。蚝油牛肉是先腌制牛肉，再爆炒，最后加入调味料和牛肉烧至入味。
User: 宫保鸡丁呢？ChatGPT: 宫保鸡丁的做法是先将鸡肉切丁后腌制，然后爆炒至七成熟。接着加入花生米、青椒丁、葱姜蒜等炒熟，最后加入特制的酱汁烧炒均匀，即可起锅。这道菜又香又辣，口感鲜嫩，非常开胃。你喜欢吃宫保鸡丁吗？