prompt.py

import json
from huggingface_hub import InferenceClient
import re

from config import MODEL_NAME
from dotenv import load_dotenv
import os

# Load environment variables from .env file
load_dotenv()
api_key = os.getenv("HUGGINGFACE_TOKEN")

# Prompt配置
SYSTEM_PROMPT = "你是一位精通英语和中文的教学专家，专门帮助用户，单词量为8000的语言学习者，将中文翻译成地道的英语表达，并提取重要词汇，比如，形容词，名词和动词，提供解释和用法说明。只用输出json格式的数据，确保json完整性和正确性，不要在开头加任何内容，并且注意json格式，control character。"

PROMPT_TEMPLATE = """请将以下中文翻译成英语，并提取重要词汇，比如，形容词，名词和动词，针对每个单词，提供解释和简单的造句说明：
    中文文本：
    <text>
    {text}
    <text>
    
    请按照以下JSON格式返回：
    {{
        "english": "英语翻译",
        "important_words": [
            {{
                "word_en": "重要英语单词或短语",
                "meaning_ch": "中文含义",
                "usage": "用法说明"
            }}
        ]
    }}
"""

USER_SHOT_1 = "我昨天去公园散步,看到很多人在那里锻炼身体"
SHOT_1_RESPONSE = """
{
  "english": "I went for a walk in the park yesterday and saw many people exercising there.",
  "important_words": [
    {
      "word_en": "exercise",
      "meaning_ch": "锻炼身体",
      "usage": "e.g. I go to the gym to exercise every day."
    },
    {
      "word_en": "park",
      "meaning_ch": "公园",
      "usage": "e.g. I took a walk in the park on Sunday."
    },
    {
      "word_en": "scatter",
      "meaning_ch": "散步",
      "usage": "e.g. I like to scatter around the city to explore new places."
    }
  ]
}
"""

USER_SHOT_2 = "我昨天吃坏肚子了，一直在拉肚子。"
SHOT_2_RESPONSE = """{
  "english": "I had a stomachache yesterday and have been experiencing diarrhea ever since.",
  "important_words": [
    {
      "word_en": "stomachache",
      "meaning_ch": "肚子痛",
      "usage": "Example sentence: I woke up with a stomachache and couldn't eat anything."
    },
    {
      "word_en": "diarrhea",
      "meaning_ch": "腹泻",
      "usage": "Example sentence: She had diarrhea after eating spoiled food."
    }
  ]
}
"""


def message_builder(text):
    """构建消息，包含few-shot示例"""
    messages = [
        {
            "role": "system",
            "content": SYSTEM_PROMPT
        },
        {
            "role": "user",
            "content": PROMPT_TEMPLATE.format(text=USER_SHOT_1)
        },
        {
            "role": "assistant",
            "content": SHOT_1_RESPONSE
        },
        {
            "role": "user",
            "content": PROMPT_TEMPLATE.format(text=USER_SHOT_2)
        },
        {
            "role": "assistant",
            "content": SHOT_2_RESPONSE
        },
        {
            "role": "user",
            "content": PROMPT_TEMPLATE.format(text=text)
        }
    ]
    return messages


client = InferenceClient(api_key=api_key)


def get_llm_response(text) -> str:
    """调用LLM获取响应"""

    print("get_llm_response")

    messages = message_builder(text)

    llm_response = client.chat.completions.create(
        model=MODEL_NAME, messages=messages, max_tokens=1024, temperature=0.3)

    return llm_response.choices[0].message.content


# main
if __name__ == "__main__":
    response = get_llm_response("我昨天去公园散步,看到很多人在那里锻炼身体")
    print(response)

    # import pprint
    # pprint.pprint(response)