使用qwen-mt翻译文章

qwen_mt_translate.py

import os
from openai import OpenAI

STR_CHINESE = "Chinese"
STR_ENGLISH = "English"
MODEL_QWEN_MT_TURBO = "qwen-mt-turbo"
MODEL_QWEN_MT_PLUS = "qwen-mt-plus"


def detect_language(text) -> str:
    # 去除空格和换行符
    text = text.strip()

    if not text:
        # "Empty string"
        return STR_ENGLISH

    # 统计中英文字符数量
    chinese_count = 0
    english_count = 0

    for char in text:
        if "\u4e00" <= char <= "\u9fff":
            chinese_count += 1
        elif char.isalpha():
            english_count += 1

    # 判断主要语言
    if chinese_count > 0 and chinese_count > english_count:
        return STR_CHINESE
    elif english_count > 0 and english_count > chinese_count:
        return STR_ENGLISH
    else:
        # "Mixed or Undetermined"
        return STR_ENGLISH


def translate(
    dashscope_api_key: str, model: str, source_lang: str, target_lang: str, content: str
) -> str:
    """
    使用阿里云百炼翻译服务进行翻译
    """
    client = OpenAI(
        # 若没有配置环境变量，请用阿里云百炼API Key将下行替换为：api_key="sk-xxx",
        api_key=dashscope_api_key,
        base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
    )
    messages = [
        {
            "role": "user",
            "content": content,
        }
    ]

    translation_options = {"source_lang": source_lang, "target_lang": target_lang}

    completion = client.chat.completions.create(
        model=model,
        messages=messages,
        extra_body={"translation_options": translation_options},
    )
    return completion.choices[0].message.content


def main(dashscope_api_key: str, quality: int, content: str) -> dict:
    """
    翻译文本
    :param dashscope_api_key: 阿里云百炼API Key
    :param quality: 翻译质量，1为高{QWEN_MT_PLUS}，其它为速度优先{QWEN_MT_TURBO}
    :param content: 待翻译的文本
    """
    source_lang = detect_language(content)
    target_lang = STR_CHINESE if source_lang == STR_ENGLISH else STR_ENGLISH
    model = MODEL_QWEN_MT_PLUS if quality == 1 else MODEL_QWEN_MT_TURBO
    text = translate(dashscope_api_key, model, source_lang, target_lang, content)
    return {"text": text}