translate.py

import pandas as pd
import openai
from utils.openai import load_client

client = load_client('<path-to-your-key-file>')

# 初始化OpenAI API密钥
openai.api_key = 'dE6qTITLy3WCoMPMCPr8tUFBHBaec5wN'

# 指定parquet文件的路径
file_path = '/home/yiyangai/stephenqs/datasets/physics_big/data/combined_images_non_empty.parquet'

# 读取parquet文件
df = pd.read_parquet(file_path)

# 定义翻译函数，调用GPT-4 API
def translate_text(text):
    if isinstance(text, list):  # 如果是列表，递归处理每个元素
        return [translate_text(item) for item in text]
    elif isinstance(text, dict):  # 如果是字典，递归处理字典中的值
        return {key: translate_text(value) for key, value in text.items()}
    elif isinstance(text, str):  # 如果是字符串，翻译
        response = openai.ChatCompletion.create(
            model="gpt-4o-mini",
            messages=[
                {"role": "system", "content": "You are a helpful assistant."},
                {"role": "user", "content": f"Translate this Russian text to English: {text}"}
            ]
        )
        return response.choices[0].message['content'].strip()
    else:
        return text

# 初始化一个空的DataFrame来存储翻译后的结果
df_translated = pd.DataFrame(columns=df.columns)

# 每次翻译20条记录
batch_size = 20

# 逐批次翻译数据
for i in range(0, len(df), batch_size):
    print(f"正在翻译样本 {i+1} 到 {min(i+batch_size, len(df))}...")
    batch = df.iloc[i:i+batch_size].copy()
    
    # 对批次内的每一列进行翻译
    for column in batch.columns:
        batch[column] = batch[column].apply(translate_text)
    
    # 将翻译后的批次数据添加到翻译结果的DataFrame中
    df_translated = pd.concat([df_translated, batch], ignore_index=True)

    # 保存中间结果（可选，以防脚本中途失败）
    df_translated.to_parquet('/home/yiyangai/stephenqs/datasets/physics_big/data/partial_translation.parquet')

# 最终保存完整的翻译结果到新的Parquet文件
output_file_path = '/home/yiyangai/stephenqs/datasets/physics_big/data/translated_combined_images_non_empty.parquet'
df_translated.to_parquet(output_file_path)

print(f"翻译完成，已将结果保存到 {output_file_path}")