Skip to content

[Bug] Lobe i18n在使用模型 gpt-40|gpt-4o-mini 翻译markdown文件时,会计算出大量token并翻译出错误的文档 #110

@tianlelyd

Description

@tianlelyd

💻 系统环境 | Operating System

Windows

🌐 CLI 版本 | CLI Version

1.18.1

🐛 问题描述 | Bug Description

在翻译markdown时,将模型修改为gpt-40|gpt-4o-mini , 一篇五千字的md文档,计算出要处理81万token,处理完成后消耗了大量token,并且翻译后生成的文档中含有大量以下文内容:

It seems like your message is incomplete. Could you
please provide the content ot the markdown Tile
that you would like to nave translated Trom chinese
Z-N) TO Engish en-us)

image

我的配置文件:

module.exports = {
	modelName: 'gpt-4o-mini',
	markdown: {
		entry: ['README.zh-CN.md'],
		entryLocale: 'zh-CN',
		entryExtension: '.zh-CN.md',
		outputLocales: ['en-US'],
		outputExtensions: (locale, { getDefaultExtension }) => {
			if (locale === 'en-US') return '.md';
			return getDefaultExtension(locale);
		},
	},
	experimental: {
		jsonMode: true,
	},
};

经过测试无论experimental中的配置是否开启,都会存在该问题,但是使用gpt3.5或者gpt4时可正常完成翻译!

🚦 期望结果 | Expected Behavior

  • token计算正确
  • markdown内容正常翻译

📷 复现步骤 | Recurrence Steps

我使用的是中文的markdown内容,文字大概5000字,我实际测试,不管使用什么markdown内容,都会复现该问题!

📝 补充信息 | Additional Information

No response

Metadata

Metadata

Assignees

No one assigned

    Labels

    InactiveNo response in 30 days | 超过 30 天未活跃🐛 BugSomething isn't working | 缺陷

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions