Hy-MT2-1.8B Live Translation Demo

A single-page web app that translates English text into eight target languages in parallel through an OpenAI-compatible chat completions endpoint.

The app was built for tencent/Hy-MT2-1.8B, but it works with any endpoint that accepts OpenAI-style /v1/chat/completions requests and returns choices[0].message.content.

Features

• 3x3 responsive translation grid.
• English source input plus eight live output tiles.
• Debounced input, with all translation requests sent concurrently.
• Previous translations stay visible while replacement translations are loading.
• Per-tile loading, ready, and quiet error states.
• Server-side proxy keeps API keys out of the browser.
• No frontend build step and no runtime dependencies.

Target Languages

The default target tiles are:

• Chinese (Simplified)
• Spanish
• Hindi
• Arabic
• French
• Portuguese
• Japanese
• Korean

Set LLM_TRANSLATE_LANGUAGES in .env to change the target tiles without editing code:

LLM_TRANSLATE_LANGUAGES=chinese,spanish,hindi,arabic,french,portuguese,japanese,korean

Entries can use built-in keys, display names, prompt names, or language codes from public/languages.js. The first eight valid entries are used.

For a language that is not in the built-in catalog, use:

LLM_TRANSLATE_LANGUAGES=chinese,custom-key|Display Name|Prompt Name|lang|ltr

Use rtl as the final field for right-to-left languages. Restart the server after changing .env.

Requirements

• Node.js 20 or newer.
• A running OpenAI-compatible chat completions endpoint for Hy-MT2 or another translation-capable model.

Configuration

Create a local .env from the example:

cp .env.example .env

Set either an exact chat completions URL:

LLM_TRANSLATE_ENDPOINT=http://127.0.0.1:8000/v1/chat/completions

or a base URL:

OPENAI_BASE_URL=http://127.0.0.1:8000/v1

Then set the model id or serving alias exposed by your endpoint:

LLM_TRANSLATE_MODEL=tencent/Hy-MT2-1.8B

If your endpoint requires bearer auth:

LLM_TRANSLATE_API_KEY=your_api_key

OPENAI_API_KEY, OPENAI_MODEL, and OPENAI_BASE_URL are also accepted for compatibility.

Environment Variables

Name	Required	Default	Notes
PORT	No	3000	Local server port.
LLM_TRANSLATE_ENDPOINT	One endpoint setting required		Exact /v1/chat/completions URL.
OPENAI_BASE_URL	One endpoint setting required		Base URL; /chat/completions is appended when needed.
LLM_TRANSLATE_API_KEY	No		Bearer token for the translation endpoint.
OPENAI_API_KEY	No		Fallback bearer token name.
LLM_TRANSLATE_MODEL	No	tencent/Hy-MT2-1.8B	Model id or serving alias.
LLM_TRANSLATE_LANGUAGES	No	Built-in eight-language demo set	Comma-separated target language list. Uses the first eight valid entries.
LLM_TRANSLATE_TEMPERATURE	No	0.7	Inference setting from the Hy-MT2 docs.
LLM_TRANSLATE_TOP_P	No	0.6	Inference setting from the Hy-MT2 docs.
LLM_TRANSLATE_TOP_K	No	20	Inference setting from the Hy-MT2 docs.
LLM_TRANSLATE_REPETITION_PENALTY	No	1.05	Inference setting from the Hy-MT2 docs.
LLM_TRANSLATE_MAX_TOKENS	No	4096	Maximum output tokens per request.

Run

npm run dev

Open http://localhost:3000.

For production-style local serving:

npm start

API

POST /api/translate

{
  "sourceText": "Good morning.",
  "targetLanguage": "French"
}

The server sends this prompt to the configured model:

Translate the following text into {target_language}. Note that you should only output the translated result without any additional explanation:

{source_text}

Security

Do not commit .env. It is ignored by default. Only .env.example should be published.

The browser talks to the local server, and the local server talks to the model endpoint. This keeps provider keys and private endpoint details out of client-side code.

License

MIT