NLLB Translator

NLLB Translator

Translates text between 200 languages with a choice of engines, from open models running in the Space to cloud models you reach with your own key. For writers, students and developers.

Made by Asanari on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: NLLB Translator is a web app that translates text across 200 languages and lets you pick the engine for each request. Several open models run inside the Space with no outside API. Cloud models work too when you supply your own key. The backend is FastAPI and the front end is Next.js.

Who it suits: It suits people who translate often and want to compare engines, such as writers, students and researchers. It also suits developers who want a self-hosted translator with a clean interface and room to add engines.

Industry | Who uses it | How they use it | Example
Writing | Authors and editors | Compare machine translations of a passage | Run one paragraph through two engines
Education | Language students | Check meaning in a language they are learning | Translate a sentence and read it back
Software | Developers | Host a translator on their own server | Run the Docker image and add an engine
Research | Linguists | Test how engines treat less common languages | Switch engines mid-session on the same text

Anyone who needs a certified legal translation can skip it, since these are machine engines.

Where to use it: The app runs in three places.

• A web browser, on the page of the Space at Hugging Face
• Windows, macOS or Linux, as a local copy built with Docker
• Your own server, using the Dockerfile and compose file in the repository

Use case | Who | What they get back
Quick translation | Anyone | Translated text with live output as it arrives
Engine comparison | Researchers | The same text from different local and cloud engines
Self-hosting | Developers | A translator that keeps local engines on your server

How to set it up: The online page needs nothing installed. A local copy needs the following.

• Docker
• Python with pip, to install the Hugging Face hub package
• Your own keys, only if you want cloud engines

1. Install the Hugging Face hub package and download a copy of the Space.
2. Move into the new folder and build the Docker image.
3. Run the container with port 7860 published.
4. Optionally set keys as server settings, or paste them in the app's API keys panel.

pip install -U huggingface_hub
hf download Asanari/nllb-translator --repo-type space --local-dir nllb-translator
cd nllb-translator
docker build -t nllb-translator .
docker run -it -p 7860:7860 nllb-translator

How to use it: Open the app, pick a source and target language, choose an engine, and type or paste your text. Translation starts as you type, and the output streams in. Swap, copy and example phrases are one click away.

Tool | What it does
NLLB-200 600M | Runs on the server with int8 CPU inference across 200 languages
Hy-MT2 1.8B | Tencent's translation-focused model, run on the server
TranslateGemma 4B | A Google model that runs in the Space through Ollama and is slower on free CPU
Qwen3.5 2B and Qwen3 1.7B | Small Qwen models run in the Space through Ollama
Gemini, Groq and OpenRouter | Cloud engines you reach with your own key
GPT-6 Luna | A paid OpenAI engine that needs an OpenAI key

Example: You pick Korean as the target, choose the Hy-MT2 engine, and paste a short English paragraph. The translation streams into the output pane as you type, and you can switch to a cloud engine to compare.

For newcomers and experts: If you are new to this, leave the default engine and translate one short sentence. Then switch the engine and read how the wording changes.

If you know the field, five local engines translate on the server and cloud engines are labelled with where your text goes. LLM output streams token by token, NLLB shows each finished line, and a new edit cancels the older stream. Bounded caches on the server and in the browser reuse completed translations. If a Groq model is retired, the backend finds the newest model of the same family.

Pick languages and an engine
Type or paste your text
Local or cloud engine translates
Output streams into the page

Good to know: Each engine reports whether it is ready, and the interface offers only the ready ones. Keys pasted in the browser stay in that browser and are passed per request, so the server never stores them.

• License: MIT
• Cost to run: local engines need no key, while Gemini, Groq, OpenRouter and OpenAI engines need your own key, and GPT-6 Luna is paid
• Data it reaches: your text goes to the cloud provider only when you pick a cloud engine, and local engines translate on the server
• Limits: TranslateGemma is slower on the free CPU

On Hugging Face, 13 November 2026

• Likes: 3
• Built with: docker
• License: mit

Links

• Space: https://huggingface.co/spaces/Asanari/nllb-translator
• Files: https://huggingface.co/spaces/Asanari/nllb-translator/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.