opus-mt-en-zh

opus-mt-en-zh

Translates English text into Chinese, covering Mandarin and several other Chinese varieties, with a compact transformer model you run yourself. For developers and teams who need local English to Chinese translation.

Made by Helsinki-NLP on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: opus-mt-en-zh is an English to Chinese translation model from Helsinki-NLP, part of the OPUS machine translation work. It is a transformer trained on the Tatoeba Challenge data. One model covers many Chinese varieties, so you tell it which one you want with a short tag at the start of each sentence.

Who it suits: It suits developers and teams who want English to Chinese translation under their own control, and researchers who need a plain baseline to compare against.

Industry | Who uses it | How they use it | Example
Software localization | Product teams | Draft Chinese strings for an app from English text | Translating interface labels
Publishing | Editors | Get a rough Chinese version of English copy to review | A first pass on an article
Education | Language learners | Compare an English sentence with a Chinese rendering | Checking how a phrase is translated
Research | Machine translation researchers | Use a known baseline for English to Chinese | Comparing BLEU against a new model

Skip it if you need any other language pair, since this one only goes from English to Chinese.

Where to use it:

• A Python script or notebook with the transformers library
• Windows, macOS or Linux, running on your own machine
• A backend service that translates English text into Chinese

Use case | Who | What they get back
Sentence translation | Developers | A Chinese version of each English sentence
Variety selection | Localization teams | Output in Mandarin or another listed Chinese variety, chosen by tag
Baseline testing | Researchers | A reference score to compare other systems against

How to set it up: You need these first.

• Python with pip
• The transformers library, which the model is listed for
• A list of the target tags you plan to use, taken from the card

1. Install the Hugging Face hub library with pip.
2. Download the model files with the hf command.
3. Load the model with the transformers library, following the model card.

pip install -U huggingface_hub
hf download Helsinki-NLP/opus-mt-en-zh

How to use it: Put a target tag in front of every English sentence you send, in the form of two angle brackets around an ID, written as >>id<<. The ID must be one of the thirteen the card lists, such as cmn_Hans, cmn_Hant, yue or wuu.

hf download Helsinki-NLP/opus-mt-en-zh

Tool | What it does
>>id<< | Sentence-start tag that chooses the Chinese variety to translate into
hf download | Fetches the model files to your computer

Example: Send the text ">>cmn_Hans<< How are you today?" to the model. It returns the sentence as Chinese text. If you swap the tag for another ID from the card, the output changes to that variety.

For newcomers and experts: If you are new, the one thing to remember is the tag at the start of each sentence. Without it the model does not know which Chinese to produce. A first small try is one short sentence with the cmn_Hans tag.

If you know the field, this is a one-direction transformer with SentencePiece at 32k and text normalization. It is not multilingual on the source side, which is English only. The card reports a BLEU of 31.4 and a chr-F of 0.268 on Tatoeba-test.eng.zho, with a brevity penalty near 0.9, so outputs run a little shorter than the references.

English sentence
Add the >>id<< target tag
Normalize and split with SentencePiece
Transformer translates
Chinese text in the chosen variety

Good to know:

• Direction: English to Chinese only
• Targets: thirteen Chinese variety IDs, including Mandarin, Cantonese and Wu
• Required tag: every sentence needs the target tag at its start
• Benchmark: BLEU 31.4 and chr-F 0.268 on Tatoeba-test.eng.zho
• Other builds: other people have made 7 quantized and 35 fine-tuned versions

• License: Apache 2.0
• Cost to run: none named, the source lists no API key, account or paid service
• Data it reaches: the text you pass in and the model files on your disk, and the source names no web or account access
• Not for: other language pairs or translating without a target tag

On Hugging Face, 13 November 2026

• Task: translation
• Library: transformers
• Downloads in the last 30 days: 69,966
• Likes: 411
• License: apache-2.0
• Languages: en, zh
• Quantized versions: 7
• Fine-tunes: 35
• Tags: transformers, pytorch, rust, marian, text2text-generation, translation

Links

• Model: https://huggingface.co/Helsinki-NLP/opus-mt-en-zh
• Files: https://huggingface.co/Helsinki-NLP/opus-mt-en-zh/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download Helsinki-NLP/opus-mt-en-zh

Then follow the usage in its model card. Nothing in this kit runs by itself.