---
language:
- en
- ko
tags:
- translation
license: cc-by-4.0
datasets:
- quickmt/quickmt-train.ko-en
model-index:
- name: quickmt-ko-en
  results:
  - task:
      name: Translation kor-eng
      type: translation
      args: kor-eng
    dataset:
      name: flores101-devtest
      type: flores_101
      args: kor_Hang eng_Latn devtest
    metrics:
    - name: CHRF
      type: chrf
      value: 56.25
    - name: BLEU
      type: bleu
      value: 27.03
    - name: COMET
      type: comet
      value: 86.11 
---


# `quickmt-ko-en` Neural Machine Translation Model 

`quickmt-ko-en` is a reasonably fast and reasonably accurate neural machine translation model for translation from `ko` into `en`.


## Model Information

* Trained using [`eole`](https://github.com/eole-nlp/eole)
* 185M parameter transformer 'big' with 8 encoder layers and 2 decoder layers
* 20k sentencepiece vocabularies
* Exported for fast inference to [CTranslate2](https://github.com/OpenNMT/CTranslate2) format
* Training data: https://huggingface.co/datasets/quickmt/quickmt-train.ko-en/tree/main

See the `eole` model configuration in this repository for further details and the `eole-model` for the raw `eole` (pytorch) model.


## Usage with `quickmt`

You must install the Nvidia cuda toolkit first, if you want to do GPU inference.

Next, install the `quickmt` python library and download the model:

```bash
git clone https://github.com/quickmt/quickmt.git
pip install ./quickmt/

quickmt-model-download quickmt/quickmt-ko-en ./quickmt-ko-en
```

Finally use the model in python:

```python
from quickmt import Translator

# Auto-detects GPU, set to "cpu" to force CPU inference
t = Translator("./quickmt-ko-en/", device="auto")

# Translate - set beam size to 5 for higher quality (but slower speed)
sample_text = '노바스코샤주 핼리팩스의 댈하우지대학교 의과 교수이자 캐나다 당뇨 협회 임상과학부 의장인 Ehud Ur 박사는 이 연구가 아직 초기 단계라고 경고했습니다.'
t(sample_text, beam_size=5)

> 'Dr. Ehud Ur, a medical professor at Dalhousie University in Halifax, Nova Scotia and chair of the Canadian Diabetes Association Clinical Sciences Department, warned that the study is still in its early stages.'

# Get alternative translations by sampling
# You can pass any cTranslate2 `translate_batch` arguments
t([sample_text], sampling_temperature=1.2, beam_size=1, sampling_topk=50, sampling_topp=0.9)

> 'Dr. Ehud Ur, professor of medicine and professor of medicine from the Dalhowes Institute and chair of the Canadian Diabetes Association Clinical Science Department in Halifax, Nova Scotia, warned the study is still an early step forward.'
```

The model is in `ctranslate2` format, and the tokenizers are `sentencepiece`, so you can use `ctranslate2` directly instead of through `quickmt`. It is also possible  to get this model to work with e.g. [LibreTranslate](https://libretranslate.com/) which also uses `ctranslate2` and `sentencepiece`.


## Metrics

`bleu` and `chrf2` are calculated with [sacrebleu](https://github.com/mjpost/sacrebleu) on the [Flores200 `devtest` test set](https://huggingface.co/datasets/facebook/flores) ("kor_Hang"->"eng_Latn"). `comet22` with the [`comet`](https://github.com/Unbabel/COMET) library and the [default model](https://huggingface.co/Unbabel/wmt22-comet-da). "Time (s)" is the time in seconds to translate (using `ctranslate2`) the flores-devtest dataset (1012 sentences) on an RTX 4070s GPU with batch size 32 (faster speed is possible using a large batch size).

|                                  |   bleu |   chrf2 |   comet22 |   Time (s) |
|:---------------------------------|-------:|--------:|----------:|-----------:|
| quickmt/quickmt-ko-en            |  27.03 |   56.25 |     86.11 |       1.05 |
| Helsink-NLP/opus-mt-ko-en        |  20.78 |   50.39 |     83.06 |       3.62 |
| facebook/nllb-200-distilled-600M |  26.53 |   55.04 |     85.83 |      21.28 |
| facebook/nllb-200-distilled-1.3B |  29.61 |   57.58 |     87.24 |      37.42 |
| facebook/m2m100_418M             |  20.75 |   50.65 |     82.07 |      18.21 |
| facebook/m2m100_1.2B             |  24.59 |   54.17 |     85.15 |      34.82 |

`quickmt-ko-en` is the fastest and is higher quality than `opus-mt-ko-en`, `m2m100_418m`, `m2m100_1.2B` and `nllb-200-distilled-600M` but lower quality than `nllb-200-distilled-1.3B`.