quickmt-hi-en Neural Machine Translation Model

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

Model Information

• Trained using eole
• 185M parameter transformer 'big' with 8 encoder layers and 2 decoder layers
• 50k joint Sentencepiece vocabulary
• Exported for fast inference to CTranslate2 format
• Training data: https://huggingface.co/datasets/quickmt/quickmt-train.hi-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:

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

quickmt-model-download quickmt/quickmt-hi-en ./quickmt-hi-en

Finally use the model in python:

from quickmt import Translator

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

# Translate - set beam size to 1 for faster speed (but lower quality)
sample_text = 'डॉ. एहुड उर, नोवा स्कोटिया के हैलिफ़ैक्स में डलहौज़ी विश्वविद्यालय में चिकित्सा के प्रोफ़ेसर और कनाडाई डायबिटीज़ एसोसिएशन के नैदानिक \u200b\u200bऔर वैज्ञानिक विभाग के अध्यक्ष ने आगाह किया कि शोध अभी भी अपने शुरुआती दिनों में है.'
t(sample_text, beam_size=5)

> 'On the other hand, Dr Ehud ur, professor of medicine at Dalhousie University in Halifax, Nova Scotia and chairman of the clinical and scientific department of the Canadian Diabetes Association, cautioned that the research is still in its early days.'

# 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)

> 'A group of young men Ehud ur, professor of medicine at Dalhousie University in Halifax, Nova Scotia, and president of the Diagnostics and Scientific Department of the Canadian Diabetes Association warned that the research is still in its early days.'

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 which also uses ctranslate2 and sentencepiece.

Metrics

bleu and chrf2 are calculated with sacrebleu on the Flores200 devtest test set ("hin_Deva"->"eng_Latn"). comet22 with the comet library and the default model. "Time (s)" is the time in seconds to translate the flores-devtest dataset (1012 sentences) on an RTX 4070s GPU with batch size 32 (faster speed is possible using a larger batch size).

it -> en flores-devtest metrics

model	bleu	chrf2	comet22	Time (s)
quickmt/quickmt-hi-en	39.9	65.04	88.77	1.14
Helsink-NLP/opus-mt-hi-en	18.83	45.9	75.81	4.38
facebook/nllb-200-distilled-600M	38.8	64.29	88.87	22.49
facebook/nllb-200-distilled-1.3B	41.71	66.67	89.69	38.91
facebook/m2m100_418M	29.81	57.66	85	19.65
facebook/m2m100_1.2B	33.79	60.21	86.3	38.33

quickmt-hi-en is the fastest and second highest quality next to nllb-200-distilled-1.3B (which has a non-commercial license).