quickmt-it-en Neural Machine Translation Model

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

Finally use the model in python:

from quickmt import Translator

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

# Translate - set beam size to 1 for faster speed (but lower quality)
sample_text = "Lo studio è ancora in fase iniziale, come dichiarato cautelativamente dal dottor Ehud Ur, professore di medicina alla Dalhousie University di Halifax, Nuova Scozia, e direttore del dipartimento clinico e scientifico della Canadian Diabetes Association."
t(sample_text, beam_size=5)

> 'The study is still in its early stages, as cautiously stated by Dr. Ehud Ur, professor of medicine at Dalhousie University in Halifax, Nova Scotia, and director of the clinical and scientific department of the Canadian Diabetes Association.'

# 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)

> 'The study is still in the early stages, as indicated cautiously by Dr. Ehud Ur, Medical professor at Dalhousie University of Halifax, Nova Scotia, and director of clinical and scientific department at the Canadian Diabetes Association.'

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 ("ita_Latn"->"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

	bleu	chrf2	comet22	Time (s)
quickmt/quickmt-it-en	32.1	61.48	87.31	1.04
Helsink-NLP/opus-mt-it-en	29.48	59.96	86.59	4.39
facebook/nllb-200-distilled-600M	33.49	61.97	87.39	22.16
facebook/nllb-200-distilled-1.3B	34.97	63.23	88.14	38.26
facebook/m2m100_418M	25.92	56.94	83.14	18.61
facebook/m2m100_1.2B	30.81	60.43	86.43	36.28

quickmt-it-en is the fastest and is higher quality than opus-mt-it-en, m2m100_418m and m2m100_1.2B.