.ipynb_checkpoints/README-checkpoint.md

metadata

language:
  - en
  - el
tags:
  - translation
license: cc-by-4.0
datasets:
  - quickmt/quickmt-train.el-en
model-index:
  - name: quickmt-en-el
    results:
      - task:
          name: Translation eng-ell
          type: translation
          args: eng-ell
        dataset:
          name: flores101-devtest
          type: flores_101
          args: eng_Latn ell_Grek devtest
        metrics:
          - name: BLEU
            type: bleu
            value: 28.86
          - name: CHRF
            type: chrf
            value: 54.94
          - name: COMET
            type: comet
            value: 88.87

quickmt-en-el Neural Machine Translation Model

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

Model Information

• Trained using eole
• 195M parameter transformer 'big' with 8 encoder layers and 2 decoder layers
• 20k sentencepiece vocabularies
• Exported for fast inference to CTranslate2 format
• Training data: https://huggingface.co/datasets/quickmt/quickmt-train.el-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-en-el ./quickmt-en-el

Finally use the model in python:

from quickmt import Translator

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

# Translate - set beam size to 5 for higher quality (but slower speed)
sample_text = 'Dr. Ehud Ur, professor of medicine at Dalhousie University in Halifax, Nova Scotia and chair of the clinical and scientific division of the Canadian Diabetes Association cautioned that the research is still in its early days.'
t(sample_text, beam_size=5)

'Ο Δρ Ehud Ur, καθηγητής ιατρικής στο Πανεπιστήμιο Dalhousie στο Halifax της Νέας Σκωτίας και πρόεδρος του κλινικού και επιστημονικού τμήματος της Καναδικής Διαβητολογικής Εταιρείας, προειδοποίησε ότι η έρευνα βρίσκεται ακόμα στις πρώτες ημέρες της.'

# 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)

'Ο Δρ. Ehud Ur, καθηγητής Ιατρικής στο Πανεπιστήμιο Dalhousie του Halifax της Νέας Σκωτίας και πρόεδρος του κλινικού και επιστημονικού τμήματος της Καναδικής Διαβητικής Εταιρείας, προειδοποίησε ότι η έρευνα παραμένει ακόμη σε πρώιμο στάδιο.'

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 ("eng_Latn"->"ell_Grek"). comet22 with the comet library and the default model. "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 with a larger batch size).

	bleu	chrf2	comet22	Time (s)
quickmt/quickmt-en-el	28.86	54.94	88.87	1.48
Helsinki-NLP/opus-mt-en-el	25	52.16	86.9	4.63
facebook/nllb-200-distilled-600M	24.98	52.32	88.15	33.38
facebook/nllb-200-distilled-1.3B	27.36	54.43	89.35	58.82
facebook/m2m100_418M	21.98	49.09	84.37	27.13
facebook/m2m100_1.2B	26.59	53.09	88.18	52.98