Views
No views yet
quickmt-en-hi Neural Machine Translation Modelquickmt-en-hi is a reasonably fast and reasonably accurate neural machine translation model for translation from en into hi.eoleeole model configuration in this repository for further details and the eole-model for the raw eole (pytorch) model.quickmtquickmt python library.1git clone https://github.com/quickmt/quickmt.git
2pip install ./quickmt/1from quickmt import Translator
2from huggingface_hub import snapshot_download
3
4# Download Model (if not downloaded already) and return path to local model
5# Device is either 'auto', 'cpu' or 'cuda'
6t = Translator(
7 snapshot_download("quickmt/quickmt-en-hi", ignore_patterns="eole-model/*"),
8 device="cpu"
9)
10
11# Translate - set beam size to 5 for higher quality (but slower speed)
12sample_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.'
13t(sample_text, beam_size=5)
14
15> 'हैलिफ़ैक्स, नोवा स्कोटिया में डलहौजी विश्वविद्यालय में चिकित्सा के प्रोफेसर और कनाडाई मधुमेह संघ के नैदानिक और वैज्ञानिक विभाजन के अध्यक्ष डॉ एहुद उर ने चेतावनी दी कि अनुसंधान अभी भी अपने शुरुआती दिनों में है।'
16
17# Get alternative translations by sampling
18# You can pass any cTranslate2 `translate_batch` arguments
19t([sample_text], sampling_temperature=1.2, beam_size=1, sampling_topk=50, sampling_topp=0.9)
20
21> 'हैलिफ़ैक्स, नोवा स्कोटिया में डलहौजी विश्वविद्यालय में चिकित्सा के प्रोफेसर और कनाडाई मधुमेह संघ के नैदानिक और वैज्ञानिक डिवीजन के अध्यक्ष डॉ। एहुद उर ने आगाह किया कि अनुसंधान अभी भी अपने प्रारंभिक दिनों में है।'
22ctranslate2 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.bleu and chrf2 are calculated with sacrebleu on the Flores200 devtest test set ("eng_Latn"->"hin_Deva"). 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 using a large batch size).| bleu | chrf2 | comet22 | Time (s) | |
|---|---|---|---|---|
| quickmt/quickmt-en-hi | 35.9 | 59.94 | 79.04 | 1.22 |
| Helsink-NLP/opus-mt-en-hi | 13.53 | 35.51 | 60.13 | 3.79 |
| facebook/nllb-200-distilled-600M | 31.13 | 56.66 | 80.26 | 27.08 |
| facebook/nllb-200-distilled-1.3B | 32.43 | 57.64 | 81.06 | 45.79 |
| facebook/m2m100_418M | 26.17 | 51.24 | 74.65 | 27.14 |
| facebook/m2m100_1.2B | 26.57 | 51.88 | 76.09 | 49.76 |
quickmt-en-hi is the fastest, considerably higher quality than opus-mt-en-hi, m2m100_418M and m2m100_1.2B and nearly as high quality as the two nllb models (which have a non-commercial license).