You can deploy this model and start earning money today!

michaelfeil/
ct2fast-opus-mt-en-fr

Name: ct2fast-opus-mt-en-fr
Availability: InStock

Views

No views yet

# Fast-Inference with Ctranslate2

Speedup inference by 2x-8x using int8 inference in C++

quantized version of Helsinki-NLP/opus-mt-en-fr

pip install hf-hub-ctranslate2>=1.0.0 ctranslate2>=3.13.0

Converted using

ct2-transformers-converter --model Helsinki-NLP/opus-mt-en-fr --output_dir /home/michael/tmp-ct2fast-opus-mt-en-fr --force --copy_files README.md generation_config.json tokenizer_config.json vocab.json source.spm .gitattributes target.spm --quantization float16

Checkpoint compatible to ctranslate2 and hf-hub-ctranslate2

compute_type=int8_float16 for device="cuda"
compute_type=int8 for device="cpu"

python

from hf_hub_ctranslate2 import TranslatorCT2fromHfHub, GeneratorCT2fromHfHub
from transformers import AutoTokenizer

model_name = "michaelfeil/ct2fast-opus-mt-en-fr"
# use either TranslatorCT2fromHfHub or GeneratorCT2fromHfHub here, depending on model.
model = TranslatorCT2fromHfHub(
        # load in int8 on CUDA
        model_name_or_path=model_name, 
        device="cuda",
        compute_type="int8_float16",
        tokenizer=AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-en-fr")
)
outputs = model.generate(
    text=["How do you call a fast Flan-ingo?", "User: How are you doing?"],
)
print(outputs)

Licence and other remarks:

This is just a quantized version. Licence conditions are intended to be idential to original huggingface repo.

Original description

opus-mt-en-fr

source languages: en
target languages: fr
OPUS readme: en-fr
dataset: opus
model: transformer-align
pre-processing: normalization + SentencePiece
download original weights: opus-2020-02-26.zip
test set translations: opus-2020-02-26.test.txt
test set scores: opus-2020-02-26.eval.txt

Benchmarks

testset	BLEU	chr-F
newsdiscussdev2015-enfr.en.fr	33.8	0.602
newsdiscusstest2015-enfr.en.fr	40.0	0.643
newssyscomb2009.en.fr	29.8	0.584
news-test2008.en.fr	27.5	0.554
newstest2009.en.fr	29.4	0.577
newstest2010.en.fr	32.7	0.596
newstest2011.en.fr	34.3	0.611
newstest2012.en.fr	31.8	0.592
newstest2013.en.fr	33.2	0.589
Tatoeba.en.fr	50.5	0.672