I would like to perform Machine Translation in tensorflow.js, in the browser. The issue is that state-of-the-art models have many gigabytes (fairseq ensemble), and translation is slow.

Do you know some good small models for Machine Translation? Probably something below the 50 million parameter threshold.

I found some older papers with models below 100 MB for training on small datasets (IWSLT'15 English-Vietnamese attention-based models), but these are probably superseded.

  • 1
    $\begingroup$ There's a 300MB model in OpenNMT, which would perhaps be compressed with ERNIE. The paper states that 5x compression keeps 98% accuracy. $\endgroup$ – MrMartin Sep 4 '19 at 13:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.