search

danielinux7 / anana

Anana group projects to revive the Abkhazian language.
1 stars 0 forks source link

[NMT] Fine tune ab-ru and ru-ab models #16

Open danielinux7 opened 2 years ago

danielinux7 commented 2 years ago

Ахцәажәара

Fine tune the current model on short sentences

Ауадаҩрақәа

Setting the notebook on Kaggle

Аӡбарақәа

See how it performs translating on this Huggingface dataset with statistics.

danielinux7 commented 2 years ago

There was't much enhancement, around 17% accuracy, here is the link to the stats file Now it's 29% accuracy. I m going to add casing in the training data. (It doesn ot seem that is has helped)

danielinux7 commented 2 years ago

Large Russian corpus: https://github.com/omnia-russica/omnia-russica.github.io

danielinux7 commented 2 years ago

If I need to deploy a model: https://www.tensorflow.org/tfx/tutorials/serving/rest_simple#add_tensorflow_serving_distribution_uri_as_a_package_source