rubert-tiny / README.md
cointegrated's picture
Update README.md
d12dfcb
|
raw
history blame
1.04 kB
metadata
language:
  - ru
  - en
tags:
  - russian
  - fill-mask
  - pretraining
  - embeddings
  - masked-lm
  - tiny
license: mit
widget:
  - text: Миниатюрная модель для [MASK] разных задач.

This is a very small distilled version of the bert-base-multilingual-cased model for Russian and English.

This model is useful if you want to fine-tune it for a relatively simple Russian task (e.g. NER or sentiment classification), and you care more about speed and size than about accuracy. It is approximately x10 smaller and faster than DeepPavlov/rubert-base-cased-sentence.

It was trained on the Yandex Translate corpus using MLM loss (partially distilled from bert-base-multilingual-cased) and translation ranking loss (partially distilled from LaBSE).