cointegrated

/

rubert-tiny

like 26

This is a very small distilled version of the bert-base-multilingual-cased model for Russian and English (45 MB, 12M parameters). There is also an updated version of this model, rubert-tiny2, with a larger vocabulary and better quality on practically all Russian NLU tasks.

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 a base-sized BERT. Its [CLS] embeddings can be used as a sentence representation aligned between Russian and English.

It was trained on the Yandex Translate corpus, OPUS-100 and Tatoeba, using MLM loss (distilled from bert-base-multilingual-cased), translation ranking loss, and [CLS] embeddings distilled from LaBSE, rubert-base-cased-sentence, Laser and USE.

There is a more detailed description in Russian.

Sentence embeddings can be produced as follows:

# pip install transformers sentencepiece
import torch
from transformers import AutoTokenizer, AutoModel
tokenizer = AutoTokenizer.from_pretrained("cointegrated/rubert-tiny")
model = AutoModel.from_pretrained("cointegrated/rubert-tiny")
# model.cuda()  # uncomment it if you have a GPU

def embed_bert_cls(text, model, tokenizer):
    t = tokenizer(text, padding=True, truncation=True, return_tensors='pt')
    with torch.no_grad():
        model_output = model(**{k: v.to(model.device) for k, v in t.items()})
    embeddings = model_output.last_hidden_state[:, 0, :]
    embeddings = torch.nn.functional.normalize(embeddings)
    return embeddings[0].cpu().numpy()

print(embed_bert_cls('привет мир', model, tokenizer).shape)
# (312,)