Welcome to the Model Zoo!

Here you can find NLP models for Russian, implemented in HF transformers🤗

Models:

ruT5

Text2Text Generation task T5 paper

• Large: HF Model
• Base: HF Model

Model parameters

ruRoBerta

fill-mask task Roberta paper

• Large: HF Model

ruBert

fill-mask task Bert paper

• Large: HF Model
• Base: HF Model

How to:

Use this to explore the models or run them on your machine.

Model set up:

pip install -r requirements.txt

Pipeline usage

from transformers import pipeline

unmasker = pipeline("fill-mask", model="sberbank-ai/ruRoberta-large")
unmasker("Евгений Понасенков назвал <mask> величайшим маэстро.", top_k=1)

Classical usage

# ruRoberta-large example 
from transformers import RobertaForMaskedLM,RobertaTokenizer

model=RobertaForMaskedLM.from_pretrained('sberbank-ai/ruRoberta-large')

tokenizer=RobertaTokenizer.from_pretrained('sberbank-ai/ruRoberta-large')

unmasker = pipeline('fill-mask', model=model,tokenizer=tokenizer)
unmasker("Стоит чаще писать на Хабр про <mask>.")

Use BertViz to obtain model visualizations

Roberta model_view:

/ !

from transformers import RobertaModel, RobertaTokenizer
from bertviz import model_view

model_version = 'sberbank-ai/ruRoberta-large'
model = RobertaModel.from_pretrained(model_version, output_attentions=True)
tokenizer = RobertaTokenizer.from_pretrained(model_version)

sentence_a = "The cat sat on the mat"
sentence_b = "The cat lay on the rug"
inputs = tokenizer.encode_plus(sentence_a, sentence_b, return_tensors='pt', add_special_tokens=True)
input_ids = inputs['input_ids']
attention = model(input_ids)[-1]
input_id_list = input_ids[0].tolist() # Batch index 0
tokens = tokenizer.convert_ids_to_tokens(input_id_list)
model_view(attention, tokens)