BLEURT

Pretrained model on English language. It was introduced in this paper, described in this blogpost and first released in this repository.

The team releasing BLEURT did not write a model card for this model so this model card has been written by the Surfer team.

Original TensorFlow implementation has been converted to PyTorch with help of this article by Surfer team.

Visit us at surferseo.com.

How to use

Since BLEURT is not implemented in transformers library yet, you have to import BleurtModel from bleurt_model.py

import torch
from bleurt_model import BleurtModel
from transformers import BertTokenizerFast

model = BleurtModel.from_pretrained("SurferSEO/bleurt")
tokenizer = BertTokenizerFast.from_pretrained("SurferSEO/bleurt")

sentence_pairs = [("I love surfing.", "I'd like to surf.")]
encoded = tokenizer(sentence_pairs, padding=True, truncation=True, return_tensors="pt")
input_ids, attention_mask, token_type_ids = (
    encoded["input_ids"],
    encoded["attention_mask"],
    encoded["token_type_ids"],
)
with torch.set_grad_enabled(False):
    predictions = model(
        input_ids=input_ids,
        attention_mask=attention_mask,
        token_type_ids=token_type_ids,
    )
print(predictions)