File size: 1,464 Bytes
adb10ae
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
---
language: ko
---

# KoBigBird

Pretrained BigBird Model for Korean (**kobigbird-bert-base**)

## About

BigBird, is a sparse-attention based transformer which extends Transformer based models, such as BERT to much longer sequences.

BigBird relies on **block sparse attention** instead of normal attention (i.e. BERT's attention) and can handle sequences up to a length of 4096 at a much lower compute cost compared to BERT. It has achieved SOTA on various tasks involving very long sequences such as long documents summarization, question-answering with long contexts.

Model is warm started from Korean BERT’s checkpoint.

## How to use

WARN: Please use `BertTokenizer` instead of `BigBirdTokenizer`.

```python
from transformers import AutoModel, AutoTokenizer

# by default its in `block_sparse` mode with num_random_blocks=3, block_size=64
model = AutoModel.from_pretrained("monologg/kobigbird-bert-base")

# you can change `attention_type` to full attention like this:
model = AutoModel.from_pretrained("monologg/kobigbird-bert-base", attention_type="original_full")

# you can change `block_size` & `num_random_blocks` like this:
model = AutoModel.from_pretrained("monologg/kobigbird-bert-base", block_size=16, num_random_blocks=2)

tokenizer = AutoTokenizer.from_pretrained("monologg/kobigbird-bert-base")
text = "한국어 BigBird 모델을 공개합니다!"
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```