Table of contents
PhoBERT: Pre-trained language models for Vietnamese
Pre-trained PhoBERT models are the state-of-the-art language models for Vietnamese (Pho, i.e. "Phở", is a popular food in Vietnam):
- Two PhoBERT versions of "base" and "large" are the first public large-scale monolingual language models pre-trained for Vietnamese. PhoBERT pre-training approach is based on RoBERTa which optimizes the BERT pre-training procedure for more robust performance.
- PhoBERT outperforms previous monolingual and multilingual approaches, obtaining new state-of-the-art performances on four downstream Vietnamese NLP tasks of Part-of-speech tagging, Dependency parsing, Named-entity recognition and Natural language inference.
The general architecture and experimental results of PhoBERT can be found in our paper:
@inproceedings{phobert,
title = {{PhoBERT: Pre-trained language models for Vietnamese}},
author = {Dat Quoc Nguyen and Anh Tuan Nguyen},
booktitle = {Findings of the Association for Computational Linguistics: EMNLP 2020},
year = {2020},
pages = {1037--1042}
}
Please CITE our paper when PhoBERT is used to help produce published results or is incorporated into other software.
Using PhoBERT with transformers
Installation
- Install
transformers
with pip:pip install transformers
, or installtransformers
from source.
Note that we merged a slow tokenizer for PhoBERT into the maintransformers
branch. The process of merging a fast tokenizer for PhoBERT is in the discussion, as mentioned in this pull request. If users would like to utilize the fast tokenizer, the users might installtransformers
as follows:
git clone --single-branch --branch fast_tokenizers_BARTpho_PhoBERT_BERTweet https://github.com/datquocnguyen/transformers.git
cd transformers
pip3 install -e .
- Install
tokenizers
with pip:pip3 install tokenizers
Pre-trained models
Model | #params | Arch. | Max length | Pre-training data | License |
---|---|---|---|---|---|
vinai/phobert-base-v2 | 135M | base | 256 | 20GB of Wikipedia and News texts + 120GB of texts from OSCAR-2301 | GNU Affero GPL v3 |
vinai/phobert-base | 135M | base | 256 | 20GB of Wikipedia and News texts | MIT License |
vinai/phobert-large | 370M | large | 256 | 20GB of Wikipedia and News texts | MIT License |
Example usage
import torch
from transformers import AutoModel, AutoTokenizer
phobert = AutoModel.from_pretrained("vinai/phobert-base-v2")
tokenizer = AutoTokenizer.from_pretrained("vinai/phobert-base-v2")
# INPUT TEXT MUST BE ALREADY WORD-SEGMENTED!
sentence = 'Chúng_tôi là những nghiên_cứu_viên .'
input_ids = torch.tensor([tokenizer.encode(sentence)])
with torch.no_grad():
features = phobert(input_ids) # Models outputs are now tuples
## With TensorFlow 2.0+:
# from transformers import TFAutoModel
# phobert = TFAutoModel.from_pretrained("vinai/phobert-base")
Using PhoBERT with fairseq
Please see details at HERE!
Notes
In case the input texts are raw
, i.e. without word segmentation, a word segmenter must be applied to produce word-segmented texts before feeding to PhoBERT. As PhoBERT employed the RDRSegmenter from VnCoreNLP to pre-process the pre-training data (including Vietnamese tone normalization and word and sentence segmentation), it is recommended to also use the same word segmenter for PhoBERT-based downstream applications w.r.t. the input raw texts.
Installation
pip install py_vncorenlp
Example usage
import py_vncorenlp
# Automatically download VnCoreNLP components from the original repository
# and save them in some local machine folder
py_vncorenlp.download_model(save_dir='/absolute/path/to/vncorenlp')
# Load the word and sentence segmentation component
rdrsegmenter = py_vncorenlp.VnCoreNLP(annotators=["wseg"], save_dir='/absolute/path/to/vncorenlp')
text = "Ông Nguyễn Khắc Chúc đang làm việc tại Đại học Quốc gia Hà Nội. Bà Lan, vợ ông Chúc, cũng làm việc tại đây."
output = rdrsegmenter.word_segment(text)
print(output)
# ['Ông Nguyễn_Khắc_Chúc đang làm_việc tại Đại_học Quốc_gia Hà_Nội .', 'Bà Lan , vợ ông Chúc , cũng làm_việc tại đây .']