Implementing SimCSE(paper, official repository) using TensorFlow 2 and KR-BERT.

Overview

KR-BERT-SimCSE

Implementing SimCSE(paper, official repository) using TensorFlow 2 and KR-BERT.

Training

Unsupervised

python train_unsupervised.py --mixed_precision

I used Korean Wikipedia Corpus that is divided into sentences in advance. (Check out tfds-korean catalog page for details)

  • Settings
    • KR-BERT character
    • peak learning rate 3e-5
    • batch size 64
    • Total steps: 25,000
    • 0.05 warmup rate, and linear decay learning rate scheduler
    • temperature 0.05
    • evalaute on KLUE STS and KorSTS every 250 steps
    • max sequence length 64
    • Use pooled outputs for training, and [CLS] token's representations for inference

The hyperparameters were not tuned and mostly followed the values in the paper.

Supervised

python train_supervised.py --mixed_precision

I used KorNLI for supervised training. (Check out tfds-korean catalog page)

  • Settings
    • KR-BERT character
    • batch size 128
    • epoch 3
    • peak learning rate 5e-5
    • 0.05 warmup rate, and linear decay learning rate scheduler
    • temperature 0.05
    • evalaute on KLUE STS and KorSTS every 125 steps
    • max sequence length 48
    • Use pooled outputs for training, and [CLS] token's representations for inference

The hyperparameters were not tuned and mostly followed the values in the paper.

Results

KorSTS (dev set results)

model 100 X Spearman correlation
KR-BERT base
SimCSE
unsupervised bi encoding 79.99
KR-BERT base
SimCSE-supervised
trained on KorNLI bi encoding 84.88
SRoBERTa base* unsupervised bi encoding 63.34
SRoBERTa base* trained on KorNLI bi encoding 76.48
SRoBERTa base* trained on KorSTS bi encoding 83.68
SRoBERTa base* trained on KorNLI -> KorSTS bi encoding 83.54
SRoBERTa large* trained on KorNLI bi encoding 77.95
SRoBERTa large* trained on KorSTS bi encoding 84.74
SRoBERTa large* trained on KorNLI -> KorSTS bi encoding 84.21

KorSTS (test set results)

model 100 X Spearman correlation
KR-BERT base
SimCSE
unsupervised bi encoding 73.25
KR-BERT base
SimCSE-supervised
trained on KorNLI bi encoding 80.72
SRoBERTa base* unsupervised bi encoding 48.96
SRoBERTa base* trained on KorNLI bi encoding 74.19
SRoBERTa base* trained on KorSTS bi encoding 78.94
SRoBERTa base* trained on KorNLI -> KorSTS bi encoding 80.29
SRoBERTa large* trained on KorNLI bi encoding 75.46
SRoBERTa large* trained on KorSTS bi encoding 79.55
SRoBERTa large* trained on KorNLI -> KorSTS bi encoding 80.49
SRoBERTa base* trained on KorSTS cross encoding 83.00
SRoBERTa large* trained on KorSTS cross encoding 85.27

KLUE STS (dev set results)

model 100 X Pearson's correlation
KR-BERT base
SimCSE
unsupervised bi encoding 74.45
KR-BERT base
SimCSE-supervised
trained on KorNLI bi encoding 79.42
KR-BERT base* supervised cross encoding 87.50

References

@misc{gao2021simcse,
    title={SimCSE: Simple Contrastive Learning of Sentence Embeddings},
    author={Tianyu Gao and Xingcheng Yao and Danqi Chen},
    year={2021},
    eprint={2104.08821},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}
@misc{ham2020kornli,
    title={KorNLI and KorSTS: New Benchmark Datasets for Korean Natural Language Understanding},
    author={Jiyeon Ham and Yo Joong Choe and Kyubyong Park and Ilji Choi and Hyungjoon Soh},
    year={2020},
    eprint={2004.03289},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}
@misc{park2021klue,
    title={KLUE: Korean Language Understanding Evaluation},
    author={Sungjoon Park and Jihyung Moon and Sungdong Kim and Won Ik Cho and Jiyoon Han and Jangwon Park and Chisung Song and Junseong Kim and Yongsook Song and Taehwan Oh and Joohong Lee and Juhyun Oh and Sungwon Lyu and Younghoon Jeong and Inkwon Lee and Sangwoo Seo and Dongjun Lee and Hyunwoo Kim and Myeonghwa Lee and Seongbo Jang and Seungwon Do and Sunkyoung Kim and Kyungtae Lim and Jongwon Lee and Kyumin Park and Jamin Shin and Seonghyun Kim and Lucy Park and Alice Oh and Jung-Woo Ha and Kyunghyun Cho},
    year={2021},
    eprint={2105.09680},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}
Owner
Jeong Ukjae
Jeong Ukjae
DziriBERT: a Pre-trained Language Model for the Algerian Dialect

DziriBERT is the first Transformer-based Language Model that has been pre-trained specifically for the Algerian Dialect.

117 Jan 07, 2023
Mkdocs + material + cool stuff

Modern-Python-Doc-Example mkdocs + material + cool stuff Doc is live here Features out of the box amazing good looking website thanks to mkdocs.org an

Francesco Saverio Zuppichini 61 Oct 26, 2022
DeepPavlov Tutorials

DeepPavlov tutorials DeepPavlov: Sentence Classification with Word Embeddings DeepPavlov: Transfer Learning with BERT. Classification, Tagging, QA, Ze

Neural Networks and Deep Learning lab, MIPT 28 Sep 13, 2022
A single model that parses Universal Dependencies across 75 languages.

A single model that parses Universal Dependencies across 75 languages. Given a sentence, jointly predicts part-of-speech tags, morphology tags, lemmas, and dependency trees.

Dan Kondratyuk 189 Nov 29, 2022
Topic Inference with Zeroshot models

zeroshot_topics Table of Contents Installation Usage License Installation zeroshot_topics is distributed on PyPI as a universal wheel and is available

Rita Anjana 55 Nov 28, 2022
Faster, modernized fork of the language identification tool langid.py

py3langid py3langid is a fork of the standalone language identification tool langid.py by Marco Lui. Original license: BSD-2-Clause. Fork license: BSD

Adrien Barbaresi 12 Nov 05, 2022
🏖 Easy training and deployment of seq2seq models.

Headliner Headliner is a sequence modeling library that eases the training and in particular, the deployment of custom sequence models for both resear

Axel Springer Ideas Engineering GmbH 231 Nov 18, 2022
Addon for adding subtitle files to blender VSE as Text sequences. Using pysub2 python module.

Import Subtitles for Blender VSE Addon for adding subtitle files to blender VSE as Text sequences. Using pysub2 python module. Supported formats by py

4 Feb 27, 2022
Searching keywords in PDF file folders

keyword_searching Steps to use this Python scripts: (1)Paste this script into the file folder containing the PDF files you need to search from; (2)Thi

1 Nov 08, 2021
Dust model dichotomous performance analysis

Dust-model-dichotomous-performance-analysis Using a collated dataset of 90,000 dust point source observations from 9 drylands studies from around the

1 Dec 17, 2021
LUKE -- Language Understanding with Knowledge-based Embeddings

LUKE (Language Understanding with Knowledge-based Embeddings) is a new pre-trained contextualized representation of words and entities based on transf

Studio Ousia 587 Dec 30, 2022
Transformer training code for sequential tasks

Sequential Transformer This is a code for training Transformers on sequential tasks such as language modeling. Unlike the original Transformer archite

Meta Research 578 Dec 13, 2022
Tensorflow Implementation of A Generative Flow for Text-to-Speech via Monotonic Alignment Search

Tensorflow Implementation of A Generative Flow for Text-to-Speech via Monotonic Alignment Search

Ankur Dhuriya 10 Oct 13, 2022
天池中药说明书实体识别挑战冠军方案;中文命名实体识别;NER; BERT-CRF & BERT-SPAN & BERT-MRC;Pytorch

天池中药说明书实体识别挑战冠军方案;中文命名实体识别;NER; BERT-CRF & BERT-SPAN & BERT-MRC;Pytorch

zxx飞翔的鱼 751 Dec 30, 2022
Extract city and country mentions from Text like GeoText without regex, but FlashText, a Aho-Corasick implementation.

flashgeotext ⚡ 🌍 Extract and count countries and cities (+their synonyms) from text, like GeoText on steroids using FlashText, a Aho-Corasick impleme

Ben 57 Dec 16, 2022
NLPIR tutorial: pretrain for IR. pre-train on raw textual corpus, fine-tune on MS MARCO Document Ranking

pretrain4ir_tutorial NLPIR tutorial: pretrain for IR. pre-train on raw textual corpus, fine-tune on MS MARCO Document Ranking 用作NLPIR实验室, Pre-training

ZYMa 12 Apr 07, 2022
A full spaCy pipeline and models for scientific/biomedical documents.

This repository contains custom pipes and models related to using spaCy for scientific documents. In particular, there is a custom tokenizer that adds

AI2 1.3k Jan 03, 2023
The official implementation of VAENAR-TTS, a VAE based non-autoregressive TTS model.

VAENAR-TTS This repo contains code accompanying the paper "VAENAR-TTS: Variational Auto-Encoder based Non-AutoRegressive Text-to-Speech Synthesis". Sa

THUHCSI 138 Oct 28, 2022
RecipeReduce: Simplified Recipe Processing for Lazy Programmers

RecipeReduce This repo will help you figure out the amount of ingredients to buy for a certain number of meals with selected recipes. RecipeReduce Get

Qibin Chen 9 Apr 22, 2022
Yuqing Xie 2 Feb 17, 2022