使用pytorch+transformers复现了SimCSE论文中的有监督训练和无监督训练方法

Related tags

Text Data & NLPSimCSE
Overview

SimCSE复现

项目描述

SimCSE是一种简单但是很巧妙的NLP对比学习方法,创新性地引入Dropout的方式,对样本添加噪声,从而达到对正样本增强的目的。 该框架的训练目的为:对于batch中的每个样本,拉近其与正样本之间的距离,拉远其与负样本之间的距离,使得模型能够在大规模无监督语料(也可以使用有监督的语料)中学习到文本相似关系。 详见论文:Simple Contrastive Learning of Sentence EmbeddingsSimCSE官方代码仓库

本项目使用pytorch+transformers复现了SimCSE论文中的有监督训练和无监督训练方法,并且在STS-B数据集上进行消融实验,评价指标为Spearman相关系数,预训练模型为Bert-base-uncased, 验证了SimCSE的有效性。在STS-B数据集上,有监督训练和无监督训练的复现效果如下表。

在无监督训练中,dropout=0.1时,复现效果比原文略差,但也比较接近。当dropout=0.2时,复现效果比原文略高。 ** 但在有监督训练中,不知是否由于batch size过小(原论文使用512),复现效果与论文的效果相差较远,后续会进行排查。 **

训练方法 learning rate batch size dropout Spearman’s correlation
原论文 无监督 3e-5 64 0.1 0.763
复现 无监督 3e-5 64 0.2 0.771
复现 无监督 3e-5 64 0.1 0.748
原论文 有监督 5e-5 512 0.1 0.816
复现 有监督 5e-5 64 0.1 0.764

运行环境

python==3.6、transformers==3.1.0、torch==1.6.0

项目结构

  • data:存放训练数据
    • stsbenchmark:STS-B数据集
      • sts-dev.csv:STS-B验证集
      • sts-test.csv:STS-B验测试集
    • nli_for_simcse.csv:数量275601为的NLI数据集
    • wiki1m_for_simcse.txt:维基百科上获取的100w的文本
  • output:输出目录
  • pretrain_model:预训练模型存放位置
  • script:脚本存放位置。
  • dataset.py
  • model.py:模型代码,包含有监督和无监督损失函数的计算方式
  • train.py:训练代码

使用方法

Quick Start

下载训练数据:

bash script/download_nli.sh
bash script/download_wiki.sh

无监督训练,运行脚本

bash script/run_unsup_train.sh

有监督训练,运行脚本

bash script/run_sup_train.sh

实验

无监督训练

从前四条实验数据中可以看到,较大的batch size在一定程度上可以增加模型的泛化性。

dropout为0.2的时候,训练效果比0.1与0.3更好,有可能dropout=0.1加入的噪声过小,而dropout=0.3加入的噪声过大,增强得到的样本与原始样本差异较大。

learning rate batch size dropout 在哪一步得到best checkpoint 验证集上的得分 测试集上的得分
3e-5 256 0.1 6000 0.800 0.761
3e-5 128 0.1 4200 0.799 0.747
3e-5 64 0.1 10900 0.803 0.748
3e-5 32 0.1 21300 0.787 0.714
3e-5 64 0.2 11200 0.811 0.771
3e-5 64 0.3 6300 0.781 0.745
1e-5 64 0.1 16400 0.798 0.751

有监督训练

有监督实验的复现结果未达到预期,超参数相同时,在验证集上的得分略高于无监督,但是在测试集上,得分基本没有差异。增大有监督训练的学习率,有监督的训练的得分略高于无监督训练, 但还是与论文声称的0.816相差较远,原论文使用512的batch size, 不知是否由于batch size的设置有关,后续会对有监督的训练代码进一步排查。

不过从训练曲线可以看到,有监督训练的收敛速度明显快于无监督训练,这也符合我们的认知。

训练方法 learning rate batch size dropout 在哪一步得到best checkpoint 验证集上的得分 测试集上的得分
无监督 3e-5 64 0.1 10900 0.803 0.748
有监督 3e-5 64 0.1 200 0.810 0.748
有监督 5e-5 64 0.1 2300 0.809 0.764
有监督 3e-5 32 0.1 200 0.808 0.743
有监督 5e-5 32 0.1 200 0.806 0.746

无监督训练过程中,验证集得分的变化曲线: avatar

有监督训练过程中,验证集得分的变化曲线: avatar

REFERENCE

TODO

  • 排查有监督学习的效果不符合预期的原因
This is my reading list for my PhD in AI, NLP, Deep Learning and more.

This is my reading list for my PhD in AI, NLP, Deep Learning and more.

Zhong Peixiang 156 Dec 21, 2022
A paper list of pre-trained language models (PLMs).

Large-scale pre-trained language models (PLMs) such as BERT and GPT have achieved great success and become a milestone in NLP.

RUCAIBox 124 Jan 02, 2023
An A-SOUL Text Generator Based on CPM-Distill.

ASOUL-Generator-Backend 本项目为 https://asoul.infedg.xyz/ 的后端。 模型为基于 CPM-Distill 的 transformers 转化版本 CPM-Generate-distill 训练而成。

infinityedge 46 Dec 11, 2022
Named Entity Recognition API used by TEI Publisher

TEI Publisher Named Entity Recognition API This repository contains the API used by TEI Publisher's web-annotation editor to detect entities in the in

e-editiones.org 14 Nov 15, 2022
100+ Chinese Word Vectors 上百种预训练中文词向量

Chinese Word Vectors 中文词向量 中文 This project provides 100+ Chinese Word Vectors (embeddings) trained with different representations (dense and sparse),

embedding 10.4k Jan 09, 2023
Code for paper "Role-oriented Network Embedding Based on Adversarial Learning between Higher-order and Local Features"

Role-oriented Network Embedding Based on Adversarial Learning between Higher-order and Local Features Train python main.py --dataset brazil-flights C

wang zhang 0 Jun 28, 2022
IMS-Toucan is a toolkit to train state-of-the-art Speech Synthesis models

IMS-Toucan is a toolkit to train state-of-the-art Speech Synthesis models. Everything is pure Python and PyTorch based to keep it as simple and beginner-friendly, yet powerful as possible.

Digital Phonetics at the University of Stuttgart 247 Jan 05, 2023
Baseline code for Korean open domain question answering(ODQA)

Open-Domain Question Answering(ODQA)는 다양한 주제에 대한 문서 집합으로부터 자연어 질의에 대한 답변을 찾아오는 task입니다. 이때 사용자 질의에 답변하기 위해 주어지는 지문이 따로 존재하지 않습니다. 따라서 사전에 구축되어있는 Knowl

VUMBLEB 69 Nov 04, 2022
Simple python code to fix your combo list by removing any text after a separator or removing duplicate combos

Combo List Fixer A simple python code to fix your combo list by removing any text after a separator or removing duplicate combos Removing any text aft

Hamidreza Dehghan 3 Dec 05, 2022
Code associated with the Don't Stop Pretraining ACL 2020 paper

dont-stop-pretraining Code associated with the Don't Stop Pretraining ACL 2020 paper Citation @inproceedings{dontstoppretraining2020, author = {Suchi

AI2 449 Jan 04, 2023
DLO8012: Natural Language Processing & CSL804: Computational Lab - II

NATURAL-LANGUAGE-PROCESSING-AND-COMPUTATIONAL-LAB-II DLO8012: NLP & CSL804: CL-II [SEMESTER VIII] Syllabus NLP - Reference Books THE WALL MEGA SATISH

AMEY THAKUR 7 Apr 28, 2022
Neural network models for joint POS tagging and dependency parsing (CoNLL 2017-2018)

Neural Network Models for Joint POS Tagging and Dependency Parsing Implementations of joint models for POS tagging and dependency parsing, as describe

Dat Quoc Nguyen 152 Sep 02, 2022
A high-level yet extensible library for fast language model tuning via automatic prompt search

ruPrompts ruPrompts is a high-level yet extensible library for fast language model tuning via automatic prompt search, featuring integration with Hugg

Sber AI 37 Dec 07, 2022
DVC-NLP-Simple-usecase

dvc-NLP-simple-usecase DVC NLP project Reference repository: official reference repo DVC STUDIO MY View Bag of Words- Krish Naik TF-IDF- Krish Naik ST

SUNNY BHAVEEN CHANDRA 2 Oct 02, 2022
Pre-training with Extracted Gap-sentences for Abstractive SUmmarization Sequence-to-sequence models

PEGASUS library Pre-training with Extracted Gap-sentences for Abstractive SUmmarization Sequence-to-sequence models, or PEGASUS, uses self-supervised

Google Research 1.4k Dec 22, 2022
Levenshtein and Hamming distance computation

distance - Utilities for comparing sequences This package provides helpers for computing similarities between arbitrary sequences. Included metrics ar

112 Dec 22, 2022
A library for Multilingual Unsupervised or Supervised word Embeddings

MUSE: Multilingual Unsupervised and Supervised Embeddings MUSE is a Python library for multilingual word embeddings, whose goal is to provide the comm

Facebook Research 3k Jan 06, 2023
Python interface for converting Penn Treebank trees to Stanford Dependencies and Universal Depenencies

PyStanfordDependencies Python interface for converting Penn Treebank trees to Universal Dependencies and Stanford Dependencies. Example usage Start by

David McClosky 64 May 08, 2022
SimpleChinese2 集成了许多基本的中文NLP功能,使基于 Python 的中文文字处理和信息提取变得简单方便。

SimpleChinese2 SimpleChinese2 集成了许多基本的中文NLP功能,使基于 Python 的中文文字处理和信息提取变得简单方便。 声明 本项目是为方便个人工作所创建的,仅有部分代码原创。

Ming 30 Dec 02, 2022