Diffcse Roberta Base Sts

🚀 DiffCSE: Difference-based Contrastive Learning for Sentence Embeddings

DiffCSE is an unsupervised contrastive learning framework for learning sentence embeddings. It learns embeddings sensitive to the difference between original and edited sentences, achieving state - of - the - art results among unsupervised sentence representation learning methods.

🚀 Quick Start

Our code is mainly based on the code of SimCSE. Please refer to their repository for more detailed information.

✨ Features

• Novel Framework: Propose DiffCSE, an unsupervised contrastive learning framework for sentence embeddings.
• State - of - the - art Results: Achieve state - of - the - art results among unsupervised sentence representation learning methods, outperforming unsupervised SimCSE by 2.3 absolute points on semantic textual similarity tasks.
• Equivariant Contrastive Learning: DiffSCE is an instance of equivariant contrastive learning (Dangovski et al., 2021).

📦 Installation

Requirements

• 
  • Python 3.9.5

Install our customized Transformers package

cd transformers-4.2.1
pip install .

⚠️ Important Note

If you have already installed transformers==4.2.1 through pip, you need to put modeling_bert.py into <your_python_env>/site-packages/transformers/models/bert/modeling_bert.py and modeling_roberta.py into <your_python_env>/site-packages/transformers/models/bert/modeling_roberta.py. We modify these two files in the package so that we can perform conditional pretraining tasks using BERT/RoBERTa. If possible, please directly pip install our customized Transformers package.

Install other packages

pip install -r requirements.txt

Download the pretraining dataset

cd data
bash download_wiki.sh

Download the downstream dataset

cd SentEval/data/downstream/
bash download_dataset.sh

💻 Usage Examples

Basic Usage

python train.py \
    --model_name_or_path bert-base-uncased \
    --generator_name distilbert-base-uncased \
    --train_file data/wiki1m_for_simcse.txt \
    --output_dir <your_output_model_dir> \
    --num_train_epochs 2 \
    --per_device_train_batch_size 64 \
    --learning_rate 7e-6 \
    --max_seq_length 32 \
    --evaluation_strategy steps \
    --metric_for_best_model stsb_spearman \
    --load_best_model_at_end \
    --eval_steps 125 \
    --pooler_type cls \
    --mlp_only_train \
    --overwrite_output_dir \
    --logging_first_step \
    --logging_dir <your_logging_dir> \
    --temp 0.05 \
    --do_train \
    --do_eval \
    --batchnorm \
    --lambda_weight 0.005 \
    --fp16 --masking_ratio 0.30

Advanced Usage

Our new arguments:

• --lambda_weight: the lambda coefficient mentioned in Section 3 of our paper.
• --masking_ratio: the masking ratio for MLM generator to randomly replace tokens.
• --generator_name: the model name of generator. For bert-base-uncased, we use distilbert-base-uncased. For roberta-base, we use distilroberta-base.

Arguments from SimCSE:

• --train_file: Training file path (data/wiki1m_for_simcse.txt).
• --model_name_or_path: Pre - trained checkpoints to start with such as BERT - based models (bert-base-uncased, bert-large-uncased, etc.) and RoBERTa - based models (RoBERTa-base, RoBERTa-large).
• --temp: Temperature for the contrastive loss. We always use 0.05.
• --pooler_type: Pooling method.
• --mlp_only_train: For unsupervised SimCSE or DiffCSE, it works better to train the model with MLP layer but test the model without it. You should use this argument when training unsupervised SimCSE/DiffCSE models.

For the results in our paper, we use a NVidia 2080Ti GPU with CUDA 11.2. Using different types of devices or different versions of CUDA/Python/PyTorch may lead to slightly different performance.

Evaluation

We provide a simple colab notebook to reproduce our results easily. We can also run the commands below for evaluation:

python evaluation.py \
    --model_name_or_path <your_output_model_dir> \
    --pooler cls_before_pooler \
    --task_set <sts|transfer|full> \
    --mode test

To evaluate our pretrained DiffCSE checkpoints, we can use the following scripts:

BERT

STS

python evaluation.py \
    --model_name_or_path voidism/diffcse-bert-base-uncased-sts \
    --pooler cls_before_pooler \
    --task_set sts \
    --mode test

Transfer Tasks

python evaluation.py \
    --model_name_or_path voidism/diffcse-bert-base-uncased-trans \
    --pooler cls_before_pooler \
    --task_set transfer \
    --mode test

RoBERTa

STS

python evaluation.py \
    --model_name_or_path voidism/diffcse-roberta-base-sts \
    --pooler cls_before_pooler \
    --task_set sts \
    --mode test

Transfer Tasks

python evaluation.py \
    --model_name_or_path voidism/diffcse-roberta-base-trans \
    --pooler cls_before_pooler \
    --task_set transfer \
    --mode test

For more detailed information, please check SimCSE's GitHub repo.

Pretrained models

• DiffCSE - BERT - base (STS): https://huggingface.co/voidism/diffcse-bert-base-uncased-sts
• DiffCSE - BERT - base (transfer tasks): https://huggingface.co/voidism/diffcse-bert-base-uncased-trans
• DiffCSE - RoBERTa - base (STS): https://huggingface.co/voidism/diffcse-roberta-base-sts
• DiffCSE - RoBERTa - base (transfer tasks): https://huggingface.co/voidism/diffcse-roberta-base-trans

We can load the models using the API provided by SimCSE. See Getting Started for more information.

from diffcse import DiffCSE
model_bert_sts = DiffCSE("voidism/diffcse-bert-base-uncased-sts")
model_bert_trans = DiffCSE("voidism/diffcse-bert-base-uncased-trans")
model_roberta_sts = DiffCSE("voidism/diffcse-roberta-base-sts")
model_roberta_trans = DiffCSE("voidism/diffcse-roberta-base-trans")

📄 License

This project is licensed under the Apache - 2.0 license.

Citations

Please cite our paper and the SimCSE paper if they are helpful to your work!

@inproceedings{chuang2022diffcse,
   title={{DiffCSE}: Difference-based Contrastive Learning for Sentence Embeddings},
   author={Chuang, Yung-Sung and Dangovski, Rumen and Luo, Hongyin and Zhang, Yang and Chang, Shiyu and Soljacic, Marin and Li, Shang-Wen and Yih, Wen-tau and Kim, Yoon and Glass, James},
   booktitle={Annual Conference of the North American Chapter of the Association for Computational Linguistics (NAACL)},
   year={2022}
}

@inproceedings{gao2021simcse,
   title={{SimCSE}: Simple Contrastive Learning of Sentence Embeddings},
   author={Gao, Tianyu and Yao, Xingcheng and Chen, Danqi},
   booktitle={Empirical Methods in Natural Language Processing (EMNLP)},
   year={2021}
}