Home

Awesome

Patient Knowledge Distillation for BERT Model Compression

Knowledge distillation for BERT model

Installation

Run command below to install the environment

conda install pytorch torchvision cudatoolkit=10.0 -c pytorch
pip install -r requirements.txt

Training

Objective Function

L = (1 - \alpha) L_CE + \alpha * L_DS + \beta * L_PT,

where L_CE is the CrossEntropy loss, DS is the usual Distillation loss, and PT is the proposed loss. Please see our paper below for more details.

Data Preprocess

Modify the HOME_DATA_FOLDER in envs.py and put all data under it (by default it is ./data), RTE data is uploaded for your convenience.

Predefinted Training

Run NLI_KD_training.py to start training, you can set DEBUG = True to run some pre-defined arguments

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

Citation

If you find this code useful for your research, please consider citing:

@article{sun2019patient,
title={Patient Knowledge Distillation for BERT Model Compression},
author={Sun, Siqi and Cheng, Yu and Gan, Zhe and Liu, Jingjing},
journal={arXiv preprint arXiv:1908.09355},
year={2019}
}

Paper is available at here.