BiT

This repository contains the training code of BiT introduced in our work: "BiT: Robustly Binarized Multi-distilled Transformer"

In this work, we identify a series of improvements which enables binary transformers at a much higher accuracy than what was possible previously. These include a two-set binarization scheme, a novel elastic binary activation function with learned parameters, and a multi-step distilation method. These approaches allow for the first time, fully binarized transformer models that are at a practical level of accuracy, approaching a full-precision BERT baseline on the GLUE language understanding benchmark within as little as 5.9%.

Citation

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

@article{liu2022bit,
title={BiT: Robustly Binarized Multi-distilled Transformer},
author={Liu, Zechun and Oguz, Barlas and Pappu, Aasish and Xiao, Lin and Yih, Scott and Li, Meng and Krishnamoorthi, Raghuraman and Mehdad, Yashar},
journal={arXiv preprint arXiv:2205.13016},
year={2022}
}

Run

1. Requirements:

• python 3.6, pytorch 1.7.1

2. Data:

• Download GLUE dataset and SQuAD dataset

  For data augmentation on GLUE, please follow the instruction in TinyBERT.

3. Pretrained models:

• Download pretrained models from hugging face model zoo.

  Dataset	Finetuned full-precision model
  MNLI	bert-base-uncased-MNLI
  QQP	bert-base-uncased-QQP
  QNLI	bert-base-uncased-QNLI
  SST-2	bert-base-uncased-SST-2
  CoLA	bert-base-uncased-CoLA
  STS-B	bert-base-uncased-STS-B
  MRPC	bert-base-uncased-MRPC
  RTE	bert-base-uncased-RTE
  Squad v1	bert-base-uncased-squad-v1

4. Steps to run:

• Specify the num_bits, data path and the pre-trained model path in scrips/run.sh file.

• Run bash scrips/run_glue.sh GLUE_dataset or Run bash scrips/run_squad.sh .

  E.g., bash scrips/run_glue.sh MNLI for running the MNLI dataset in GLUE dataset.

Models

1. GLUE dataset

(1) Without data augmentation

Method	#Bits	Size (M)	FLOPs (G)	MNLI m/mm	QQP	QNLI	SST-2	CoLA	STS-B	MRPC	RTE	Avg
BERT	32-32-32	418	22.5	84.9/85.5	91.4	92.1	93.2	59.7	90.1	86.3	72.2	83.9
BinaryBert	1-1-4	16.5	1.5	83.9/84.2	91.2	90.9	92.3	44.4	87.2	83.3	65.3	79.9
BinaryBert	1-1-2	16.5	0.8	62.7/63.9	79.9	52.6	82.5	14.6	6.5	68.3	52.7	53.7
BinaryBert	1-1-1	16.5	0.4	35.6/35.3	66.2	51.5	53.2	0	6.1	68.3	52.7	41.0
BiBert	1-1-1	13.4	0.4	66.1/67.5	84.8	72.6	88.7	25.4	33.6	72.5	57.4	63.2
BiT *	1-1-4	13.4	1.5	83.6/84.4	87.8	91.3	91.5	42.0	86.3	86.8	66.4	79.5
BiT *	1-1-2	13.4	0.8	82.1/82.5	87.1	89.3	90.8	32.1	82.2	78.4	58.1	75.0
BiT *	1-1-1	13.4	0.4	77.1/77.5	82.9	85.7	87.7	25.1	71.1	79.7	58.8	71.0
BiT	1-1-1	13.4	0.4	79.5/79.4	85.4	86.4	89.9	32.9	72	79.9	62.1	73.5

(2) With data augmentation

Method	#Bits	Size (M)	FLOPs (G)	MNLI m/mm	QQP	QNLI	SST-2	CoLA	STS-B	MRPC	RTE	Avg
BinaryBert	1-1-2	16.5	0.8	62.7/63.9*	79.9*	51.0	89.6	33.0	11.4	71.0	55.9	57.6
BinaryBert	1-1-1	16.5	0.4	35.6/35.3*	66.2*	66.1	78.3	7.3	22.1	69.3	57.7	48.7
BiBert	1-1-1	13.4	0.4	66.1/67.5*	84.8*	76.0	90.9	37.8	56.7	78.8	61.0	68.8
BiT *	1-1-2	13.4	0.8	82.1/82.5*	87.1*	88.8	92.5	43.2	86.3	90.4	72.9	80.4
BiT *	1-1-1	13.4	0.4	77.1/77.5*	82.9*	85.0	91.5	32.0	84.1	88.0	67.5	76.0
BiT	1-1-1	13.4	0.4	79.5/79.4*	85.4*	86.5	92.3	38.2	84.2	88	69.7	78.0

2. SQuAD dataset

Method	#Bits	SQuADv1.1 em/f1
BERT	32-32-32	82.6/89.7
BinaryBert	1-1-4	77.9/85.8
BinaryBert	1-1-2	72.3/81.8
BinaryBert	1-1-1	1.5/8.2
BiBert	1-1-1	8.5/18.9
BiT	1-1-1	63.1/74.9

Acknowledgement

The original code is borrowed from BinaryBERT.

Contact

Zechun Liu, Reality Labs, Meta Inc (liuzechun0216 at gmail.com)

License

BiT is CC-BY-NC 4.0 licensed as of now.