SensiMix: Sensitivity-Aware 8-bit index & 1-bit value mixed precision quantization for BERT compression.

Given a pre-trained BERT, how can we compress it to a fast and lightweight one while maintaining its accuracy? Pre-training language model, such as BERT, is effective for improving the performance of natural language processing (NLP) tasks. However, heavy models like BERT have problems of large memory cost and long inference time. In this paper, we propose SensiMix (Sensitivity-Aware Mixed Precision Quantization), a novel quantization-based BERT compression method that considers the sensitivity of different modules of BERT. SensiMix effectively applies 8-bit index quantization and 1-bit value quantization to the sensitive and insensitive parts of BERT, maximizing the compression rate while minimizing the accuracy drop. We also propose three novel 1-bit training methods to minimize the accuracy drop: Absolute Binary Weight Regularization, Prioritized Training, and Inverse Layer-wise Fine-tuning. Moreover, for fast inference, we apply FP16 general matrix multiplication (GEMM) and XNOR-Count GEMM for 8-bit and 1-bit quantization parts of the model, respectively. Experiments on four GLUE downstream tasks show that SensiMix compresses the original BERT model to an equally effective but lightweight one, reducing the model size by a factor of 8× and shrinking the inference time by around 80% without noticeable accuracy drop.