COMQ: A Backpropagation-Free Algorithm for Post-Training Quantization
This work addresses the problem of efficient deployment of neural networks for practitioners, offering a hyper-parameter-free method that is incremental but provides strong specific gains in quantization accuracy.
The paper tackles the challenge of compressing large neural networks via post-training quantization without accuracy loss, proposing COMQ, a backpropagation-free algorithm that achieves less than 1% accuracy drop in 4-bit Vision Transformers and 0.3% drop in convolutional neural networks.
Post-training quantization (PTQ) has emerged as a practical approach to compress large neural networks, making them highly efficient for deployment. However, effectively reducing these models to their low-bit counterparts without compromising the original accuracy remains a key challenge. In this paper, we propose an innovative PTQ algorithm termed COMQ, which sequentially conducts coordinate-wise minimization of the layer-wise reconstruction errors. We consider the widely used integer quantization, where every quantized weight can be decomposed into a shared floating-point scalar and an integer bit-code. Within a fixed layer, COMQ treats all the scaling factor(s) and bit-codes as the variables of the reconstruction error. Every iteration improves this error along a single coordinate while keeping all other variables constant. COMQ is easy to use and requires no hyper-parameter tuning. It instead involves only dot products and rounding operations. We update these variables in a carefully designed greedy order, significantly enhancing the accuracy. COMQ achieves remarkable results in quantizing 4-bit Vision Transformers, with a negligible loss of less than 1% in Top-1 accuracy. In 4-bit INT quantization of convolutional neural networks, COMQ maintains near-lossless accuracy with a minimal drop of merely 0.3% in Top-1 accuracy.