Q-Rater: Non-Convex Optimization for Post-Training Uniform Quantization

TL;DR

This paper introduces Q-Rater, a non-convex optimization-based post-training quantization method that improves model accuracy in low-bit scenarios by optimizing hyper-parameters through task loss monitoring.

Contribution

It proposes a novel non-convex optimization approach for post-training quantization, leveraging hyper-parameter exploration and incremental optimization for better low-bit model accuracy.

Findings

Achieves higher accuracy in low-bit quantization.

Effective hyper-parameter tuning via task loss monitoring.

Demonstrates superiority over convex methods in experiments.

Abstract

Various post-training uniform quantization methods have usually been studied based on convex optimization. As a result, most previous ones rely on the quantization error minimization and/or quadratic approximations. Such approaches are computationally efficient and reasonable when a large number of quantization bits are employed. When the number of quantization bits is relatively low, however, non-convex optimization is unavoidable to improve model accuracy. In this paper, we propose a new post-training uniform quantization technique considering non-convexity. We empirically show that hyper-parameters for clipping and rounding of weights and activations can be explored by monitoring task loss. Then, an optimally searched set of hyper-parameters is frozen to proceed to the next layer such that an incremental non-convex optimization is enabled for post-training quantization. Throughout extensive experimental results using various models, our proposed technique presents higher model accuracy, especially for a low-bit quantization.