Bit-Mixer: Mixed-precision networks with runtime bit-width selection

TL;DR

Bit-Mixer introduces a novel meta-quantized network that enables dynamic, runtime adjustment of layer bit-widths in mixed-precision networks without sacrificing accuracy, enhancing deployment flexibility.

Contribution

It is the first method to train a network allowing layer-wise bit-width changes at runtime, using transitional Batch-Norms and a 3-stage optimization process.

Findings

Enables flexible bit-width adjustment during inference.

Maintains high accuracy despite runtime bit-width changes.

Demonstrates effective deployment on devices with variable characteristics.

Abstract

Mixed-precision networks allow for a variable bit-width quantization for every layer in the network. A major limitation of existing work is that the bit-width for each layer must be predefined during training time. This allows little flexibility if the characteristics of the device on which the network is deployed change during runtime. In this work, we propose Bit-Mixer, the very first method to train a meta-quantized network where during test time any layer can change its bid-width without affecting at all the overall network's ability for highly accurate inference. To this end, we make 2 key contributions: (a) Transitional Batch-Norms, and (b) a 3-stage optimization process which is shown capable of training such a network. We show that our method can result in mixed precision networks that exhibit the desirable flexibility properties for on-device deployment without compromising accuracy. Code will be made available.