First Open Source Width Benchmark

Recent research focuses on optimizing the width of neural networks, aiming to improve efficiency and performance without sacrificing accuracy. This involves developing novel methods for automatically searching optimal layer widths, often employing supernets and evolutionary algorithms, and addressing issues like "over-squashing" through techniques such as width-aware message passing. A newly established open-source benchmark facilitates comparative analysis of different width optimization strategies, accelerating progress in this crucial area of deep learning model design. The resulting more efficient models have significant implications for deploying deep learning on resource-constrained devices and accelerating the training process.