Narrow bit-width data formats are key to reducing the computational and storage costs of modern deep learning applications. This paper evaluates Microscaling (MX) data formats that combine a per-block scaling factor with narrow floating-point and integer types for individual elements. MX formats balance the competing needs of hardware efficiency, model accuracy, and user friction. Empirical results on over two dozen benchmarks demonstrate practicality of MX data formats as a drop-in replacement for baseline FP32 for AI inference and training with low user friction. We also show the first instance of training generative language models at sub-8-bit weights, activations, and gradients with minimal accuracy loss and no modifications to the training recipe.

This paper introduces Block Data Representations (BDR), a framework for exploring and evaluating a wide spectrum of narrow-precision formats for deep learning. It enables comparison of popular quantization standards, and through BDR, new formats based on shared microexponents (MX) are identified, which outperform other state-of-the-art quantization approaches, including narrow-precision floating-point and block floating-point. MX utilizes multiple levels of quantization scaling with ultra-fine scaling factors based on shared microexponents in the hardware. The effectiveness of MX is demonstrated on real-world models including large-scale generative pretraining and inferencing, and production-scale recommendation systems.

We introduce a technique that compresses deep neural networks both during and after training by constraining the total number of weights updated during backpropagation to those with the highest total gradients. The remaining weights are forgotten and their initial value is regenerated at every access to avoid storing them in memory. This dramatically reduces the number of off-chip memory accesses during both training and inference, a key component of the energy needs of DNN accelerators. By ensuring that the total weight diffusion remains close to that of baseline unpruned SGD, networks pruned using DropBack are able to maintain high accuracy across network architectures. We observe weight compression of 25x with LeNet-300-100 on MNIST while maintaining accuracy. On CIFAR-10, we see an approximately 5x weight compression on 3 models: an already 9x-reduced VGG-16, Densenet, and WRN-28-10 - all with zero or negligible accuracy loss. On Densenet and WRN, which are particularly challenging to compress, Both Densenet and WRN improve on the state of the art, achieving higher compression with better accuracy than prior pruning techniques.