Accelerating Sparse Deep Neural Networks
arXiv:2104.08378
Abstract
As neural network model sizes have dramatically increased, so has the interest in various techniques to reduce their parameter counts and accelerate their execution. An active area of research in this field is sparsity - encouraging zero values in parameters that can then be discarded from storage or computations. While most research focuses on high levels of sparsity, there are challenges in universally maintaining model accuracy as well as achieving significant speedups over modern matrix-math hardware. To make sparsity adoption practical, the NVIDIA Ampere GPU architecture introduces sparsity support in its matrix-math units, Tensor Cores. We present the design and behavior of Sparse Tensor Cores, which exploit a 2:4 (50%) sparsity pattern that leads to twice the math throughput of dense matrix units. We also describe a simple workflow for training networks that both satisfy 2:4 sparsity pattern requirements and maintain accuracy, verifying it on a wide range of common tasks and model architectures. This workflow makes it easy to prepare accurate models for efficient deployment on Sparse Tensor Cores.
References in corpus (24)
- Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation
- GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium
- MobileNetV2: Inverted Residuals and Linear Bottlenecks
- Pruning Convolutional Neural Networks for Resource Efficient Inference
- SQuAD: 100,000+ Questions for Machine Comprehension of Text
- To prune, or not to prune: exploring the efficacy of pruning for model compression
- Learning Structured Sparsity in Deep Neural Networks
- Channel Pruning for Accelerating Very Deep Neural Networks
- Billion-scale semi-supervised learning for image classification
- Integer Quantization for Deep Learning Inference: Principles and Empirical Evaluation
- Dilated Residual Networks
- Exploring Sparsity in Recurrent Neural Networks
- Rigging the Lottery: Making All Tickets Winners
- Block-Sparse Recurrent Neural Networks
- Accelerator-Aware Pruning for Convolutional Neural Networks
- Balanced Sparsity for Efficient DNN Inference on GPU
- Learning N:M Fine-grained Structured Sparse Neural Networks From Scratch
- Mesh-TensorFlow: Deep Learning for Supercomputers
- Stacked U-Nets: A No-Frills Approach to Natural Image Segmentation
- Accelerated Sparse Neural Training: A Provable and Efficient Method to Find N:M Transposable Masks
- SQuantizer: Simultaneous Learning for Both Sparse and Low-precision Neural Networks
- Campfire: Compressible, Regularization-Free, Structured Sparse Training for Hardware Accelerators
- Structured Pruning for Efficient ConvNets via Incremental Regularization
- Filter Sketch for Network Pruning