Bi-directional Masks for Efficient N:M Sparse Training
This work addresses training efficiency for sparse neural networks, particularly in scenarios requiring hardware acceleration, but it is incremental as it builds on existing N:M sparsity methods.
The paper tackles the dense backward propagation issue in training N:M fine-grained sparsity, which limits efficiency despite hardware support, by introducing Bi-directional Masks (Bi-Mask) that use separate masks for forward and backward propagation to accelerate training and maintain performance through weight row permutation, achieving superior performance compared to existing methods that enable backward acceleration and performing on par or better than those that do not.
We focus on addressing the dense backward propagation issue for training efficiency of N:M fine-grained sparsity that preserves at most N out of M consecutive weights and achieves practical speedups supported by the N:M sparse tensor core. Therefore, we present a novel method of Bi-directional Masks (Bi-Mask) with its two central innovations in: 1) Separate sparse masks in the two directions of forward and backward propagation to obtain training acceleration. It disentangles the forward and backward weight sparsity and overcomes the very dense gradient computation. 2) An efficient weight row permutation method to maintain performance. It picks up the permutation candidate with the most eligible N:M weight blocks in the backward to minimize the gradient gap between traditional uni-directional masks and our bi-directional masks. Compared with existing uni-directional scenario that applies a transposable mask and enables backward acceleration, our Bi-Mask is experimentally demonstrated to be more superior in performance. Also, our Bi-Mask performs on par with or even better than methods that fail to achieve backward acceleration. Project of this paper is available at \url{https://github.com/zyxxmu/Bi-Mask}.