Pre-training Transformer models is resource-intensive, and recent studies have shown that sign momentum is an efficient technique for training large-scale deep learning models, particularly Transformers. However, its application in distributed training or federated learning remains underexplored. This paper investigates a novel communication-efficient distributed sign momentum method with local updates. Our proposed method allows for a broad class of base optimizers for local updates, and uses sign momentum in global updates, where momentum is generated from differences accumulated during local steps. We evaluate our method on the pre-training of various GPT-2 models, and the empirical results show significant improvement compared to other distributed methods with local updates. Furthermore, by approximating the sign operator with a randomized version that acts as a continuous analog in expectation, we present an $O(1/\sqrt{T})$ convergence for one instance of the proposed method for nonconvex smooth functions.

Mixture-of-Experts (MoE) architectures face challenges such as high memory consumption and redundancy in experts. Pruning MoE can reduce network weights while maintaining model performance. Motivated by the recent observation of emergent large magnitude features in Large Language Models (LLM) and MoE routing policy, we propose MoE-Pruner, a method that prunes weights with the smallest magnitudes multiplied by the corresponding input activations and router weights, on each output neuron. Our pruning method is one-shot, requiring no retraining or weight updates. We evaluate our method on Mixtral-8x7B and Mixtral-8x22B across multiple language benchmarks. Experimental results show that our pruning method significantly outperforms state-of-the-art LLM pruning methods. Furthermore, our pruned MoE models can benefit from a pretrained teacher model through expert-wise knowledge distillation, improving performance post-pruning. Experimental results demonstrate that the Mixtral-8x7B model with 50% sparsity maintains 99% of the performance of the original model after the expert-wise knowledge distillation.

The privacy-preserving federated learning for vertically partitioned data has shown promising results as the solution of the emerging multi-party joint modeling application, in which the data holders (such as government branches, private finance and e-business companies) collaborate throughout the learning process rather than relying on a trusted third party to hold data. However, existing federated learning algorithms for vertically partitioned data are limited to synchronous computation. To improve the efficiency when the unbalanced computation/communication resources are common among the parties in the federated learning system, it is essential to develop asynchronous training algorithms for vertically partitioned data while keeping the data privacy. In this paper, we propose an asynchronous federated SGD (AFSGD-VP) algorithm and its SVRG and SAGA variants on the vertically partitioned data. Moreover, we provide the convergence analyses of AFSGD-VP and its SVRG and SAGA variants under the condition of strong convexity. We also discuss their model privacy, data privacy, computational complexities and communication costs. To the best of our knowledge, AFSGD-VP and its SVRG and SAGA variants are the first asynchronous federated learning algorithms for vertically partitioned data. Extensive experimental results on a variety of vertically partitioned datasets not only verify the theoretical results of AFSGD-VP and its SVRG and SAGA variants, but also show that our algorithms have much higher efficiency than the corresponding synchronous algorithms.

Although the distributed machine learning methods show the potential for the speed-up of training large deep neural networks, the communication cost has been the notorious bottleneck to constrain the performance. To address this challenge, the gradient compression based communication-efficient distributed learning methods were designed to reduce the communication cost, and more recently the local error feedback was incorporated to compensate for the performance loss. However, in this paper, we will show the "gradient mismatch" problem of the local error feedback in centralized distributed training and this issue can lead to degraded performance compared with full-precision training. To solve this critical problem, we propose two novel techniques: 1) step ahead; 2) error averaging. Both our theoretical and empirical results show that our new methods can alleviate the "gradient mismatch" problem. Experiments show that we can even train \textbf{faster with compressed gradient} than full-precision training \textbf{regarding training epochs}.

The backpropagation algorithm is the most popular algorithm training neural networks nowadays. However, it suffers from the forward locking, backward locking and update locking problems, especially when a neural network is so large that its layers are distributed across multiple devices. Existing solutions either can only handle one locking problem or lead to severe accuracy loss or memory inefficiency. Moreover, none of them consider the straggler problem among devices. In this paper, we propose Layer-wise Staleness and a novel efficient training algorithm, Diversely Stale Parameters (DSP), which can address all these challenges without loss of accuracy nor memory issue. We also analyze the convergence of DSP with two popular gradient-based methods and prove that both of them are guaranteed to converge to critical points for non-convex problems. Finally, extensive experimental results on training deep convolutional neural networks demonstrate that our proposed DSP algorithm can achieve significant training speedup with stronger robustness and better generalization than compared methods.