Mixture of Experts (MoE) models have emerged as a promising approach to scale language models efficiently by activating only a subset of parameters for each input. However, deploying these models under GPU memory constraints remains challenging, as existing offloading strategies incur significant overhead from CPU-GPU data transfers. While prior work has explored prefetching techniques to mitigate this bottleneck, these methods require costly fallback mechanisms when predictions fail. Since expert transfers cannot be canceled once initiated, the correct experts need to be loaded on demand sequentially, introducing additional latency. To address this, we present CommitMoE, a novel approach featuring a Commit Router that makes execution decisions based on expert predictions without fallback mechanisms. Our key insight reveals that router certainty strongly correlates with prediction accuracy, while in low-certainty scenarios, the model output demonstrates inherent robustness to expert selection. Leveraging this insight to design a systems-level solution, CommitMoE achieves 1.3× to 9.4× faster inference across different environments and datasets compared to state-of-the-art offloading frameworks while maintaining model quality.

Large Language Models (LLMs) have demonstrated remarkable capabilities, yet their extensive parameter scales pose significant challenges for practical deployment. Unstructured pruning has emerged as an effective model compression strategy with minimal performance loss, which introduces fine-grained sparsity for weight parameters. While existing methods employ a layer-wise pruning strategy to avoid the complexity of global pruning for billion-scale LLMs, they require appropriate sparsity allocation for the layer-wise pruning objectives and often lead to suboptimal solutions for the overall model. In this paper, we propose Lua-LLM ($\textbf{L}$earning $\textbf{u}$nstructured-sparsity $\textbf{a}$llocation in LLMs), a learning-based global pruning framework that explores the optimal unstructured sparsity allocation. Unlike existing pruning methods, which primarily focus on allocating per-layer sparsity, Lua-LLM achieves flexible allocation for both layer-wise and intra-layer sparsity. Furthermore, Lua-LLM leverages a soft Top-K operator to approximate the importance-based mask selection mechanism, enabling efficient binary mask learning. Experimental results on LLaMA and OPT families demonstrate significant performance improvements over existing methods.

As deep neural networks (DNNs) become increasingly large and complicated, pruning techniques are proposed for lower memory footprint and more efficient inference. The most critical kernel to execute pruned sparse DNNs on GPUs is Sparse-dense Matrix Multiplication (SpMM). To maximize the performance of SpMM, despite the high-performance implementation generated from advanced tensor compilers, they often take a long time to iteratively search tuning configurations. Such a long time slows down the cycle of exploring better DNN architectures or pruning algorithms. In this article, we propose LO-SpMM to efficiently generate high-performance SpMM implementations for sparse DNN inference. Based on the analysis of nonzero elements’ layout, the characterization of the GPU architecture, and a rank-based cost model, LO-SpMM can effectively reduce the search space and eliminate possibly low-performance candidates. Besides, rather than generating complete SpMM implementations for evaluation, LO-SpMM constructs simplified proxies to quickly estimate performance, thereby substantially reducing compilation and execution costs. Experimental results show that LO-SpMM can reduce the search time by 281× at most, while the performance of generated SpMM implementations is comparable to or better than the state-of-the-art sparse tensor compiling solutions.

As deep neural networks (DNNs) become increasingly large and complicated, pruning techniques are proposed for lower memory footprint and more efficient inference. The most critical kernel to execute pruned sparse DNNs on GPUs is Sparse-dense Matrix Multiplication (SpMM). To maximize the performance of SpMM, despite the high-performance code generated from recent tensor compilers, they often take a long time for iteratively searching candidate configurations. Such a long time slows down the cycle of exploring better DNN architectures or pruning algorithms. In this paper, we propose EC-SpMM to efficiently generate high-performance SpMM kernels for sparse DNN inference. Based on the analysis of nonzero elements’ layout, the characterization of GPU architecture, and a rank-based cost model, EC-SpMM can effectively reduce the search space and eliminate possibly low-performance candidates. Experimental results show that EC-SpMM can reduce the compilation time by a factor of 35 ×, while the performance of generated SpMM kernels is comparable or even better, compared with the state-of-the-art sparse tensor compiling solution.