Mixture of experts (MoE) models achieve state-of-the-art results in language modeling but suffer from inefficient hardware utilization due to imbalanced token routing and communication overhead. While prior work has focused on optimizing MoE training and decoder architectures, inference for encoder-based MoE models in a multi-GPU with expert parallelism setting remains underexplored. We introduce MoEShard, an inference system that achieves perfect load balancing through tensor sharding of MoE experts. Unlike existing approaches that rely on heuristic capacity factors or drop tokens, MoEShard evenly distributes computation across GPUs and ensures full token retention, maximizing utilization regardless of routing skewness. We achieve this through a strategic row- and column-wise decomposition of expert matrices. This reduces idle time and avoids bottlenecks caused by imbalanced expert assignments. Furthermore, MoEShard minimizes kernel launches by fusing decomposed expert computations, significantly improving throughput. We evaluate MoEShard against DeepSpeed on encoder-based architectures, demonstrating speedups of up to 6.4$\times$ in time to first token (TTFT). Our results show that tensor sharding, when properly applied to experts, is a viable and effective strategy for efficient MoE inference.

Recent large language models (LLMs) have tended to leverage sparsity to reduce computations, employing the sparsely activated mixture-of-experts (MoE) technique. MoE introduces four modules, including token routing, token communication, expert computation, and expert parallelism, that impact model quality and training efficiency. To enable versatile usage of MoE models, we introduce FSMoE, a flexible training system optimizing task scheduling with three novel techniques: 1) Unified abstraction and online profiling of MoE modules for task scheduling across various MoE implementations. 2) Co-scheduling intra-node and inter-node communications with computations to minimize communication overheads. 3) To support near-optimal task scheduling, we design an adaptive gradient partitioning method for gradient aggregation and a schedule to adaptively pipeline communications and computations. We conduct extensive experiments with configured MoE layers and real-world MoE models on two GPU clusters. Experimental results show that 1) our FSMoE supports four popular types of MoE routing functions and is more efficient than existing implementations (with up to a 1.42$\times$ speedup), and 2) FSMoE outperforms the state-of-the-art MoE training systems (DeepSpeed-MoE and Tutel) by 1.18$\times$-1.22$\times$ on 1458 MoE layers and 1.19$\times$-3.01$\times$ on real-world MoE models based on GPT-2 and Mixtral using a popular routing function.

Mixture-of-Experts (MoE) architectures have recently gained popularity in enabling efficient scaling of large language models. However, we uncover a fundamental tension: while MoEs are designed for selective expert activation, production serving requires request batching, which forces the activation of all experts and negates MoE's efficiency benefits during the decode phase. We present Lynx, a system that enables efficient MoE inference through dynamic, batch-aware expert selection. Our key insight is that expert importance varies significantly across tokens and inference phases, creating opportunities for runtime optimization. Lynx leverages this insight through a lightweight framework that dynamically reduces active experts while preserving model accuracy. Our evaluations show that Lynx achieves up to 1.55x reduction in inference latency while maintaining negligible accuracy loss from baseline model across complex code generation and mathematical reasoning tasks.

The Mixture of Experts (MoE) is an advanced model architecture in the industry that combines multiple specialized expert models from various domains into a single supermodel. This approach enables the model to scale without significantly increasing the computational costs of training and inference, while maximizing model performance. However, current distributed training frameworks do not consider the ultimate optimization of communication, especially for large base models. This paper proposes a network-traffic-aware parallel optimization method that selects the optimal parallel strategy based on the communication volume, and the training cluster's inter-node and intra-node network topologies. Compared to the DeepSpeed, MoNTA achieves an 8x increase in AllToAll communication performance under 8-card tensor parallelism. Compared to the baseline, training a 2x70B model using 16 A800 cards, with an 8K sequence, results in a 13% overall latency performance improvement. Project Page: https://github.com/EnflameTechnology/DeepSpeed.

Mixture-of-Experts (MoE) large language models (LLM) have memory requirements that often exceed the GPU memory capacity, requiring costly parameter movement from secondary memories to the GPU for expert computation. In this work, we present Mixture of Near-Data Experts (MoNDE), a near-data computing solution that efficiently enables MoE LLM inference. MoNDE reduces the volume of MoE parameter movement by transferring only the $\textit{hot}$ experts to the GPU, while computing the remaining $\textit{cold}$ experts inside the host memory device. By replacing the transfers of massive expert parameters with the ones of small activations, MoNDE enables far more communication-efficient MoE inference, thereby resulting in substantial speedups over the existing parameter offloading frameworks for both encoder and decoder operations.