LatentMoE: Toward Optimal Accuracy per FLOP and Parameter in Mixture of Experts
This work addresses the efficiency of MoEs in large language models, which is crucial for reducing computational costs in both offline and online inference scenarios, representing an incremental improvement in model design.
The paper tackled the problem of optimizing Mixture of Experts (MoE) architectures for inference cost, measured by accuracy per FLOP and per parameter, and introduced LatentMoE, which consistently outperforms standard MoEs in these metrics based on empirical exploration up to 95B parameters and 1T tokens.
Mixture of Experts (MoEs) have become a central component of many state-of-the-art open-source and proprietary large language models. Despite their widespread adoption, it remains unclear how close existing MoE architectures are to optimal with respect to inference cost, as measured by accuracy per floating-point operation and per parameter. In this work, we revisit MoE design from a hardware-software co-design perspective, grounded in empirical and theoretical considerations. We characterize key performance bottlenecks across diverse deployment regimes, spanning offline high-throughput execution and online, latency-critical inference. Guided by these insights, we introduce LatentMoE, a new model architecture resulting from systematic design exploration and optimized for maximal accuracy per unit of compute. Empirical design space exploration at scales of up to 95B parameters and over a 1T-token training horizon, together with supporting theoretical analysis, shows that LatentMoE consistently outperforms standard MoE architectures in terms of accuracy per FLOP and per parameter. Given its strong performance, the LatentMoE architecture has been adopted by the flagship Nemotron-3 Super and Ultra models and scaled to substantially larger regimes, including longer token horizons and larger model sizes, as reported in Nvidia et al. (arXiv:2512.20856).