Search papers, labs, and topics across Lattice.
This paper introduces expert upcycling, a novel method for progressively expanding the capacity of Mixture-of-Experts (MoE) models during continued pre-training, which addresses the high memory and communication costs associated with training large MoEs. By duplicating existing experts and extending the router while maintaining fixed top-K routing, the approach enables a warm initialization that significantly reduces loss compared to random initialization. The results demonstrate that the upcycled models can achieve validation loss comparable to fixed-size baselines while saving 32% in GPU hours, offering a practical solution for scaling MoE architectures efficiently.
Upcycling MoE models can achieve the same performance as larger fixed-size models while cutting GPU costs by 32%.
Mixture-of-Experts (MoE) has become the dominant architecture for scaling large language models: frontier models routinely decouple total parameters from per-token computation through sparse expert routing. Scaling laws show that under fixed active computation, model quality scales predictably with total parameters, and MoEs realize this by increasing expert count. However, training large MoEs is expensive, as memory requirements and inter-device communication both scale with total parameter count. We propose expert upcycling, a method for progressively expanding MoE capacity by increasing the number of experts during continued pre-training (CPT). Given a trained E-expert model, the upcycling operator constructs an mE-expert model through expert duplication and router extension while holding top-K routing fixed, preserving per-token inference cost. Duplication provides a warm initialization: the expanded model inherits the source checkpoint's learned representations, starting from a substantially lower loss than random initialization. Subsequent CPT then breaks the symmetry among duplicated experts to drive specialization. We formalize the upcycling operator and develop a theoretical framework decomposing the quality gap into a capacity term and an initialization term. We further introduce utility-based expert selection, which uses gradient-based importance scores to guide non-uniform duplication, more than tripling gap closure when CPT is limited. In our 7B-13B total parameter experiments, the upcycled model matches the fixed-size baseline on validation loss while saving 32% of GPU hours. Comprehensive ablations across model scales, activation ratios, MoE architectures, and training budgets yield a practical recipe for deploying expert upcycling, establishing it as a principled, compute-efficient alternative to training large MoE models from scratch.
Amazon Science