Higher-order pruning enables better compression of large language models
Higher-order pruning of experts in mixture-of-experts language models
Machine LearningArtificial Intelligence
Summary
Large language models with many parts called experts sometimes use too much memory. The authors show that by considering how these experts work together, rather than separately, they can pick which experts to remove with less harm to the model’s ability. Their new method, called HOPE, works better especially when cutting the model size in half and on complex tasks like coding and following instructions. This lets big models run more efficiently without losing much performance.
What this means in practice
- •For machine learning engineers: Reduce memory usage of large expert-based language models while maintaining accuracy on complex tasks like coding and instruction following.
- •For natural language processing teams: Improve runtime efficiency of large-scale MoE models in applications requiring diverse and cooperative expert contributions.
Authors
Alex M. Tseng, Prannay Kaul, Luca Zancato, Wei Xia, Stefano Soatto
Abstract
Mixture-of-Experts (MoE) language models suffer from large parameter counts, which create a significant memory bottleneck. Expert pruning is the most direct approach for reducing this parameter count, yet existing methods make pruning decisions for each expert independently, and assume experts' contributions are purely additive. In reality, expert usage in MoEs is inherently cooperative. We derive HOPE (Higher-Order Pruning of Experts), a second-order pruning objective which provably minimizes an upper bound on the error resulting from pruning. We show that REAP (a state-of-the-art first-order pruning method) is a special case of HOPE where interaction terms are ignored. Across three frontier MoE models (up to 122B parameters), two distinct calibration sets, and multiple benchmarks (including math, instruction following, coding, and an agentic suite), we demonstrate that HOPE produces better pruning decisions than existing methods, and its advantage is most pronounced at high pruning rates and on challenging agentic workloads. At 50% pruning, HOPE outperforms all baselines and achieves an average rank of 1.58 out of 5 methods (versus 2.42 for the next-best method, REAP), with gains of up to +6.1% on agentic coding. Over all conditions, HOPE again achieves the best average rank and surpasses every other method in the majority of head-to-head comparisons. By preserving cooperative expert structure that first-order methods ignore, HOPE enables aggressive compression with minimal degradation, particularly on complex tasks where diverse expert combinations are invoked over long sequences.