CartesianMoE: Boosting Knowledge Sharing among Experts via Cartesian Product Routing in Mixture-of-Experts
This addresses a key bottleneck in scaling MoE models for LLMs, offering incremental improvements in efficiency and performance for AI researchers and practitioners.
The paper tackles the problem of limited knowledge sharing among experts in Mixture-of-Experts (MoE) models for large language models, proposing CartesianMoE to improve this via a multiplication-based routing approach, resulting in better perplexity and downstream task performance with enhanced routing robustness.
Large language models (LLM) have been attracting much attention from the community recently, due to their remarkable performance in all kinds of downstream tasks. According to the well-known scaling law, scaling up a dense LLM enhances its capabilities, but also significantly increases the computational complexity. Mixture-of-Experts (MoE) models address that by allowing the model size to grow without substantially raising training or inference costs. Yet MoE models face challenges regarding knowledge sharing among experts, making their performance somehow sensitive to routing accuracy. To tackle that, previous works introduced shared experts and combined their outputs with those of the top $K$ routed experts in an ``addition'' manner. In this paper, inspired by collective matrix factorization to learn shared knowledge among data, we propose CartesianMoE, which implements more effective knowledge sharing among experts in more like a ``multiplication'' manner. Extensive experimental results indicate that CartesianMoE outperforms previous MoE models for building LLMs, in terms of both perplexity and downstream task performance. And we also find that CartesianMoE achieves better expert routing robustness.