SliceMoE: Routing Embedding Slices Instead of Tokens for Fine-Grained and Balanced Transformer Scaling
This addresses inefficiencies in scaling transformers for language modeling and translation tasks, offering a more fine-grained and balanced approach, though it is an incremental improvement over existing MoE methods.
The paper tackled the problem of capacity bottlenecks and load-balancing issues in token-level routing for Mixture-of-Experts (MoE) layers in transformers by introducing SliceMoE, which routes slices of token embeddings instead of whole tokens, resulting in up to 1.7x faster inference, 12-18% lower perplexity than token-MoE, and improved expert balance.
Mixture-of-Experts (MoE) layers scale transformers by routing tokens to a sparse subset of feed-forward experts. Token-level routing, however, assigns an entire semantic spectrum to each expert, creating capacity bottlenecks, load-balancing pathologies, and limited specialization. We introduce SliceMoE, an architecture that routes contiguous slices of a token's hidden vector. A d-dimensional embedding is partitioned into S slices, and for each slice, a lightweight shared router predicts the top-k experts. Experts operate on their assigned slices independently, and outputs are reassembled, maintaining per-token FLOP efficiency. Because slices from different tokens interleave within an expert, utilization is naturally smoother. We propose a slice-level capacity loss, cross-slice dropout, and efficient fused batched GEMM kernels. Experiments on WikiText-103 language modeling, WMT En-De translation, and three text-classification datasets show SliceMoE attains up to 1.7x faster inference than dense baselines, 12 to 18 percent lower perplexity than parameter-matched token-MoE, and improved expert balance, with interpretable expertise over syntactic versus semantic subspaces.