How LatentMoE Reshapes the MoE Paradigm
青稞AI · wechat · 2026-07-19
Using Kimi K3 and NVIDIA Nemotron3 Super as examples, this article explains why LatentMoE could be the key paradigm for the next generation of MoE models.
The core argument is that traditional MoE bottlenecks aren't just about the number of experts. When routing tokens to experts, large-dimension weights must be read and all-to-all communication is required—costs that scale linearly as the model grows. LatentMoE addresses this by first projecting tokens from their original dimension d into a smaller latent space ℓ, performing routing and expert computations in this lower-dimensional space, and then projecting back. This allows it to:
- Reduce expert weight reads from d×m to ℓ×m
- Proportionally decrease all-to-all communication volume
- Activate more experts and expand expert combination space under the same inference cost
The article details two case studies:
- Nemotron3 Super: Features 512 experts, Top-22 routing, ℓ = 1024, and combines Mamba-2, minimal attention, and MTP for inference acceleration.
- Kimi K3: Features 896 experts, Top-16 routing, and an extremely high sparsity ratio. It introduces QuantileBalancing—using quantiles instead of heuristic hyperparameters for load balancing—to maintain training stability at a 2.8T scale.
The author concludes that while NVIDIA leans towards "efficiency first" and Kimi pushes for "extreme scale," both are essentially compressing expert computation dimensions to trade saved resources for greater expert combinatorial capacity. The article also looks ahead to adaptive compression ratios, dynamic Top-K routing, and incorporating structured sparsity into architectural design.
More from Infra
- AWS on agent prompt caching: explicit + implicit patterns to slash token costs — AWS ML Blog · 2026-10-27
- Thieves spoofing GPS steal $100M+ in data center cargo since June — Ars Technica AI · 2026-10-27
- Lam Research: weekly token usage up ~250x in 18 months — Beth_Kindig · 2026-10-27
- AWS shows how to build enterprise agentic retrieval on Bedrock with CDK — AWS ML Blog · 2026-10-27
- Encrypted GPT-2 speedrun challenges FHE inference: best is 3.031 s/token, goal sub-1s — tensorqt · 2026-10-27
- Open-source Latch compiles PyTorch into training megakernels, 2.75x faster on H100 — sguada · 2026-10-27