BWhat changed
Projected routed payloads and expert computation into a lower-dimensional latent space while retaining routing decisions in the original representation, trading communication and weight traffic for expert capacity.
Scientific scope
Projected routed payloads and expert computation into a lower-dimensional latent space while retaining routing decisions in the original representation, trading communication and weight traffic for expert capacity. Method-level design distinct from later Nemotron NVFP4 training/system integration. Accuracy-cost evidence is source-reported; trillion-scale serving gains include projections, not measured universal speedups. Excess compression can degrade quality.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Method-level design distinct from later Nemotron NVFP4 training/system integration. Accuracy-cost evidence is source-reported; trillion-scale serving gains include projections, not measured universal speedups. Excess compression can degrade quality.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
- LatentMoE: Toward Optimal Accuracy per FLOP and Parameter in Mixture of Experts
- Elango et al.: LatentMoE compressed routed-expert computation — official method source
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction