Group limited topk routing
Group-limited top-k expert routing for Mixture of Experts. Implements a two-stage selection: first organizes 256 experts into 8 groups, computes group scores by summing top-2 expert scores within each group, selects top-4 groups, then picks top-8 experts from selected groups. Returns selected expert indices and normalized routing weights scaled by routed_scaling_factor.
Reported evidence · last observed 2026-08-01. Reported by source; not independently reproduced. No public source. License unknown.
Current records
Workloadsuite of 16 cases · mean latencyn_group = 8 · gating_dim = 4096 · topk_group = 4 · num_experts = 256 · experts_per_group = 32 · num_experts_per_tok = 816 cases
Not measured on H100 for this workload. Challenges →
121.1µs88% SOL1.00×Reported · license unknown · no source2026-08-01
Measured exactly what you asked. Reported by source; not independently reproduced.
221.5µs88% SOL1.02×Reported · license unknown · no source2026-08-15
Measured exactly what you asked. Reported by source; not independently reproduced.
321.6µs88% SOL1.02×Reported · license unknown · no source2026-08-15
Measured exactly what you asked. Reported by source; not independently reproduced.
421.6µs88% SOL1.02×Reported · license unknown · no source2026-08-15
Measured exactly what you asked. Reported by source; not independently reproduced.
521.6µs88% SOL1.02×Reported · license unknown · no source2026-07-23
Measured exactly what you asked. Reported by source; not independently reproduced.
621.8µs88% SOL1.03×Reported · license unknown · no source2026-07-17
Measured exactly what you asked. Reported by source; not independently reproduced.
722.8µs87% SOL1.08×Reported · license unknown · no source2026-07-16
Measured exactly what you asked. Reported by source; not independently reproduced.
822.9µs87% SOL1.08×Reported · license unknown · no source2026-07-25
Measured exactly what you asked. Reported by source; not independently reproduced.
924.2µs86% SOL1.14×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
1024.6µs86% SOL1.16×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
1127.9µs84% SOL1.32×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
1228.2µs84% SOL1.34×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
1355.3µs71% SOL2.62×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
14148.9µs46% SOL7.04×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.