FP8 shared expert MLP
FP8-quantized shared expert MLP with SwiGLU activation. Processes ALL tokens in MoE layers with blockwise FP8 quantization (1x128 for activations, 128x128 for weights). Architecture: x -> [gate_proj, up_proj] -> silu(gate) * up -> down_proj -> output.
Reported evidence · last observed 2026-08-23. Reported by source; not independently reproduced. No public source. License unknown.
Current records
Workloadsuite of 16 cases · mean latencyhidden_size = 7168 · hidden_blocks = 56 · intermediate_size = 2048 · intermediate_blocks = 1616 cases
Not measured on H100 for this workload. Challenges →
1167.2µs98% SOL1.00×Reported · license unknown · no source2026-08-23
Measured exactly what you asked. Reported by source; not independently reproduced.
2171.8µs98% SOL1.03×Reported · license unknown · no source2026-08-01
Measured exactly what you asked. Reported by source; not independently reproduced.
3193.5µs97% SOL1.16×Reported · license unknown · no source2026-08-01
Measured exactly what you asked. Reported by source; not independently reproduced.
4213.7µs97% SOL1.28×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
5216.4µs97% SOL1.29×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
6266.2µs96% SOL1.59×Reported · license unknown · no source2026-08-19
Measured exactly what you asked. Reported by source; not independently reproduced.
7267.0µs96% SOL1.60×Reported · license unknown · no source2026-08-22
Measured exactly what you asked. Reported by source; not independently reproduced.
8270.6µs96% SOL1.62×Reported · license unknown · no source2026-08-18
Measured exactly what you asked. Reported by source; not independently reproduced.
9274.9µs96% SOL1.64×Reported · license unknown · no source2026-08-04
Measured exactly what you asked. Reported by source; not independently reproduced.
10277.8µs96% SOL1.66×Reported · license unknown · no source2026-08-18
Measured exactly what you asked. Reported by source; not independently reproduced.
11277.9µs96% SOL1.66×Reported · license unknown · no source2026-08-04
Measured exactly what you asked. Reported by source; not independently reproduced.
12282.2µs96% SOL1.69×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
13325.6µs95% SOL1.95×Reported · license unknown · no source2026-08-04
Measured exactly what you asked. Reported by source; not independently reproduced.
14409.2µs94% SOL2.45×Reported · license unknown · no source2026-08-02
Measured exactly what you asked. Reported by source; not independently reproduced.
15425.3µs93% SOL2.54×Reported · license unknown · no source2026-07-18
Measured exactly what you asked. Reported by source; not independently reproduced.
16964.5µs85% SOL5.77×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
175.49ms50% SOL32.83×Reported · license unknown · no source2026-08-19
Measured exactly what you asked. Reported by source; not independently reproduced.