FP8 vision attention output projection
FP8-quantized vision attention output projection that combines multi-head attention outputs back to hidden dimension. Projects concatenated attention outputs from all heads through a linear layer with blockwise FP8 quantization (1x128 for activation, 128x128 for weight).
Reported evidence · last observed 2026-08-20. Reported by source; not independently reproduced. No public source. License unknown.
Current records
Workloadsuite of 16 cases · mean latencyhidden_size = 1280 · num_hidden_blocks = 1016 cases
Not measured on H100 for this workload. Challenges →
117.3µs84% SOL1.00×Reported · license unknown · no source2026-08-20
Measured exactly what you asked. Reported by source; not independently reproduced.
217.6µs83% SOL1.02×Reported · license unknown · no source2026-08-17
Measured exactly what you asked. Reported by source; not independently reproduced.
317.8µs83% SOL1.03×Reported · license unknown · no source2026-08-04
Measured exactly what you asked. Reported by source; not independently reproduced.
418.1µs83% SOL1.04×Reported · license unknown · no source2026-07-31
Measured exactly what you asked. Reported by source; not independently reproduced.
518.4µs83% SOL1.06×Reported · license unknown · no source2026-08-10
Measured exactly what you asked. Reported by source; not independently reproduced.
620.4µs81% SOL1.18×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
720.7µs81% SOL1.20×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
821.1µs80% SOL1.22×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.
923.2µs79% SOL1.34×Reported · license unknown · no source2026-08-04
Measured exactly what you asked. Reported by source; not independently reproduced.
1023.2µs79% SOL1.34×Reported · license unknown · no source2026-08-03
Measured exactly what you asked. Reported by source; not independently reproduced.
1125.1µs77% SOL1.45×Reported · license unknown · no source2026-07-18
Measured exactly what you asked. Reported by source; not independently reproduced.
1228.3µs75% SOL1.63×Reported · license unknown · no source2026-07-25
Measured exactly what you asked. Reported by source; not independently reproduced.
1328.6µs75% SOL1.65×Reported · license unknown · no source2026-07-23
Measured exactly what you asked. Reported by source; not independently reproduced.
1451.6µs61% SOL2.98×Reported · license unknown · no source2026-07-10
Measured exactly what you asked. Reported by source; not independently reproduced.