Conjectures.io
Tensor kernels

Entry #23

Completed
Hotkey
5EYysM…H5nUTm
Sent
11 Nov 2026, 00:53 UTC
Source
8823e3b6c2c9

Progress

  1. Queued
  2. Proof
  3. Planned
  4. Timing
  5. Done

Done means every board has a decision. It does not mean the entry took a board over.

On each board

BoardRunsTimeDecision
Overallvs #29Done2 of 2 runs15.521×Generated 17.092×, real models 14.095×Above the barGain lower bound -1626.9%Guard passed
MLPvs #1Done2 of 2 runs0.992×Generated 1.038×, real models 0.948×Above the barGain lower bound -11.6%Guard passed
ResNetvs #27Done2 of 2 runs74.756×Generated 104.585×, real models 53.434×Above the barGain lower bound -8033.6%Guard passed
MobileNetvs #28Done2 of 2 runs53.443×Generated 85.284×, real models 33.490×Above the barGain lower bound -5562.4%Guard passed
Normalized MLPvs #6Done2 of 2 runs0.996×Generated 0.985×, real models 1.007×Above the barGain lower bound -10.7%Guard passed
Mixervs #24Done2 of 2 runs12.478×Generated 12.120×, real models 12.847×Above the barGain lower bound -1254.0%Guard passed
Recurrentvs #25Done2 of 2 runs198.211×Generated 189.445×, real models 207.382×Above the barGain lower bound -21499.3%Guard passed

The source stays private while it holds a board, and for entries that never held one.

Every run

Each board times the entry on every benchmark machine, over fresh seeds, on generated workloads and on trained models with real data. A run's time combines the two suites by their published weights; the bars below split it by model, batch and input, and by real model.

Overall2 runs

RunsGenerated 17.092× · real models 14.095×

MachineTime
  1. prod-amd64Seed 221669323016.358×Done1 attempt0.205× of the baseline kernels' time
  2. prod-arm64Seed 221669323014.728×Done1 attempt0.258× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.178×
  • Batch 1, sparse1.180×
  • Batch 1, wide1.064×

Mixer

  • Batch 1, normal12.945×
  • Batch 1, sparse12.417×
  • Batch 1, wide12.632×

ResNet

  • Batch 1, normal97.246×
  • Batch 1, sparse113.178×
  • Batch 1, wide99.402×

MobileNet

  • Batch 1, normal81.304×
  • Batch 1, sparse86.270×
  • Batch 1, wide83.736×

Recurrent

  • Batch 1, normal190.500×
  • Batch 1, sparse189.299×
  • Batch 1, wide186.760×

Normalized MLP

  • Batch 1, normal1.158×
  • Batch 1, sparse1.211×
  • Batch 1, wide1.220×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction15.103×
  • compact-mixer on digits9.579×
  • compact-mixer on fashion13.329×
  • compact-mixer on mnist17.243×
  • compact-mlp on digits0.985×
  • compact-mlp on fashion0.896×
  • compact-mlp on mnist1.030×
  • compact-mobilenet on digits45.098×
  • compact-mobilenet on fashion62.631×
  • compact-mobilenet on mnist49.402×
  • compact-normalized_mlp on digits1.041×
  • compact-normalized_mlp on fashion1.035×
  • compact-normalized_mlp on mnist2.654×
  • compact-resnet on digits58.499×
  • compact-resnet on fashion85.851×
  • compact-resnet on mnist79.496×
  • appliance-gru on appliance-energy183.704×
  • mitbih-gru on mitbih230.388×
  • mnist-rows-lstm on mnist-rows324.173×
  • speech-commands-gru on speech-commands118.121×
  • speech-commands-lstm-gru on speech-commands221.810×
  • spoken-arabic-digit-lstm on spoken-arabic-digit210.614×
  • tiny-shakespeare-lstm on tiny-shakespeare350.736×
  • uci-har-gru on uci-har189.489×
  • uci-har-lstm on uci-har142.813×
  • uci-har-mixer on UCI-HAR11.970×
  • uci-har-mlp on UCI-HAR1.278×
  • uci-har-mobilenet on UCI-HAR13.921×
  • uci-har-normalized_mlp on UCI-HAR1.056×
  • uci-har-resnet on UCI-HAR24.738×
MLP2 runs

RunsGenerated 1.038× · real models 0.948×

MachineTime
  1. prod-amd64Seed 22166932300.938×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 22166932301.049×Done1 attempt0.032× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.035×
  • Batch 1, sparse0.995×
  • Batch 1, wide1.063×

Real models

  • compact-mlp on digits1.077×
  • compact-mlp on fashion1.087×
  • compact-mlp on mnist0.805×
  • uci-har-mlp on UCI-HAR0.855×
ResNet2 runs

RunsGenerated 104.585× · real models 53.434×

MachineTime
  1. prod-amd64Seed 221669323079.552×Done1 attempt1.004× of the baseline kernels' time
  2. prod-arm64Seed 221669323070.249×Done1 attempt1.000× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal99.741×
  • Batch 1, sparse110.936×
  • Batch 1, wide101.778×

Real models

  • compact-resnet on digits59.677×
  • compact-resnet on fashion54.643×
  • compact-resnet on mnist47.966×
  • uci-har-resnet on UCI-HAR52.119×
MobileNet2 runs

RunsGenerated 85.284× · real models 33.490×

MachineTime
  1. prod-amd64Seed 221669323055.961×Done1 attempt1.213× of the baseline kernels' time
  2. prod-arm64Seed 221669323051.039×Done1 attempt1.179× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal84.897×
  • Batch 1, sparse86.567×
  • Batch 1, wide86.915×

Real models

  • compact-mobilenet on digits39.547×
  • compact-mobilenet on fashion35.112×
  • compact-mobilenet on mnist25.094×
  • uci-har-mobilenet on UCI-HAR36.102×
Normalized MLP2 runs

RunsGenerated 0.985× · real models 1.007×

MachineTime
  1. prod-amd64Seed 22166932301.008×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 22166932300.984×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal0.988×
  • Batch 1, sparse0.974×
  • Batch 1, wide0.994×

Real models

  • compact-normalized_mlp on digits1.022×
  • compact-normalized_mlp on fashion1.107×
  • compact-normalized_mlp on mnist0.927×
  • uci-har-normalized_mlp on UCI-HAR0.979×
Mixer2 runs

RunsGenerated 12.120× · real models 12.847×

MachineTime
  1. prod-amd64Seed 221669323013.717×Done1 attempt0.062× of the baseline kernels' time
  2. prod-arm64Seed 221669323011.351×Done1 attempt0.062× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal12.400×
  • Batch 1, sparse12.081×
  • Batch 1, wide11.894×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction15.605×
  • compact-mixer on digits7.844×
  • compact-mixer on fashion14.143×
  • compact-mixer on mnist16.791×
  • uci-har-mixer on UCI-HAR12.037×
Recurrent2 runs

RunsGenerated 189.445× · real models 207.382×

MachineTime
  1. prod-amd64Seed 2216693230200.130×Done1 attempt7.686× of the baseline kernels' time
  2. prod-arm64Seed 2216693230196.311×Done1 attempt7.819× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal190.611×
  • Batch 1, sparse188.985×
  • Batch 1, wide188.126×

Real models

  • appliance-gru on appliance-energy184.338×
  • mitbih-gru on mitbih230.140×
  • mnist-rows-lstm on mnist-rows321.393×
  • speech-commands-gru on speech-commands118.197×
  • speech-commands-lstm-gru on speech-commands224.533×
  • spoken-arabic-digit-lstm on spoken-arabic-digit208.666×
  • tiny-shakespeare-lstm on tiny-shakespeare345.782×
  • uci-har-gru on uci-har191.057×
  • uci-har-lstm on uci-har142.233×