Conjectures.io
Tensor kernels

Entry #106

Completed
Hotkey
5E6mCx…etusQN
Sent
12 Nov 2026, 12:16 UTC
Source
b0748216c2f7

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 #50Done2 of 2 runs23.319×Generated 24.413×, real models 22.273×Above the barGain lower bound -2587.4%Guard passed
MLPvs #44Done2 of 2 runs0.929×Generated 1.054×, real models 0.818×Not significantGain lower bound -24.8%Guard passed
ResNetvs #51Done2 of 2 runs109.632×Generated 135.401×, real models 88.767×Above the barGain lower bound -13131.1%Guard passed
MobileNetvs #53Done2 of 2 runs103.103×Generated 128.561×, real models 82.686×Above the barGain lower bound -12527.5%Guard passed
Normalized MLPvs #48Done2 of 2 runs1.023×Generated 0.958×, real models 1.093×Above the barGain lower bound -17.7%Guard passed
Mixervs #58Done2 of 2 runs28.607×Generated 27.550×, real models 29.705×Above the barGain lower bound -3252.5%Guard passed
Recurrentvs #57Done2 of 2 runs913.290×Generated 787.407×, real models 1059.298×Above the barGain lower bound -95874.4%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 24.413× · real models 22.273×

MachineTime
  1. prod-amd64Seed 298017177123.769×Done1 attempt0.220× of the baseline kernels' time
  2. prod-arm64Seed 298017177122.877×Done1 attempt0.259× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.829×
  • Batch 1, sparse0.986×
  • Batch 1, wide1.180×

Mixer

  • Batch 1, normal23.852×
  • Batch 1, sparse23.153×
  • Batch 1, wide24.082×

ResNet

  • Batch 1, normal113.540×
  • Batch 1, sparse135.049×
  • Batch 1, wide115.090×

MobileNet

  • Batch 1, normal93.399×
  • Batch 1, sparse96.450×
  • Batch 1, wide93.582×

Recurrent

  • Batch 1, normal689.432×
  • Batch 1, sparse661.046×
  • Batch 1, wide641.227×

Normalized MLP

  • Batch 1, normal1.235×
  • Batch 1, sparse1.385×
  • Batch 1, wide1.121×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction18.913×
  • compact-mixer on digits10.826×
  • compact-mixer on fashion36.411×
  • compact-mixer on mnist39.546×
  • compact-mlp on digits0.944×
  • compact-mlp on fashion0.986×
  • compact-mlp on mnist0.450×
  • compact-mobilenet on digits66.472×
  • compact-mobilenet on fashion81.923×
  • compact-mobilenet on mnist77.800×
  • compact-normalized_mlp on digits0.542×
  • compact-normalized_mlp on fashion1.799×
  • compact-normalized_mlp on mnist0.735×
  • compact-resnet on digits72.461×
  • compact-resnet on fashion118.546×
  • compact-resnet on mnist117.395×
  • appliance-gru on appliance-energy753.893×
  • mitbih-gru on mitbih900.320×
  • mnist-rows-lstm on mnist-rows710.063×
  • speech-commands-gru on speech-commands701.325×
  • speech-commands-lstm-gru on speech-commands857.776×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1141.277×
  • tiny-shakespeare-lstm on tiny-shakespeare818.629×
  • uci-har-gru on uci-har849.304×
  • uci-har-lstm on uci-har1161.288×
  • uci-har-mixer on UCI-HAR53.340×
  • uci-har-mlp on UCI-HAR1.016×
  • uci-har-mobilenet on UCI-HAR77.610×
  • uci-har-normalized_mlp on UCI-HAR0.857×
  • uci-har-resnet on UCI-HAR80.618×
MLP2 runs

RunsGenerated 1.054× · real models 0.818×

MachineTime
  1. prod-amd64Seed 29801717711.013×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 29801717710.851×Done1 attempt0.019× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.117×
  • Batch 1, sparse1.062×
  • Batch 1, wide1.136×

Real models

  • compact-mlp on digits0.420×
  • compact-mlp on fashion0.941×
  • compact-mlp on mnist1.091×
  • uci-har-mlp on UCI-HAR1.039×
ResNet2 runs

RunsGenerated 135.401× · real models 88.767×

MachineTime
  1. prod-amd64Seed 2980171771105.798×Done1 attempt0.983× of the baseline kernels' time
  2. prod-arm64Seed 2980171771113.605×Done1 attempt1.004× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal147.526×
  • Batch 1, sparse130.107×
  • Batch 1, wide111.731×

Real models

  • compact-resnet on digits63.768×
  • compact-resnet on fashion118.249×
  • compact-resnet on mnist117.742×
  • uci-har-resnet on UCI-HAR69.933×
MobileNet2 runs

RunsGenerated 128.561× · real models 82.686×

MachineTime
  1. prod-amd64Seed 298017177190.253×Done1 attempt1.167× of the baseline kernels' time
  2. prod-arm64Seed 2980171771117.782×Done1 attempt1.157× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal129.882×
  • Batch 1, sparse97.249×
  • Batch 1, wide134.892×

Real models

  • compact-mobilenet on digits62.716×
  • compact-mobilenet on fashion95.970×
  • compact-mobilenet on mnist93.400×
  • uci-har-mobilenet on UCI-HAR83.151×
Normalized MLP2 runs

RunsGenerated 0.958× · real models 1.093×

MachineTime
  1. prod-amd64Seed 29801717711.105×Done1 attempt0.007× of the baseline kernels' time
  2. prod-arm64Seed 29801717710.947×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.073×
  • Batch 1, sparse0.786×
  • Batch 1, wide0.783×

Real models

  • compact-normalized_mlp on digits1.168×
  • compact-normalized_mlp on fashion1.054×
  • compact-normalized_mlp on mnist1.038×
  • uci-har-normalized_mlp on UCI-HAR1.116×
Mixer2 runs

RunsGenerated 27.550× · real models 29.705×

MachineTime
  1. prod-amd64Seed 298017177130.971×Done1 attempt0.121× of the baseline kernels' time
  2. prod-arm64Seed 298017177126.424×Done1 attempt0.116× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal25.667×
  • Batch 1, sparse26.684×
  • Batch 1, wide30.132×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction21.256×
  • compact-mixer on digits13.112×
  • compact-mixer on fashion37.750×
  • compact-mixer on mnist40.722×
  • uci-har-mixer on UCI-HAR53.981×
Recurrent2 runs

RunsGenerated 787.407× · real models 1059.298×

MachineTime
  1. prod-amd64Seed 2980171771967.261×Done1 attempt7.991× of the baseline kernels' time
  2. prod-arm64Seed 2980171771862.330×Done1 attempt8.168× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal785.058×
  • Batch 1, sparse798.878×
  • Batch 1, wide783.660×

Real models

  • appliance-gru on appliance-energy919.173×
  • mitbih-gru on mitbih1098.392×
  • mnist-rows-lstm on mnist-rows1029.840×
  • speech-commands-gru on speech-commands833.454×
  • speech-commands-lstm-gru on speech-commands959.172×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1258.350×
  • tiny-shakespeare-lstm on tiny-shakespeare1203.619×
  • uci-har-gru on uci-har1018.617×
  • uci-har-lstm on uci-har1309.658×