Conjectures.io
Tensor kernels

Entry #70

Completed
Hotkey
5CWyYV…brthL1
Sent
11 Nov 2026, 22:22 UTC
Source
8b559efa97ad

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 runs23.344×Generated 24.245×, real models 22.476×Above the barGain lower bound -2406.0%Guard passed
MLPvs #44Done2 of 2 runs1.036×Generated 0.957×, real models 1.121×Above the barGain lower bound -33.1%Guard passed
ResNetvs #51Done2 of 2 runs106.530×Generated 119.469×, real models 94.992×Above the barGain lower bound -34249.4%Guard passed
MobileNetvs #53Done2 of 2 runs98.341×Generated 116.953×, real models 82.691×Above the barGain lower bound -11559.6%Guard passed
Normalized MLPvs #48Done2 of 2 runs19.585×Generated 21.988×, real models 17.445×Above the barGain lower bound -2417.3%Guard passed
Mixervs #47Done2 of 2 runs0.957×Generated 1.020×, real models 0.898×Not significantGain lower bound -6.4%Guard passed
Recurrentvs #52Done2 of 2 runs786.161×Generated 700.065×, real models 882.845×Above the barGain lower bound -81481.1%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 24.245× · real models 22.476×

MachineTime
  1. prod-amd64Seed 331704625223.318×Done1 attempt0.234× of the baseline kernels' time
  2. prod-arm64Seed 331704625223.369×Done1 attempt0.254× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.123×
  • Batch 1, sparse0.973×
  • Batch 1, wide1.258×

Mixer

  • Batch 1, normal0.968×
  • Batch 1, sparse1.016×
  • Batch 1, wide1.035×

ResNet

  • Batch 1, normal112.721×
  • Batch 1, sparse132.901×
  • Batch 1, wide114.267×

MobileNet

  • Batch 1, normal93.738×
  • Batch 1, sparse95.002×
  • Batch 1, wide93.266×

Recurrent

  • Batch 1, normal713.271×
  • Batch 1, sparse660.989×
  • Batch 1, wide666.716×

Normalized MLP

  • Batch 1, normal24.103×
  • Batch 1, sparse24.496×
  • Batch 1, wide23.046×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.964×
  • compact-mixer on digits1.054×
  • compact-mixer on fashion1.011×
  • compact-mixer on mnist0.990×
  • compact-mlp on digits1.512×
  • compact-mlp on fashion1.952×
  • compact-mlp on mnist0.730×
  • compact-mobilenet on digits62.092×
  • compact-mobilenet on fashion80.530×
  • compact-mobilenet on mnist79.260×
  • compact-normalized_mlp on digits40.852×
  • compact-normalized_mlp on fashion11.841×
  • compact-normalized_mlp on mnist17.971×
  • compact-resnet on digits74.154×
  • compact-resnet on fashion119.168×
  • compact-resnet on mnist113.154×
  • appliance-gru on appliance-energy745.273×
  • mitbih-gru on mitbih905.186×
  • mnist-rows-lstm on mnist-rows724.205×
  • speech-commands-gru on speech-commands702.284×
  • speech-commands-lstm-gru on speech-commands858.604×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1129.534×
  • tiny-shakespeare-lstm on tiny-shakespeare808.490×
  • uci-har-gru on uci-har825.337×
  • uci-har-lstm on uci-har1142.852×
  • uci-har-mixer on UCI-HAR0.970×
  • uci-har-mlp on UCI-HAR0.987×
  • uci-har-mobilenet on UCI-HAR78.689×
  • uci-har-normalized_mlp on UCI-HAR11.438×
  • uci-har-resnet on UCI-HAR77.888×
MLP2 runs

RunsGenerated 0.957× · real models 1.121×

MachineTime
  1. prod-amd64Seed 33170462520.931×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 33170462521.152×Done1 attempt0.025× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.023×
  • Batch 1, sparse0.979×
  • Batch 1, wide0.951×

Real models

  • compact-mlp on digits0.509×
  • compact-mlp on fashion1.921×
  • compact-mlp on mnist1.608×
  • uci-har-mlp on UCI-HAR1.006×
ResNet2 runs

RunsGenerated 119.469× · real models 94.992×

MachineTime
  1. prod-amd64Seed 3317046252111.050×Done1 attempt0.954× of the baseline kernels' time
  2. prod-arm64Seed 3317046252102.193×Done1 attempt0.978× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal114.168×
  • Batch 1, sparse129.345×
  • Batch 1, wide114.166×

Real models

  • compact-resnet on digits76.821×
  • compact-resnet on fashion118.541×
  • compact-resnet on mnist114.822×
  • uci-har-resnet on UCI-HAR77.870×
MobileNet2 runs

RunsGenerated 116.953× · real models 82.691×

MachineTime
  1. prod-amd64Seed 331704625293.611×Done1 attempt1.175× of the baseline kernels' time
  2. prod-arm64Seed 3317046252103.311×Done1 attempt1.152× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal140.497×
  • Batch 1, sparse98.609×
  • Batch 1, wide95.696×

Real models

  • compact-mobilenet on digits64.208×
  • compact-mobilenet on fashion95.371×
  • compact-mobilenet on mnist93.789×
  • uci-har-mobilenet on UCI-HAR81.410×
Normalized MLP2 runs

RunsGenerated 21.988× · real models 17.445×

MachineTime
  1. prod-amd64Seed 331704625220.038×Done1 attempt0.219× of the baseline kernels' time
  2. prod-arm64Seed 331704625219.143×Done1 attempt0.228× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal19.540×
  • Batch 1, sparse24.147×
  • Batch 1, wide23.270×

Real models

  • compact-normalized_mlp on digits43.451×
  • compact-normalized_mlp on fashion10.508×
  • compact-normalized_mlp on mnist16.662×
  • uci-har-normalized_mlp on UCI-HAR12.176×
Mixer2 runs

RunsGenerated 1.020× · real models 0.898×

MachineTime
  1. prod-amd64Seed 33170462521.040×Done1 attempt0.004× of the baseline kernels' time
  2. prod-arm64Seed 33170462520.880×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal1.038×
  • Batch 1, sparse1.017×
  • Batch 1, wide0.994×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.101×
  • compact-mixer on digits0.697×
  • compact-mixer on fashion0.956×
  • compact-mixer on mnist1.007×
  • uci-har-mixer on UCI-HAR0.788×
Recurrent2 runs

RunsGenerated 700.065× · real models 882.845×

MachineTime
  1. prod-amd64Seed 3317046252823.262×Done1 attempt7.947× of the baseline kernels' time
  2. prod-arm64Seed 3317046252750.732×Done1 attempt8.123× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal719.760×
  • Batch 1, sparse687.799×
  • Batch 1, wide686.689×

Real models

  • appliance-gru on appliance-energy801.821×
  • mitbih-gru on mitbih909.812×
  • mnist-rows-lstm on mnist-rows748.345×
  • speech-commands-gru on speech-commands726.782×
  • speech-commands-lstm-gru on speech-commands876.773×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1159.109×
  • tiny-shakespeare-lstm on tiny-shakespeare816.869×
  • uci-har-gru on uci-har833.568×
  • uci-har-lstm on uci-har1186.640×