Conjectures.io
Tensor kernels

Entry #105

Completed
Hotkey
5CJ4KK…gW236t
Sent
12 Nov 2026, 11:57 UTC
Source
ac25d07e4e95

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 runs21.513×Generated 22.362×, real models 20.695×Above the barGain lower bound -2149.4%Guard passed
MLPvs #44Done2 of 2 runs0.985×Generated 0.948×, real models 1.023×Above the barGain lower bound -25.3%Guard passed
ResNetvs #51Done2 of 2 runs1.000×Generated 1.076×, real models 0.929×Above the barGain lower bound -7.3%Guard passed
MobileNetvs #53Done2 of 2 runs20.355×Generated 20.882×, real models 19.841×Above the barGain lower bound -2236.6%Guard passed
Normalized MLPvs #48Done2 of 2 runs41.311×Generated 54.067×, real models 31.565×Above the barGain lower bound -4919.3%Guard passed
Mixervs #58Done2 of 2 runs177.232×Generated 187.485×, real models 167.540×Above the barGain lower bound -22275.0%Guard passed
Recurrentvs #57Done2 of 2 runs928.885×Generated 817.944×, real models 1054.873×Above the barGain lower bound -101318.9%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 22.362× · real models 20.695×

MachineTime
  1. prod-amd64Seed 211416129821.389×Done1 attempt0.218× of the baseline kernels' time
  2. prod-arm64Seed 211416129821.637×Done1 attempt0.223× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.037×
  • Batch 1, sparse1.034×
  • Batch 1, wide0.993×

Mixer

  • Batch 1, normal155.762×
  • Batch 1, sparse167.127×
  • Batch 1, wide184.097×

ResNet

  • Batch 1, normal1.066×
  • Batch 1, sparse1.045×
  • Batch 1, wide1.063×

MobileNet

  • Batch 1, normal20.118×
  • Batch 1, sparse20.352×
  • Batch 1, wide19.992×

Recurrent

  • Batch 1, normal717.755×
  • Batch 1, sparse658.791×
  • Batch 1, wide669.370×

Normalized MLP

  • Batch 1, normal51.517×
  • Batch 1, sparse55.748×
  • Batch 1, wide53.295×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction204.738×
  • compact-mixer on digits184.618×
  • compact-mixer on fashion164.262×
  • compact-mixer on mnist151.569×
  • compact-mlp on digits1.037×
  • compact-mlp on fashion0.819×
  • compact-mlp on mnist0.792×
  • compact-mobilenet on digits18.285×
  • compact-mobilenet on fashion21.145×
  • compact-mobilenet on mnist19.911×
  • compact-normalized_mlp on digits90.532×
  • compact-normalized_mlp on fashion24.703×
  • compact-normalized_mlp on mnist25.748×
  • compact-resnet on digits0.937×
  • compact-resnet on fashion0.990×
  • compact-resnet on mnist0.971×
  • appliance-gru on appliance-energy696.722×
  • mitbih-gru on mitbih901.510×
  • mnist-rows-lstm on mnist-rows717.045×
  • speech-commands-gru on speech-commands696.568×
  • speech-commands-lstm-gru on speech-commands864.077×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1118.062×
  • tiny-shakespeare-lstm on tiny-shakespeare834.209×
  • uci-har-gru on uci-har840.191×
  • uci-har-lstm on uci-har1158.265×
  • uci-har-mixer on UCI-HAR146.283×
  • uci-har-mlp on UCI-HAR0.994×
  • uci-har-mobilenet on UCI-HAR15.138×
  • uci-har-normalized_mlp on UCI-HAR22.723×
  • uci-har-resnet on UCI-HAR0.949×
MLP2 runs

RunsGenerated 0.948× · real models 1.023×

MachineTime
  1. prod-amd64Seed 21141612981.070×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 21141612980.907×Done1 attempt0.015× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.930×
  • Batch 1, sparse0.906×
  • Batch 1, wide0.941×

Real models

  • compact-mlp on digits0.577×
  • compact-mlp on fashion1.883×
  • compact-mlp on mnist0.976×
  • uci-har-mlp on UCI-HAR1.033×
ResNet2 runs

RunsGenerated 1.076× · real models 0.929×

MachineTime
  1. prod-amd64Seed 21141612980.985×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 21141612981.015×Done1 attempt0.009× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal1.106×
  • Batch 1, sparse1.039×
  • Batch 1, wide1.085×

Real models

  • compact-resnet on digits0.968×
  • compact-resnet on fashion0.985×
  • compact-resnet on mnist0.965×
  • uci-har-resnet on UCI-HAR0.809×
MobileNet2 runs

RunsGenerated 20.882× · real models 19.841×

MachineTime
  1. prod-amd64Seed 211416129820.259×Done1 attempt0.253× of the baseline kernels' time
  2. prod-arm64Seed 211416129820.451×Done1 attempt0.248× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal22.397×
  • Batch 1, sparse20.744×
  • Batch 1, wide19.754×

Real models

  • compact-mobilenet on digits17.926×
  • compact-mobilenet on fashion24.311×
  • compact-mobilenet on mnist24.139×
  • uci-har-mobilenet on UCI-HAR14.733×
Normalized MLP2 runs

RunsGenerated 54.067× · real models 31.565×

MachineTime
  1. prod-amd64Seed 211416129840.176×Done1 attempt0.511× of the baseline kernels' time
  2. prod-arm64Seed 211416129842.479×Done1 attempt0.512× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal52.699×
  • Batch 1, sparse56.715×
  • Batch 1, wide53.388×

Real models

  • compact-normalized_mlp on digits87.788×
  • compact-normalized_mlp on fashion21.483×
  • compact-normalized_mlp on mnist25.293×
  • uci-har-normalized_mlp on UCI-HAR20.811×
Mixer2 runs

RunsGenerated 187.485× · real models 167.540×

MachineTime
  1. prod-amd64Seed 2114161298175.947×Done1 attempt0.853× of the baseline kernels' time
  2. prod-arm64Seed 2114161298178.527×Done1 attempt0.847× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal181.454×
  • Batch 1, sparse185.256×
  • Batch 1, wide218.862×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction254.061×
  • compact-mixer on digits152.222×
  • compact-mixer on fashion162.093×
  • compact-mixer on mnist140.835×
  • uci-har-mixer on UCI-HAR149.523×
Recurrent2 runs

RunsGenerated 817.944× · real models 1054.873×

MachineTime
  1. prod-amd64Seed 21141612981004.763×Done1 attempt7.792× of the baseline kernels' time
  2. prod-arm64Seed 2114161298858.737×Done1 attempt7.829× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal845.298×
  • Batch 1, sparse828.899×
  • Batch 1, wide814.819×

Real models

  • appliance-gru on appliance-energy866.598×
  • mitbih-gru on mitbih1119.331×
  • mnist-rows-lstm on mnist-rows1031.797×
  • speech-commands-gru on speech-commands843.748×
  • speech-commands-lstm-gru on speech-commands979.453×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1236.331×
  • tiny-shakespeare-lstm on tiny-shakespeare1210.678×
  • uci-har-gru on uci-har1012.726×
  • uci-har-lstm on uci-har1289.965×