Conjectures.io
Tensor kernels

Entry #100

Completed
Hotkey
5Gwzd6…eVnAzj
Sent
12 Nov 2026, 10:16 UTC
Source
c319ab518ffe

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 runs17.092×Generated 17.903×, real models 16.318×Above the barGain lower bound -1874.2%Guard passed
MLPvs #44Done2 of 2 runs1.072×Generated 1.033×, real models 1.112×Above the barGain lower bound -31.1%Guard passed
ResNetvs #51Done2 of 2 runs4.234×Generated 4.642×, real models 3.863×Above the barGain lower bound -533.6%Guard passed
MobileNetvs #53Done2 of 2 runs0.999×Generated 1.054×, real models 0.947×Above the barGain lower bound -12.3%Guard passed
Normalized MLPvs #48Done2 of 2 runs48.636×Generated 69.610×, real models 33.982×Above the barGain lower bound -6411.2%Guard passed
Mixervs #47Done2 of 2 runs167.131×Generated 178.339×, real models 156.628×Above the barGain lower bound -18713.1%Guard passed
Recurrentvs #55Done2 of 2 runs842.359×Generated 753.480×, real models 941.723×Above the barGain lower bound -87490.2%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 17.903× · real models 16.318×

MachineTime
  1. prod-amd64Seed 239414616316.534×Done1 attempt0.173× of the baseline kernels' time
  2. prod-arm64Seed 239414616317.669×Done1 attempt0.181× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.035×
  • Batch 1, sparse1.042×
  • Batch 1, wide1.193×

Mixer

  • Batch 1, normal153.761×
  • Batch 1, sparse159.780×
  • Batch 1, wide191.370×

ResNet

  • Batch 1, normal3.952×
  • Batch 1, sparse4.383×
  • Batch 1, wide3.946×

MobileNet

  • Batch 1, normal0.975×
  • Batch 1, sparse0.960×
  • Batch 1, wide0.977×

Recurrent

  • Batch 1, normal703.028×
  • Batch 1, sparse670.973×
  • Batch 1, wide663.039×

Normalized MLP

  • Batch 1, normal62.975×
  • Batch 1, sparse61.136×
  • Batch 1, wide62.559×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction204.096×
  • compact-mixer on digits169.497×
  • compact-mixer on fashion161.452×
  • compact-mixer on mnist153.042×
  • compact-mlp on digits3.472×
  • compact-mlp on fashion1.131×
  • compact-mlp on mnist0.662×
  • compact-mobilenet on digits0.849×
  • compact-mobilenet on fashion0.821×
  • compact-mobilenet on mnist0.813×
  • compact-normalized_mlp on digits88.878×
  • compact-normalized_mlp on fashion24.361×
  • compact-normalized_mlp on mnist26.019×
  • compact-resnet on digits2.735×
  • compact-resnet on fashion3.553×
  • compact-resnet on mnist3.567×
  • appliance-gru on appliance-energy714.310×
  • mitbih-gru on mitbih873.387×
  • mnist-rows-lstm on mnist-rows724.395×
  • speech-commands-gru on speech-commands698.420×
  • speech-commands-lstm-gru on speech-commands860.546×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1122.737×
  • tiny-shakespeare-lstm on tiny-shakespeare823.237×
  • uci-har-gru on uci-har832.362×
  • uci-har-lstm on uci-har1136.574×
  • uci-har-mixer on UCI-HAR167.212×
  • uci-har-mlp on UCI-HAR1.039×
  • uci-har-mobilenet on UCI-HAR0.929×
  • uci-har-normalized_mlp on UCI-HAR21.129×
  • uci-har-resnet on UCI-HAR4.889×
MLP2 runs

RunsGenerated 1.033× · real models 1.112×

MachineTime
  1. prod-amd64Seed 23941461630.935×Done1 attempt0.016× of the baseline kernels' time
  2. prod-arm64Seed 23941461631.228×Done1 attempt0.016× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.951×
  • Batch 1, sparse0.991×
  • Batch 1, wide0.981×

Real models

  • compact-mlp on digits1.035×
  • compact-mlp on fashion0.837×
  • compact-mlp on mnist1.784×
  • uci-har-mlp on UCI-HAR0.989×
ResNet2 runs

RunsGenerated 4.642× · real models 3.863×

MachineTime
  1. prod-amd64Seed 23941461634.056×Done1 attempt0.031× of the baseline kernels' time
  2. prod-arm64Seed 23941461634.421×Done1 attempt0.035× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal5.348×
  • Batch 1, sparse4.349×
  • Batch 1, wide3.921×

Real models

  • compact-resnet on digits2.803×
  • compact-resnet on fashion3.905×
  • compact-resnet on mnist3.806×
  • uci-har-resnet on UCI-HAR5.346×
MobileNet2 runs

RunsGenerated 1.054× · real models 0.947×

MachineTime
  1. prod-amd64Seed 23941461630.937×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 23941461631.064×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal0.962×
  • Batch 1, sparse0.992×
  • Batch 1, wide1.003×

Real models

  • compact-mobilenet on digits0.920×
  • compact-mobilenet on fashion0.962×
  • compact-mobilenet on mnist0.953×
  • uci-har-mobilenet on UCI-HAR0.951×
Normalized MLP2 runs

RunsGenerated 69.610× · real models 33.982×

MachineTime
  1. prod-amd64Seed 239414616351.801×Done1 attempt0.587× of the baseline kernels' time
  2. prod-arm64Seed 239414616345.665×Done1 attempt0.594× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal64.464×
  • Batch 1, sparse66.504×
  • Batch 1, wide67.675×

Real models

  • compact-normalized_mlp on digits84.892×
  • compact-normalized_mlp on fashion20.183×
  • compact-normalized_mlp on mnist34.430×
  • uci-har-normalized_mlp on UCI-HAR22.606×
Mixer2 runs

RunsGenerated 178.339× · real models 156.628×

MachineTime
  1. prod-amd64Seed 2394146163167.064×Done1 attempt0.865× of the baseline kernels' time
  2. prod-arm64Seed 2394146163167.199×Done1 attempt0.859× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal168.690×
  • Batch 1, sparse168.167×
  • Batch 1, wide197.014×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction206.425×
  • compact-mixer on digits138.465×
  • compact-mixer on fashion157.647×
  • compact-mixer on mnist150.464×
  • uci-har-mixer on UCI-HAR139.036×
Recurrent2 runs

RunsGenerated 753.480× · real models 941.723×

MachineTime
  1. prod-amd64Seed 2394146163875.820×Done1 attempt8.494× of the baseline kernels' time
  2. prod-arm64Seed 2394146163810.177×Done1 attempt8.795× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal787.107×
  • Batch 1, sparse753.851×
  • Batch 1, wide728.858×

Real models

  • appliance-gru on appliance-energy782.047×
  • mitbih-gru on mitbih963.695×
  • mnist-rows-lstm on mnist-rows766.004×
  • speech-commands-gru on speech-commands751.498×
  • speech-commands-lstm-gru on speech-commands953.404×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1292.047×
  • tiny-shakespeare-lstm on tiny-shakespeare907.602×
  • uci-har-gru on uci-har908.216×
  • uci-har-lstm on uci-har1322.316×