Conjectures.io
Tensor kernels

Entry #80

Completed
Hotkey
5CAVQq…5eCqYj
Sent
12 Nov 2026, 02:33 UTC
Source
42602a34fa9b

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 runs17.054×Generated 17.402×, real models 16.714×Above the barGain lower bound -1830.6%Guard passed
MLPvs #44Done2 of 2 runs0.908×Generated 0.982×, real models 0.839×Not significantGain lower bound -24.1%Guard passed
ResNetvs #51Done2 of 2 runs3.709×Generated 3.918×, real models 3.512×Above the barGain lower bound -338.8%Guard passed
MobileNetvs #53Done2 of 2 runs0.991×Generated 1.028×, real models 0.955×Above the barGain lower bound -10.3%Guard passed
Normalized MLPvs #48Done2 of 2 runs43.790×Generated 51.594×, real models 37.167×Above the barGain lower bound -5242.5%Guard passed
Mixervs #47Done2 of 2 runs178.109×Generated 181.960×, real models 174.340×Above the barGain lower bound -19292.0%Guard passed
Recurrentvs #54Done2 of 2 runs827.099×Generated 708.272×, real models 965.862×Above the barGain lower bound -87010.8%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 17.402× · real models 16.714×

MachineTime
  1. prod-amd64Seed 328228591917.812×Done1 attempt0.171× of the baseline kernels' time
  2. prod-arm64Seed 328228591916.329×Done1 attempt0.201× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.203×
  • Batch 1, sparse0.887×
  • Batch 1, wide1.159×

Mixer

  • Batch 1, normal166.324×
  • Batch 1, sparse167.297×
  • Batch 1, wide196.248×

ResNet

  • Batch 1, normal3.753×
  • Batch 1, sparse4.250×
  • Batch 1, wide3.765×

MobileNet

  • Batch 1, normal1.032×
  • Batch 1, sparse1.006×
  • Batch 1, wide1.015×

Recurrent

  • Batch 1, normal703.718×
  • Batch 1, sparse646.923×
  • Batch 1, wide677.862×

Normalized MLP

  • Batch 1, normal53.455×
  • Batch 1, sparse59.003×
  • Batch 1, wide62.556×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction233.399×
  • compact-mixer on digits149.177×
  • compact-mixer on fashion158.741×
  • compact-mixer on mnist152.529×
  • compact-mlp on digits1.042×
  • compact-mlp on fashion1.170×
  • compact-mlp on mnist1.048×
  • compact-mobilenet on digits0.912×
  • compact-mobilenet on fashion0.840×
  • compact-mobilenet on mnist0.818×
  • compact-normalized_mlp on digits132.987×
  • compact-normalized_mlp on fashion34.704×
  • compact-normalized_mlp on mnist44.383×
  • compact-resnet on digits2.704×
  • compact-resnet on fashion3.612×
  • compact-resnet on mnist3.522×
  • appliance-gru on appliance-energy766.052×
  • mitbih-gru on mitbih898.538×
  • mnist-rows-lstm on mnist-rows724.547×
  • speech-commands-gru on speech-commands674.531×
  • speech-commands-lstm-gru on speech-commands870.338×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1137.537×
  • tiny-shakespeare-lstm on tiny-shakespeare824.526×
  • uci-har-gru on uci-har864.032×
  • uci-har-lstm on uci-har1076.567×
  • uci-har-mixer on UCI-HAR161.098×
  • uci-har-mlp on UCI-HAR1.004×
  • uci-har-mobilenet on UCI-HAR0.964×
  • uci-har-normalized_mlp on UCI-HAR18.296×
  • uci-har-resnet on UCI-HAR5.143×
MLP2 runs

RunsGenerated 0.982× · real models 0.839×

MachineTime
  1. prod-amd64Seed 32822859190.874×Done1 attempt0.015× of the baseline kernels' time
  2. prod-arm64Seed 32822859190.943×Done1 attempt0.032× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.016×
  • Batch 1, sparse1.025×
  • Batch 1, wide0.935×

Real models

  • compact-mlp on digits0.574×
  • compact-mlp on fashion0.760×
  • compact-mlp on mnist1.165×
  • uci-har-mlp on UCI-HAR0.977×
ResNet2 runs

RunsGenerated 3.918× · real models 3.512×

MachineTime
  1. prod-amd64Seed 32822859193.743×Done1 attempt0.030× of the baseline kernels' time
  2. prod-arm64Seed 32822859193.675×Done1 attempt0.034× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal3.699×
  • Batch 1, sparse4.259×
  • Batch 1, wide3.685×

Real models

  • compact-resnet on digits2.397×
  • compact-resnet on fashion3.620×
  • compact-resnet on mnist3.527×
  • uci-har-resnet on UCI-HAR4.969×
MobileNet2 runs

RunsGenerated 1.028× · real models 0.955×

MachineTime
  1. prod-amd64Seed 32822859190.990×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 32822859190.992×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.047×
  • Batch 1, sparse0.992×
  • Batch 1, wide1.037×

Real models

  • compact-mobilenet on digits0.931×
  • compact-mobilenet on fashion0.967×
  • compact-mobilenet on mnist0.967×
  • uci-har-mobilenet on UCI-HAR0.954×
Normalized MLP2 runs

RunsGenerated 51.594× · real models 37.167×

MachineTime
  1. prod-amd64Seed 328228591955.793×Done1 attempt0.596× of the baseline kernels' time
  2. prod-arm64Seed 328228591934.369×Done1 attempt0.599× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal59.454×
  • Batch 1, sparse61.423×
  • Batch 1, wide60.442×

Real models

  • compact-normalized_mlp on digits69.373×
  • compact-normalized_mlp on fashion38.999×
  • compact-normalized_mlp on mnist40.956×
  • uci-har-normalized_mlp on UCI-HAR17.221×
Mixer2 runs

RunsGenerated 181.960× · real models 174.340×

MachineTime
  1. prod-amd64Seed 3282285919184.893×Done1 attempt0.861× of the baseline kernels' time
  2. prod-arm64Seed 3282285919171.574×Done1 attempt0.859× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal181.739×
  • Batch 1, sparse176.940×
  • Batch 1, wide200.434×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction255.824×
  • compact-mixer on digits182.907×
  • compact-mixer on fashion156.742×
  • compact-mixer on mnist151.993×
  • uci-har-mixer on UCI-HAR144.477×
Recurrent2 runs

RunsGenerated 708.272× · real models 965.862×

MachineTime
  1. prod-amd64Seed 3282285919880.043×Done1 attempt8.529× of the baseline kernels' time
  2. prod-arm64Seed 3282285919777.340×Done1 attempt8.761× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal732.415×
  • Batch 1, sparse701.799×
  • Batch 1, wide691.372×

Real models

  • appliance-gru on appliance-energy780.146×
  • mitbih-gru on mitbih900.134×
  • mnist-rows-lstm on mnist-rows1063.662×
  • speech-commands-gru on speech-commands679.382×
  • speech-commands-lstm-gru on speech-commands903.827×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1264.753×
  • tiny-shakespeare-lstm on tiny-shakespeare1219.346×
  • uci-har-gru on uci-har816.589×
  • uci-har-lstm on uci-har1266.521×