Conjectures.io
Tensor kernels

Entry #97

Completed
Hotkey
5C7inh…EvgpnY
Sent
12 Nov 2026, 09:26 UTC
Source
32b2190c30c8

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 runs16.490×Generated 17.255×, real models 15.759×Above the barGain lower bound -1644.1%Guard passed
MLPvs #44Done2 of 2 runs0.923×Generated 1.006×, real models 0.848×Not significantGain lower bound -10.5%Guard passed
ResNetvs #51Done2 of 2 runs3.695×Generated 3.708×, real models 3.683×Above the barGain lower bound -300.3%Guard passed
MobileNetvs #53Done2 of 2 runs0.962×Generated 0.958×, real models 0.965×Not significantGain lower bound -8.5%Guard passed
Normalized MLPvs #48Done2 of 2 runs48.206×Generated 61.157×, real models 37.998×Above the barGain lower bound -5705.5%Guard passed
Mixervs #47Done2 of 2 runs183.382×Generated 207.224×, real models 162.283×Above the barGain lower bound -20031.5%Guard passed
Recurrentvs #55Done2 of 2 runs841.795×Generated 737.267×, real models 961.142×Above the barGain lower bound -88253.8%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 17.255× · real models 15.759×

MachineTime
  1. prod-amd64Seed 223948066316.595×Done1 attempt0.168× of the baseline kernels' time
  2. prod-arm64Seed 223948066316.385×Done1 attempt0.185× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.095×
  • Batch 1, sparse1.075×
  • Batch 1, wide0.901×

Mixer

  • Batch 1, normal158.924×
  • Batch 1, sparse159.372×
  • Batch 1, wide190.134×

ResNet

  • Batch 1, normal3.500×
  • Batch 1, sparse3.973×
  • Batch 1, wide3.652×

MobileNet

  • Batch 1, normal0.982×
  • Batch 1, sparse0.967×
  • Batch 1, wide0.989×

Recurrent

  • Batch 1, normal684.162×
  • Batch 1, sparse625.922×
  • Batch 1, wide649.343×

Normalized MLP

  • Batch 1, normal55.247×
  • Batch 1, sparse56.691×
  • Batch 1, wide56.006×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction237.875×
  • compact-mixer on digits116.916×
  • compact-mixer on fashion159.879×
  • compact-mixer on mnist156.625×
  • compact-mlp on digits1.093×
  • compact-mlp on fashion1.197×
  • compact-mlp on mnist1.510×
  • compact-mobilenet on digits0.985×
  • compact-mobilenet on fashion0.813×
  • compact-mobilenet on mnist0.818×
  • compact-normalized_mlp on digits86.319×
  • compact-normalized_mlp on fashion20.467×
  • compact-normalized_mlp on mnist19.000×
  • compact-resnet on digits2.579×
  • compact-resnet on fashion3.667×
  • compact-resnet on mnist3.538×
  • appliance-gru on appliance-energy735.387×
  • mitbih-gru on mitbih895.221×
  • mnist-rows-lstm on mnist-rows719.774×
  • speech-commands-gru on speech-commands705.730×
  • speech-commands-lstm-gru on speech-commands864.157×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1101.096×
  • tiny-shakespeare-lstm on tiny-shakespeare828.230×
  • uci-har-gru on uci-har823.919×
  • uci-har-lstm on uci-har1165.770×
  • uci-har-mixer on UCI-HAR167.708×
  • uci-har-mlp on UCI-HAR1.053×
  • uci-har-mobilenet on UCI-HAR0.935×
  • uci-har-normalized_mlp on UCI-HAR20.485×
  • uci-har-resnet on UCI-HAR4.837×
MLP2 runs

RunsGenerated 1.006× · real models 0.848×

MachineTime
  1. prod-amd64Seed 22394806630.966×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 22394806630.882×Done1 attempt0.023× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.921×
  • Batch 1, sparse0.962×
  • Batch 1, wide0.945×

Real models

  • compact-mlp on digits0.681×
  • compact-mlp on fashion0.701×
  • compact-mlp on mnist1.153×
  • uci-har-mlp on UCI-HAR0.938×
ResNet2 runs

RunsGenerated 3.708× · real models 3.683×

MachineTime
  1. prod-amd64Seed 22394806633.858×Done1 attempt0.031× of the baseline kernels' time
  2. prod-arm64Seed 22394806633.539×Done1 attempt0.034× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal3.697×
  • Batch 1, sparse4.119×
  • Batch 1, wide3.627×

Real models

  • compact-resnet on digits2.679×
  • compact-resnet on fashion3.789×
  • compact-resnet on mnist3.737×
  • uci-har-resnet on UCI-HAR4.850×
MobileNet2 runs

RunsGenerated 0.958× · real models 0.965×

MachineTime
  1. prod-amd64Seed 22394806630.942×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 22394806630.981×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal0.953×
  • Batch 1, sparse0.996×
  • Batch 1, wide0.984×

Real models

  • compact-mobilenet on digits0.942×
  • compact-mobilenet on fashion0.965×
  • compact-mobilenet on mnist0.964×
  • uci-har-mobilenet on UCI-HAR0.990×
Normalized MLP2 runs

RunsGenerated 61.157× · real models 37.998×

MachineTime
  1. prod-amd64Seed 223948066352.470×Done1 attempt0.508× of the baseline kernels' time
  2. prod-arm64Seed 223948066344.289×Done1 attempt0.512× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal54.727×
  • Batch 1, sparse63.465×
  • Batch 1, wide66.658×

Real models

  • compact-normalized_mlp on digits75.560×
  • compact-normalized_mlp on fashion34.992×
  • compact-normalized_mlp on mnist37.921×
  • uci-har-normalized_mlp on UCI-HAR20.791×
Mixer2 runs

RunsGenerated 207.224× · real models 162.283×

MachineTime
  1. prod-amd64Seed 2239480663197.480×Done1 attempt0.868× of the baseline kernels' time
  2. prod-arm64Seed 2239480663170.291×Done1 attempt0.870× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal164.961×
  • Batch 1, sparse217.990×
  • Batch 1, wide253.645×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction268.395×
  • compact-mixer on digits126.003×
  • compact-mixer on fashion157.809×
  • compact-mixer on mnist146.121×
  • uci-har-mixer on UCI-HAR144.334×
Recurrent2 runs

RunsGenerated 737.267× · real models 961.142×

MachineTime
  1. prod-amd64Seed 2239480663885.771×Done1 attempt8.510× of the baseline kernels' time
  2. prod-arm64Seed 2239480663800.002×Done1 attempt8.534× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal767.483×
  • Batch 1, sparse751.978×
  • Batch 1, wide725.601×

Real models

  • appliance-gru on appliance-energy794.378×
  • mitbih-gru on mitbih1004.258×
  • mnist-rows-lstm on mnist-rows754.689×
  • speech-commands-gru on speech-commands780.307×
  • speech-commands-lstm-gru on speech-commands982.642×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1321.837×
  • tiny-shakespeare-lstm on tiny-shakespeare921.909×
  • uci-har-gru on uci-har926.915×
  • uci-har-lstm on uci-har1342.395×