Conjectures.io
Tensor kernels

Entry #27

Completed
Hotkey
5CvjDF…xeUQzi
Sent
11 Nov 2026, 03:07 UTC
Source
8c54deffd88b

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 #29Done2 of 2 runs0.829×Generated 0.911×, real models 0.753×QualifiesGain lower bound 7.1%Guard passed
MLPvs #1Done2 of 2 runs1.008×Generated 0.999×, real models 1.017×Above the barGain lower bound -24.4%Guard passed
ResNetvs #32Done2 of 2 runs0.984×Generated 1.005×, real models 0.964×Above the barGain lower bound -5.3%Guard passed
MobileNetvs #33Done2 of 2 runs0.988×Generated 1.019×, real models 0.959×Above the barGain lower bound -1.3%Guard passed
Normalized MLPvs #6Done2 of 2 runs1.127×Generated 1.037×, real models 1.225×Above the barGain lower bound -46.1%Guard passed
Mixervs #24Done2 of 2 runs1.028×Generated 1.051×, real models 1.004×Above the barGain lower bound -14.8%Guard passed
Recurrentvs #34Done2 of 2 runs0.615×Generated 0.564×, real models 0.672×QualifiesGain lower bound 34.5%Guard passed

Download Kernel.lean

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 0.911× · real models 0.753×

MachineTime
  1. prod-amd64Seed 10405174690.810×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 10405174690.848×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.048×
  • Batch 1, sparse0.967×
  • Batch 1, wide1.052×

Mixer

  • Batch 1, normal1.021×
  • Batch 1, sparse1.052×
  • Batch 1, wide0.960×

ResNet

  • Batch 1, normal0.977×
  • Batch 1, sparse1.029×
  • Batch 1, wide0.969×

MobileNet

  • Batch 1, normal0.977×
  • Batch 1, sparse0.983×
  • Batch 1, wide0.975×

Recurrent

  • Batch 1, normal0.569×
  • Batch 1, sparse0.590×
  • Batch 1, wide0.570×

Normalized MLP

  • Batch 1, normal1.137×
  • Batch 1, sparse1.119×
  • Batch 1, wide1.000×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.855×
  • compact-mixer on digits1.527×
  • compact-mixer on fashion0.960×
  • compact-mixer on mnist0.969×
  • compact-mlp on digits1.001×
  • compact-mlp on fashion1.047×
  • compact-mlp on mnist0.779×
  • compact-mobilenet on digits0.816×
  • compact-mobilenet on fashion0.731×
  • compact-mobilenet on mnist0.676×
  • compact-normalized_mlp on digits1.017×
  • compact-normalized_mlp on fashion0.849×
  • compact-normalized_mlp on mnist0.716×
  • compact-resnet on digits0.926×
  • compact-resnet on fashion0.820×
  • compact-resnet on mnist0.754×
  • appliance-gru on appliance-energy0.638×
  • mitbih-gru on mitbih0.621×
  • mnist-rows-lstm on mnist-rows0.628×
  • speech-commands-gru on speech-commands0.512×
  • speech-commands-lstm-gru on speech-commands0.563×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.473×
  • tiny-shakespeare-lstm on tiny-shakespeare0.603×
  • uci-har-gru on uci-har0.593×
  • uci-har-lstm on uci-har0.156×
  • uci-har-mixer on UCI-HAR1.074×
  • uci-har-mlp on UCI-HAR1.018×
  • uci-har-mobilenet on UCI-HAR0.268×
  • uci-har-normalized_mlp on UCI-HAR0.959×
  • uci-har-resnet on UCI-HAR0.478×
MLP2 runs

RunsGenerated 0.999× · real models 1.017×

MachineTime
  1. prod-amd64Seed 10405174690.899×Done1 attempt0.015× of the baseline kernels' time
  2. prod-arm64Seed 10405174691.129×Done1 attempt0.017× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.078×
  • Batch 1, sparse0.976×
  • Batch 1, wide0.964×

Real models

  • compact-mlp on digits0.954×
  • compact-mlp on fashion1.291×
  • compact-mlp on mnist0.970×
  • uci-har-mlp on UCI-HAR0.894×
ResNet2 runs

RunsGenerated 1.005× · real models 0.964×

MachineTime
  1. prod-amd64Seed 10405174690.974×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 10405174690.995×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal0.999×
  • Batch 1, sparse1.022×
  • Batch 1, wide0.986×

Real models

  • compact-resnet on digits0.969×
  • compact-resnet on fashion0.945×
  • compact-resnet on mnist0.968×
  • uci-har-resnet on UCI-HAR0.975×
MobileNet2 runs

RunsGenerated 1.019× · real models 0.959×

MachineTime
  1. prod-amd64Seed 10405174690.988×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 10405174690.989×Done1 attempt0.014× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.000×
  • Batch 1, sparse1.018×
  • Batch 1, wide1.017×

Real models

  • compact-mobilenet on digits0.974×
  • compact-mobilenet on fashion0.980×
  • compact-mobilenet on mnist0.940×
  • uci-har-mobilenet on UCI-HAR0.941×
Normalized MLP2 runs

RunsGenerated 1.037× · real models 1.225×

MachineTime
  1. prod-amd64Seed 10405174691.032×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 10405174691.231×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.011×
  • Batch 1, sparse1.045×
  • Batch 1, wide1.125×

Real models

  • compact-normalized_mlp on digits1.260×
  • compact-normalized_mlp on fashion1.963×
  • compact-normalized_mlp on mnist1.019×
  • uci-har-normalized_mlp on UCI-HAR0.893×
Mixer2 runs

RunsGenerated 1.051× · real models 1.004×

MachineTime
  1. prod-amd64Seed 10405174691.074×Done1 attempt0.004× of the baseline kernels' time
  2. prod-arm64Seed 10405174690.983×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal1.049×
  • Batch 1, sparse1.043×
  • Batch 1, wide1.037×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.995×
  • compact-mixer on digits1.092×
  • compact-mixer on fashion0.961×
  • compact-mixer on mnist0.951×
  • uci-har-mixer on UCI-HAR1.028×
Recurrent2 runs

RunsGenerated 0.564× · real models 0.672×

MachineTime
  1. prod-amd64Seed 10405174690.625×Done1 attempt0.024× of the baseline kernels' time
  2. prod-arm64Seed 10405174690.606×Done1 attempt0.024× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.555×
  • Batch 1, sparse0.573×
  • Batch 1, wide0.557×

Real models

  • appliance-gru on appliance-energy0.561×
  • mitbih-gru on mitbih0.631×
  • mnist-rows-lstm on mnist-rows0.921×
  • speech-commands-gru on speech-commands0.403×
  • speech-commands-lstm-gru on speech-commands0.545×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.917×
  • tiny-shakespeare-lstm on tiny-shakespeare0.873×
  • uci-har-gru on uci-har0.536×
  • uci-har-lstm on uci-har0.905×