Conjectures.io
Tensor kernels

Entry #31

Completed
Hotkey
5FRrox…Vy2tn3
Sent
11 Nov 2026, 04:01 UTC
Source
ba10183f32bc

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 #36Done2 of 2 runs0.931×Generated 0.928×, real models 0.934×QualifiesGain lower bound 3.6%Guard passed
MLPvs #1Done2 of 2 runs1.011×Generated 0.981×, real models 1.042×Above the barGain lower bound -19.9%Guard passed
ResNetvs #32Done2 of 2 runs0.987×Generated 0.975×, real models 1.000×Above the barGain lower bound -7.8%Guard passed
MobileNetvs #33Done2 of 2 runs0.954×Generated 0.975×, real models 0.934×QualifiesGain lower bound 1.7%Guard passed
Normalized MLPvs #6Done2 of 2 runs0.975×Generated 0.979×, real models 0.972×Above the barGain lower bound -4.7%Guard passed
Mixervs #24Done2 of 2 runs0.980×Generated 0.954×, real models 1.006×Above the barGain lower bound -9.8%Guard passed
Recurrentvs #35Done2 of 2 runs0.711×Generated 0.691×, real models 0.731×QualifiesGain lower bound 26.8%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 0.928× · real models 0.934×

MachineTime
  1. prod-amd64Seed 6338355980.935×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 6338355980.928×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.000×
  • Batch 1, sparse1.108×
  • Batch 1, wide1.033×

Mixer

  • Batch 1, normal0.938×
  • Batch 1, sparse0.982×
  • Batch 1, wide0.972×

ResNet

  • Batch 1, normal1.014×
  • Batch 1, sparse0.999×
  • Batch 1, wide0.963×

MobileNet

  • Batch 1, normal0.987×
  • Batch 1, sparse0.998×
  • Batch 1, wide0.986×

Recurrent

  • Batch 1, normal0.699×
  • Batch 1, sparse0.687×
  • Batch 1, wide0.704×

Normalized MLP

  • Batch 1, normal0.989×
  • Batch 1, sparse1.001×
  • Batch 1, wide0.984×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.881×
  • compact-mixer on digits0.575×
  • compact-mixer on fashion0.981×
  • compact-mixer on mnist0.970×
  • compact-mlp on digits1.056×
  • compact-mlp on fashion0.986×
  • compact-mlp on mnist1.629×
  • compact-mobilenet on digits0.959×
  • compact-mobilenet on fashion1.000×
  • compact-mobilenet on mnist0.998×
  • compact-normalized_mlp on digits0.999×
  • compact-normalized_mlp on fashion1.069×
  • compact-normalized_mlp on mnist0.824×
  • compact-resnet on digits1.014×
  • compact-resnet on fashion1.000×
  • compact-resnet on mnist0.994×
  • appliance-gru on appliance-energy0.661×
  • mitbih-gru on mitbih0.679×
  • mnist-rows-lstm on mnist-rows0.867×
  • speech-commands-gru on speech-commands0.573×
  • speech-commands-lstm-gru on speech-commands0.676×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.814×
  • tiny-shakespeare-lstm on tiny-shakespeare0.859×
  • uci-har-gru on uci-har0.643×
  • uci-har-lstm on uci-har0.820×
  • uci-har-mixer on UCI-HAR1.019×
  • uci-har-mlp on UCI-HAR0.879×
  • uci-har-mobilenet on UCI-HAR1.013×
  • uci-har-normalized_mlp on UCI-HAR0.977×
  • uci-har-resnet on UCI-HAR1.000×
MLP2 runs

RunsGenerated 0.981× · real models 1.042×

MachineTime
  1. prod-amd64Seed 6338355980.968×Done1 attempt0.016× of the baseline kernels' time
  2. prod-arm64Seed 6338355981.057×Done1 attempt0.016× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.965×
  • Batch 1, sparse0.984×
  • Batch 1, wide1.020×

Real models

  • compact-mlp on digits1.496×
  • compact-mlp on fashion0.969×
  • compact-mlp on mnist0.895×
  • uci-har-mlp on UCI-HAR0.910×
ResNet2 runs

RunsGenerated 0.975× · real models 1.000×

MachineTime
  1. prod-amd64Seed 6338355981.032×Done1 attempt0.010× of the baseline kernels' time
  2. prod-arm64Seed 6338355980.944×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal0.964×
  • Batch 1, sparse1.014×
  • Batch 1, wide1.035×

Real models

  • compact-resnet on digits1.081×
  • compact-resnet on fashion0.977×
  • compact-resnet on mnist0.946×
  • uci-har-resnet on UCI-HAR0.999×
MobileNet2 runs

RunsGenerated 0.975× · real models 0.934×

MachineTime
  1. prod-amd64Seed 6338355980.962×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 6338355980.947×Done1 attempt0.015× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal0.972×
  • Batch 1, sparse1.018×
  • Batch 1, wide0.991×

Real models

  • compact-mobilenet on digits0.966×
  • compact-mobilenet on fashion0.902×
  • compact-mobilenet on mnist0.937×
  • uci-har-mobilenet on UCI-HAR0.931×
Normalized MLP2 runs

RunsGenerated 0.979× · real models 0.972×

MachineTime
  1. prod-amd64Seed 6338355980.964×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 6338355980.987×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.006×
  • Batch 1, sparse0.960×
  • Batch 1, wide1.018×

Real models

  • compact-normalized_mlp on digits1.018×
  • compact-normalized_mlp on fashion0.902×
  • compact-normalized_mlp on mnist1.011×
  • uci-har-normalized_mlp on UCI-HAR0.960×
Mixer2 runs

RunsGenerated 0.954× · real models 1.006×

MachineTime
  1. prod-amd64Seed 6338355981.081×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 6338355980.888×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.999×
  • Batch 1, sparse0.989×
  • Batch 1, wide0.926×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.301×
  • compact-mixer on digits0.894×
  • compact-mixer on fashion0.950×
  • compact-mixer on mnist0.936×
  • uci-har-mixer on UCI-HAR0.996×
Recurrent2 runs

RunsGenerated 0.691× · real models 0.731×

MachineTime
  1. prod-amd64Seed 6338355980.703×Done1 attempt0.016× of the baseline kernels' time
  2. prod-arm64Seed 6338355980.719×Done1 attempt0.017× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.695×
  • Batch 1, sparse0.683×
  • Batch 1, wide0.699×

Real models

  • appliance-gru on appliance-energy0.686×
  • mitbih-gru on mitbih0.678×
  • mnist-rows-lstm on mnist-rows0.867×
  • speech-commands-gru on speech-commands0.573×
  • speech-commands-lstm-gru on speech-commands0.681×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.830×
  • tiny-shakespeare-lstm on tiny-shakespeare0.863×
  • uci-har-gru on uci-har0.645×
  • uci-har-lstm on uci-har0.822×