Conjectures.io
Tensor kernels

Entry #12

Completed
Hotkey
5Gs6J9…LVfdCn
Sent
10 Nov 2026, 20:52 UTC
Source
c003ae85e12c

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 #19Done2 of 2 runs0.425×Generated 0.305×, real models 0.593×QualifiesGain lower bound 53.0%Guard passed
MLPvs #1Done2 of 2 runs0.961×Generated 0.956×, real models 0.966×Not significantGain lower bound -7.7%Guard passed
ResNetvs #15Done2 of 2 runs0.139×Generated 0.058×, real models 0.329×QualifiesGain lower bound 79.6%Guard passed
MobileNetvs #16Done2 of 2 runs0.119×Generated 0.053×, real models 0.271×QualifiesGain lower bound 83.5%Guard passed
Normalized MLPvs #6Done2 of 2 runs0.961×Generated 0.899×, real models 1.028×Not significantGain lower bound -2.8%Guard passed
Mixervs #7Done2 of 2 runs0.960×Generated 0.930×, real models 0.990×Not significantGain lower bound -4.6%Guard passed
Recurrentvs #18Done2 of 2 runs0.756×Generated 0.699×, real models 0.818×QualifiesGain lower bound 21.9%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 0.305× · real models 0.593×

MachineTime
  1. prod-amd64Seed 38317971850.417×Done1 attempt0.016× of the baseline kernels' time
  2. prod-arm64Seed 38317971850.433×Done1 attempt0.019× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.984×
  • Batch 1, sparse1.092×
  • Batch 1, wide0.969×

Mixer

  • Batch 1, normal0.991×
  • Batch 1, sparse1.056×
  • Batch 1, wide1.086×

ResNet

  • Batch 1, normal0.046×
  • Batch 1, sparse0.039×
  • Batch 1, wide0.046×

MobileNet

  • Batch 1, normal0.032×
  • Batch 1, sparse0.030×
  • Batch 1, wide0.032×

Recurrent

  • Batch 1, normal0.654×
  • Batch 1, sparse0.640×
  • Batch 1, wide0.647×

Normalized MLP

  • Batch 1, normal0.925×
  • Batch 1, sparse0.946×
  • Batch 1, wide0.935×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.979×
  • compact-mixer on digits1.135×
  • compact-mixer on fashion0.969×
  • compact-mixer on mnist0.987×
  • compact-mlp on digits1.165×
  • compact-mlp on fashion1.234×
  • compact-mlp on mnist1.691×
  • compact-mobilenet on digits0.075×
  • compact-mobilenet on fashion0.103×
  • compact-mobilenet on mnist0.127×
  • compact-normalized_mlp on digits1.295×
  • compact-normalized_mlp on fashion1.207×
  • compact-normalized_mlp on mnist0.970×
  • compact-resnet on digits0.112×
  • compact-resnet on fashion0.098×
  • compact-resnet on mnist0.140×
  • appliance-gru on appliance-energy0.632×
  • mitbih-gru on mitbih0.674×
  • mnist-rows-lstm on mnist-rows0.819×
  • speech-commands-gru on speech-commands0.542×
  • speech-commands-lstm-gru on speech-commands0.686×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.900×
  • tiny-shakespeare-lstm on tiny-shakespeare0.834×
  • uci-har-gru on uci-har0.637×
  • uci-har-lstm on uci-har0.916×
  • uci-har-mixer on UCI-HAR0.977×
  • uci-har-mlp on UCI-HAR0.970×
  • uci-har-mobilenet on UCI-HAR0.988×
  • uci-har-normalized_mlp on UCI-HAR1.028×
  • uci-har-resnet on UCI-HAR2.247×
MLP2 runs

RunsGenerated 0.956× · real models 0.966×

MachineTime
  1. prod-amd64Seed 38317971850.918×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 38317971851.006×Done1 attempt0.024× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.046×
  • Batch 1, sparse1.022×
  • Batch 1, wide0.952×

Real models

  • compact-mlp on digits1.054×
  • compact-mlp on fashion0.940×
  • compact-mlp on mnist0.888×
  • uci-har-mlp on UCI-HAR0.990×
ResNet2 runs

RunsGenerated 0.058× · real models 0.329×

MachineTime
  1. prod-amd64Seed 38317971850.143×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 38317971850.135×Done1 attempt0.009× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal0.061×
  • Batch 1, sparse0.056×
  • Batch 1, wide0.067×

Real models

  • compact-resnet on digits0.172×
  • compact-resnet on fashion0.127×
  • compact-resnet on mnist0.174×
  • uci-har-resnet on UCI-HAR3.089×
MobileNet2 runs

RunsGenerated 0.053× · real models 0.271×

MachineTime
  1. prod-amd64Seed 38317971850.126×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 38317971850.114×Done1 attempt0.013× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal0.053×
  • Batch 1, sparse0.055×
  • Batch 1, wide0.057×

Real models

  • compact-mobilenet on digits0.130×
  • compact-mobilenet on fashion0.149×
  • compact-mobilenet on mnist0.174×
  • uci-har-mobilenet on UCI-HAR1.595×
Normalized MLP2 runs

RunsGenerated 0.899× · real models 1.028×

MachineTime
  1. prod-amd64Seed 38317971850.993×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 38317971850.931×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal0.931×
  • Batch 1, sparse0.917×
  • Batch 1, wide0.918×

Real models

  • compact-normalized_mlp on digits1.119×
  • compact-normalized_mlp on fashion1.004×
  • compact-normalized_mlp on mnist0.998×
  • uci-har-normalized_mlp on UCI-HAR0.997×
Mixer2 runs

RunsGenerated 0.930× · real models 0.990×

MachineTime
  1. prod-amd64Seed 38317971850.985×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 38317971850.935×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.969×
  • Batch 1, sparse1.087×
  • Batch 1, wide0.862×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.032×
  • compact-mixer on digits0.945×
  • compact-mixer on fashion0.989×
  • compact-mixer on mnist0.979×
  • uci-har-mixer on UCI-HAR1.008×
Recurrent2 runs

RunsGenerated 0.699× · real models 0.818×

MachineTime
  1. prod-amd64Seed 38317971850.756×Done1 attempt0.221× of the baseline kernels' time
  2. prod-arm64Seed 38317971850.756×Done1 attempt0.217× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.703×
  • Batch 1, sparse0.687×
  • Batch 1, wide0.695×

Real models

  • appliance-gru on appliance-energy0.789×
  • mitbih-gru on mitbih0.927×
  • mnist-rows-lstm on mnist-rows0.858×
  • speech-commands-gru on speech-commands0.561×
  • speech-commands-lstm-gru on speech-commands0.727×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.943×
  • tiny-shakespeare-lstm on tiny-shakespeare0.883×
  • uci-har-gru on uci-har0.801×
  • uci-har-lstm on uci-har0.963×