Conjectures.io
Tensor kernels

Entry #25

Completed
Hotkey
5FFLei…5Nouyw
Sent
11 Nov 2026, 01:46 UTC
Source
9dcb55c8ebba

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 runs0.958×Generated 1.047×, real models 0.877×Not significantGain lower bound -4.7%Guard passed
MLPvs #1Done2 of 2 runs0.984×Generated 0.940×, real models 1.031×Above the barGain lower bound -11.6%Guard passed
ResNetvs #30Done2 of 2 runs0.894×Generated 1.011×, real models 0.790×QualifiesGain lower bound 4.5%Guard passed
MobileNetvs #31Done2 of 2 runs0.820×Generated 1.009×, real models 0.667×QualifiesGain lower bound 12.6%Guard passed
Normalized MLPvs #6Done2 of 2 runs0.965×Generated 0.994×, real models 0.937×Not significantGain lower bound -5.1%Guard passed
Mixervs #24Done2 of 2 runs1.079×Generated 0.983×, real models 1.184×Above the barGain lower bound -22.1%Guard passed
Recurrentvs #25Done2 of 2 runs0.891×Generated 1.007×, real models 0.789×Not significantGain lower bound -2.4%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 1.047× · real models 0.877×

MachineTime
  1. prod-amd64Seed 33674806830.956×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 33674806830.960×Done1 attempt0.015× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.108×
  • Batch 1, sparse1.128×
  • Batch 1, wide0.793×

Mixer

  • Batch 1, normal1.011×
  • Batch 1, sparse0.974×
  • Batch 1, wide1.358×

ResNet

  • Batch 1, normal0.993×
  • Batch 1, sparse1.033×
  • Batch 1, wide0.988×

MobileNet

  • Batch 1, normal0.980×
  • Batch 1, sparse1.015×
  • Batch 1, wide0.950×

Recurrent

  • Batch 1, normal1.018×
  • Batch 1, sparse1.003×
  • Batch 1, wide1.017×

Normalized MLP

  • Batch 1, normal1.235×
  • Batch 1, sparse1.295×
  • Batch 1, wide1.087×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.185×
  • compact-mixer on digits2.098×
  • compact-mixer on fashion1.009×
  • compact-mixer on mnist0.881×
  • compact-mlp on digits1.023×
  • compact-mlp on fashion0.994×
  • compact-mlp on mnist1.969×
  • compact-mobilenet on digits0.806×
  • compact-mobilenet on fashion0.801×
  • compact-mobilenet on mnist0.734×
  • compact-normalized_mlp on digits0.971×
  • compact-normalized_mlp on fashion0.815×
  • compact-normalized_mlp on mnist0.825×
  • compact-resnet on digits0.945×
  • compact-resnet on fashion0.884×
  • compact-resnet on mnist0.816×
  • appliance-gru on appliance-energy1.118×
  • mitbih-gru on mitbih0.980×
  • mnist-rows-lstm on mnist-rows0.680×
  • speech-commands-gru on speech-commands1.289×
  • speech-commands-lstm-gru on speech-commands1.037×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.426×
  • tiny-shakespeare-lstm on tiny-shakespeare0.705×
  • uci-har-gru on uci-har1.094×
  • uci-har-lstm on uci-har0.451×
  • uci-har-mixer on UCI-HAR1.059×
  • uci-har-mlp on UCI-HAR0.916×
  • uci-har-mobilenet on UCI-HAR0.282×
  • uci-har-normalized_mlp on UCI-HAR0.958×
  • uci-har-resnet on UCI-HAR0.484×
MLP2 runs

RunsGenerated 0.940× · real models 1.031×

MachineTime
  1. prod-amd64Seed 33674806830.919×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 33674806831.055×Done1 attempt0.034× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.994×
  • Batch 1, sparse1.030×
  • Batch 1, wide0.905×

Real models

  • compact-mlp on digits1.616×
  • compact-mlp on fashion0.972×
  • compact-mlp on mnist0.854×
  • uci-har-mlp on UCI-HAR0.843×
ResNet2 runs

RunsGenerated 1.011× · real models 0.790×

MachineTime
  1. prod-amd64Seed 33674806830.886×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 33674806830.902×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal1.002×
  • Batch 1, sparse0.996×
  • Batch 1, wide1.016×

Real models

  • compact-resnet on digits0.895×
  • compact-resnet on fashion0.744×
  • compact-resnet on mnist0.596×
  • uci-har-resnet on UCI-HAR0.982×
MobileNet2 runs

RunsGenerated 1.009× · real models 0.667×

MachineTime
  1. prod-amd64Seed 33674806830.839×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 33674806830.802×Done1 attempt0.015× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.003×
  • Batch 1, sparse1.004×
  • Batch 1, wide1.007×

Real models

  • compact-mobilenet on digits0.826×
  • compact-mobilenet on fashion0.581×
  • compact-mobilenet on mnist0.501×
  • uci-har-mobilenet on UCI-HAR0.823×
Normalized MLP2 runs

RunsGenerated 0.994× · real models 0.937×

MachineTime
  1. prod-amd64Seed 33674806830.939×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 33674806830.992×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal0.980×
  • Batch 1, sparse1.000×
  • Batch 1, wide0.969×

Real models

  • compact-normalized_mlp on digits1.056×
  • compact-normalized_mlp on fashion0.964×
  • compact-normalized_mlp on mnist0.890×
  • uci-har-normalized_mlp on UCI-HAR0.851×
Mixer2 runs

RunsGenerated 0.983× · real models 1.184×

MachineTime
  1. prod-amd64Seed 33674806831.126×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 33674806831.034×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.971×
  • Batch 1, sparse1.022×
  • Batch 1, wide0.964×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.086×
  • compact-mixer on digits2.204×
  • compact-mixer on fashion0.979×
  • compact-mixer on mnist0.982×
  • uci-har-mixer on UCI-HAR1.013×
Recurrent2 runs

RunsGenerated 1.007× · real models 0.789×

MachineTime
  1. prod-amd64Seed 33674806830.883×Done1 attempt0.043× of the baseline kernels' time
  2. prod-arm64Seed 33674806830.900×Done1 attempt0.045× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.999×
  • Batch 1, sparse0.996×
  • Batch 1, wide1.008×

Real models

  • appliance-gru on appliance-energy1.100×
  • mitbih-gru on mitbih0.971×
  • mnist-rows-lstm on mnist-rows0.679×
  • speech-commands-gru on speech-commands1.285×
  • speech-commands-lstm-gru on speech-commands1.044×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.427×
  • tiny-shakespeare-lstm on tiny-shakespeare0.685×
  • uci-har-gru on uci-har1.089×
  • uci-har-lstm on uci-har0.382×