Conjectures.io
Tensor kernels

Entry #1

Completed
Hotkey
5Gs2Eh…TDn5RP
Sent
10 Nov 2026, 17:41 UTC
Source
d93430858bb4

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 the shipped referenceDone2 of 2 runs0.457×Generated 0.444×, real models 0.471×QualifiesGain lower bound 52.7%Guard passed
MLPvs the shipped referenceDone2 of 2 runs0.021×Generated 0.017×, real models 0.025×QualifiesGain lower bound 97.4%Guard passed
ResNetvs the shipped referenceDone2 of 2 runs0.967×Generated 0.973×, real models 0.961×QualifiesGain lower bound 2.3%Guard passed
MobileNetvs the shipped referenceDone2 of 2 runs1.003×Generated 1.003×, real models 1.003×Above the barGain lower bound -0.6%Guard passed
Normalized MLPvs the shipped referenceDone2 of 2 runs0.505×Generated 0.537×, real models 0.475×QualifiesGain lower bound 43.0%Guard passed
Mixervs the shipped referenceDone2 of 2 runs0.779×Generated 0.800×, real models 0.758×QualifiesGain lower bound 19.8%Guard passed
Recurrentvs the shipped referenceDone2 of 2 runs1.003×Generated 1.004×, real models 1.002×Above the barGain lower bound -0.8%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 0.444× · real models 0.471×

MachineTime
  1. prod-amd64Seed 29638907920.458×Done1 attempt0.440× of the baseline kernels' time
  2. prod-arm64Seed 29638907920.457×Done1 attempt0.444× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.018×
  • Batch 1, sparse0.020×
  • Batch 1, wide0.018×

Mixer

  • Batch 1, normal0.790×
  • Batch 1, sparse0.796×
  • Batch 1, wide0.816×

ResNet

  • Batch 1, normal0.973×
  • Batch 1, sparse0.974×
  • Batch 1, wide0.975×

MobileNet

  • Batch 1, normal1.005×
  • Batch 1, sparse1.007×
  • Batch 1, wide1.006×

Recurrent

  • Batch 1, normal0.995×
  • Batch 1, sparse1.001×
  • Batch 1, wide1.003×

Normalized MLP

  • Batch 1, normal0.520×
  • Batch 1, sparse0.557×
  • Batch 1, wide0.521×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.849×
  • compact-mixer on digits0.884×
  • compact-mixer on fashion0.704×
  • compact-mixer on mnist0.697×
  • compact-mlp on digits0.047×
  • compact-mlp on fashion0.035×
  • compact-mlp on mnist0.039×
  • compact-mobilenet on digits1.011×
  • compact-mobilenet on fashion1.013×
  • compact-mobilenet on mnist1.006×
  • compact-normalized_mlp on digits0.811×
  • compact-normalized_mlp on fashion0.436×
  • compact-normalized_mlp on mnist0.502×
  • compact-resnet on digits0.976×
  • compact-resnet on fashion0.978×
  • compact-resnet on mnist0.976×
  • appliance-gru on appliance-energy0.999×
  • mitbih-gru on mitbih1.007×
  • mnist-rows-lstm on mnist-rows0.996×
  • speech-commands-gru on speech-commands1.012×
  • speech-commands-lstm-gru on speech-commands1.005×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.995×
  • tiny-shakespeare-lstm on tiny-shakespeare1.000×
  • uci-har-gru on uci-har1.010×
  • uci-har-lstm on uci-har0.996×
  • uci-har-mixer on UCI-HAR0.676×
  • uci-har-mlp on UCI-HAR0.015×
  • uci-har-mobilenet on UCI-HAR0.992×
  • uci-har-normalized_mlp on UCI-HAR0.287×
  • uci-har-resnet on UCI-HAR0.924×
MLP2 runs

RunsGenerated 0.017× · real models 0.025×

MachineTime
  1. prod-amd64Seed 29638907920.021×Done1 attempt0.017× of the baseline kernels' time
  2. prod-arm64Seed 29638907920.021×Done1 attempt0.018× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.016×
  • Batch 1, sparse0.020×
  • Batch 1, wide0.017×

Real models

  • compact-mlp on digits0.044×
  • compact-mlp on fashion0.022×
  • compact-mlp on mnist0.027×
  • uci-har-mlp on UCI-HAR0.014×
ResNet2 runs

RunsGenerated 0.973× · real models 0.961×

MachineTime
  1. prod-amd64Seed 29638907920.963×Done1 attempt0.967× of the baseline kernels' time
  2. prod-arm64Seed 29638907920.972×Done1 attempt0.976× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal0.972×
  • Batch 1, sparse0.974×
  • Batch 1, wide0.973×

Real models

  • compact-resnet on digits0.973×
  • compact-resnet on fashion0.975×
  • compact-resnet on mnist0.975×
  • uci-har-resnet on UCI-HAR0.923×
MobileNet2 runs

RunsGenerated 1.003× · real models 1.003×

MachineTime
  1. prod-amd64Seed 29638907921.005×Done1 attempt1.001× of the baseline kernels' time
  2. prod-arm64Seed 29638907921.001×Done1 attempt1.001× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.004×
  • Batch 1, sparse1.004×
  • Batch 1, wide1.001×

Real models

  • compact-mobilenet on digits1.005×
  • compact-mobilenet on fashion1.011×
  • compact-mobilenet on mnist1.006×
  • uci-har-mobilenet on UCI-HAR0.991×
Normalized MLP2 runs

RunsGenerated 0.537× · real models 0.475×

MachineTime
  1. prod-amd64Seed 29638907920.504×Done1 attempt0.533× of the baseline kernels' time
  2. prod-arm64Seed 29638907920.506×Done1 attempt0.535× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal0.524×
  • Batch 1, sparse0.557×
  • Batch 1, wide0.520×

Real models

  • compact-normalized_mlp on digits0.810×
  • compact-normalized_mlp on fashion0.441×
  • compact-normalized_mlp on mnist0.501×
  • uci-har-normalized_mlp on UCI-HAR0.284×
Mixer2 runs

RunsGenerated 0.800× · real models 0.758×

MachineTime
  1. prod-amd64Seed 29638907920.774×Done1 attempt0.794× of the baseline kernels' time
  2. prod-arm64Seed 29638907920.783×Done1 attempt0.800× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.788×
  • Batch 1, sparse0.794×
  • Batch 1, wide0.817×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.848×
  • compact-mixer on digits0.880×
  • compact-mixer on fashion0.705×
  • compact-mixer on mnist0.700×
  • uci-har-mixer on UCI-HAR0.677×
Recurrent2 runs

RunsGenerated 1.004× · real models 1.002×

MachineTime
  1. prod-amd64Seed 29638907921.010×Done1 attempt1.005× of the baseline kernels' time
  2. prod-arm64Seed 29638907920.996×Done1 attempt0.984× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

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

Real models

  • appliance-gru on appliance-energy1.004×
  • mitbih-gru on mitbih1.011×
  • mnist-rows-lstm on mnist-rows0.996×
  • speech-commands-gru on speech-commands1.011×
  • speech-commands-lstm-gru on speech-commands0.999×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.994×
  • tiny-shakespeare-lstm on tiny-shakespeare0.997×
  • uci-har-gru on uci-har1.006×
  • uci-har-lstm on uci-har0.997×