Conjectures.io
Tensor kernels

Entry #42

Completed
Hotkey
5GmaYL…t43iyN
Sent
11 Nov 2026, 09:07 UTC
Source
03d0ff54fcab

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 #43Done2 of 2 runs1.076×Generated 1.050×, real models 1.103×Above the barGain lower bound -15.0%Guard passed
MLPvs #44Done2 of 2 runs0.999×Generated 1.004×, real models 0.994×Above the barGain lower bound -8.3%Guard passed
ResNetvs #45Done2 of 2 runs1.310×Generated 1.405×, real models 1.222×Above the barGain lower bound -48.3%Guard passed
MobileNetvs #42Done2 of 2 runs1.284×Generated 1.498×, real models 1.100×Above the barGain lower bound -64.3%Guard passed
Normalized MLPvs #6Done2 of 2 runs1.101×Generated 1.087×, real models 1.115×Above the barGain lower bound -51.9%Guard passed
Mixervs #24Done2 of 2 runs0.861×Generated 0.804×, real models 0.922×QualifiesGain lower bound 1.7%Guard passed
Recurrentvs #46Done2 of 2 runs1.131×Generated 1.098×, real models 1.165×Above the barGain lower bound -14.9%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 1.050× · real models 1.103×

MachineTime
  1. prod-amd64Seed 29230427131.086×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 29230427131.067×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.993×
  • Batch 1, sparse0.995×
  • Batch 1, wide0.928×

Mixer

  • Batch 1, normal0.892×
  • Batch 1, sparse0.895×
  • Batch 1, wide0.847×

ResNet

  • Batch 1, normal1.173×
  • Batch 1, sparse1.203×
  • Batch 1, wide1.169×

MobileNet

  • Batch 1, normal1.263×
  • Batch 1, sparse1.269×
  • Batch 1, wide1.223×

Recurrent

  • Batch 1, normal1.068×
  • Batch 1, sparse1.058×
  • Batch 1, wide1.079×

Normalized MLP

  • Batch 1, normal1.023×
  • Batch 1, sparse1.047×
  • Batch 1, wide0.877×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.827×
  • compact-mixer on digits2.097×
  • compact-mixer on fashion0.916×
  • compact-mixer on mnist0.899×
  • compact-mlp on digits0.909×
  • compact-mlp on fashion0.790×
  • compact-mlp on mnist1.070×
  • compact-mobilenet on digits1.214×
  • compact-mobilenet on fashion1.025×
  • compact-mobilenet on mnist1.048×
  • compact-normalized_mlp on digits1.457×
  • compact-normalized_mlp on fashion0.896×
  • compact-normalized_mlp on mnist1.526×
  • compact-resnet on digits1.322×
  • compact-resnet on fashion1.108×
  • compact-resnet on mnist1.137×
  • appliance-gru on appliance-energy1.170×
  • mitbih-gru on mitbih1.060×
  • mnist-rows-lstm on mnist-rows1.172×
  • speech-commands-gru on speech-commands1.146×
  • speech-commands-lstm-gru on speech-commands1.112×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.071×
  • tiny-shakespeare-lstm on tiny-shakespeare1.084×
  • uci-har-gru on uci-har1.104×
  • uci-har-lstm on uci-har1.021×
  • uci-har-mixer on UCI-HAR1.064×
  • uci-har-mlp on UCI-HAR1.004×
  • uci-har-mobilenet on UCI-HAR1.164×
  • uci-har-normalized_mlp on UCI-HAR1.094×
  • uci-har-resnet on UCI-HAR1.207×
MLP2 runs

RunsGenerated 1.004× · real models 0.994×

MachineTime
  1. prod-amd64Seed 29230427131.005×Done1 attempt0.016× of the baseline kernels' time
  2. prod-arm64Seed 29230427130.992×Done1 attempt0.016× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.991×
  • Batch 1, sparse1.031×
  • Batch 1, wide0.980×

Real models

  • compact-mlp on digits0.949×
  • compact-mlp on fashion0.997×
  • compact-mlp on mnist0.938×
  • uci-har-mlp on UCI-HAR1.098×
ResNet2 runs

RunsGenerated 1.405× · real models 1.222×

MachineTime
  1. prod-amd64Seed 29230427131.449×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 29230427131.185×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal1.696×
  • Batch 1, sparse1.282×
  • Batch 1, wide1.176×

Real models

  • compact-resnet on digits1.320×
  • compact-resnet on fashion1.143×
  • compact-resnet on mnist1.162×
  • uci-har-resnet on UCI-HAR1.271×
MobileNet2 runs

RunsGenerated 1.498× · real models 1.100×

MachineTime
  1. prod-amd64Seed 29230427131.278×Done1 attempt0.017× of the baseline kernels' time
  2. prod-arm64Seed 29230427131.290×Done1 attempt0.015× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.777×
  • Batch 1, sparse1.307×
  • Batch 1, wide1.219×

Real models

  • compact-mobilenet on digits1.173×
  • compact-mobilenet on fashion1.026×
  • compact-mobilenet on mnist1.044×
  • uci-har-mobilenet on UCI-HAR1.168×
Normalized MLP2 runs

RunsGenerated 1.087× · real models 1.115×

MachineTime
  1. prod-amd64Seed 29230427131.164×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 29230427131.042×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.087×
  • Batch 1, sparse1.101×
  • Batch 1, wide1.153×

Real models

  • compact-normalized_mlp on digits1.088×
  • compact-normalized_mlp on fashion1.402×
  • compact-normalized_mlp on mnist0.947×
  • uci-har-normalized_mlp on UCI-HAR1.070×
Mixer2 runs

RunsGenerated 0.804× · real models 0.922×

MachineTime
  1. prod-amd64Seed 29230427130.815×Done1 attempt0.004× of the baseline kernels' time
  2. prod-arm64Seed 29230427130.910×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.924×
  • Batch 1, sparse0.816×
  • Batch 1, wide0.746×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.678×
  • compact-mixer on digits1.156×
  • compact-mixer on fashion0.920×
  • compact-mixer on mnist0.902×
  • uci-har-mixer on UCI-HAR1.027×
Recurrent2 runs

RunsGenerated 1.098× · real models 1.165×

MachineTime
  1. prod-amd64Seed 29230427131.155×Done1 attempt0.015× of the baseline kernels' time
  2. prod-arm64Seed 29230427131.107×Done1 attempt0.015× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal1.104×
  • Batch 1, sparse1.095×
  • Batch 1, wide1.089×

Real models

  • appliance-gru on appliance-energy1.214×
  • mitbih-gru on mitbih1.153×
  • mnist-rows-lstm on mnist-rows1.186×
  • speech-commands-gru on speech-commands1.320×
  • speech-commands-lstm-gru on speech-commands1.193×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.055×
  • tiny-shakespeare-lstm on tiny-shakespeare1.126×
  • uci-har-gru on uci-har1.177×
  • uci-har-lstm on uci-har1.080×