Conjectures.io
Tensor kernels

Entry #77

Completed
Hotkey
5HVyMG…Nw9BA4
Sent
12 Nov 2026, 02:03 UTC
Source
50863a107f8b

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 #50Done2 of 2 runs16.350×Generated 17.400×, real models 15.363×Above the barGain lower bound -1646.3%Guard passed
MLPvs #44Done2 of 2 runs0.961×Generated 1.020×, real models 0.905×Not significantGain lower bound -24.2%Guard passed
ResNetvs #51Done2 of 2 runs3.725×Generated 3.752×, real models 3.699×Above the barGain lower bound -359.5%Guard passed
MobileNetvs #53Done2 of 2 runs0.964×Generated 0.981×, real models 0.948×Not significantGain lower bound -0.3%Guard passed
Normalized MLPvs #48Done2 of 2 runs45.117×Generated 62.690×, real models 32.469×Above the barGain lower bound -5292.2%Guard passed
Mixervs #47Done2 of 2 runs157.688×Generated 194.174×, real models 128.057×Above the barGain lower bound -18140.0%Guard passed
Recurrentvs #54Done2 of 2 runs822.213×Generated 709.614×, real models 952.678×Above the barGain lower bound -85947.9%Guard passed

The source stays private while it holds a board, and for entries that never held one.

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 17.400× · real models 15.363×

MachineTime
  1. prod-amd64Seed 340915229916.105×Done1 attempt0.165× of the baseline kernels' time
  2. prod-arm64Seed 340915229916.598×Done1 attempt0.175× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.086×
  • Batch 1, sparse1.003×
  • Batch 1, wide1.166×

Mixer

  • Batch 1, normal167.142×
  • Batch 1, sparse167.450×
  • Batch 1, wide217.869×

ResNet

  • Batch 1, normal3.542×
  • Batch 1, sparse4.076×
  • Batch 1, wide3.597×

MobileNet

  • Batch 1, normal1.005×
  • Batch 1, sparse1.001×
  • Batch 1, wide1.020×

Recurrent

  • Batch 1, normal706.057×
  • Batch 1, sparse662.429×
  • Batch 1, wide661.332×

Normalized MLP

  • Batch 1, normal62.285×
  • Batch 1, sparse61.806×
  • Batch 1, wide65.554×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction221.820×
  • compact-mixer on digits180.401×
  • compact-mixer on fashion166.510×
  • compact-mixer on mnist151.415×
  • compact-mlp on digits1.465×
  • compact-mlp on fashion1.750×
  • compact-mlp on mnist0.828×
  • compact-mobilenet on digits0.853×
  • compact-mobilenet on fashion0.854×
  • compact-mobilenet on mnist0.814×
  • compact-normalized_mlp on digits82.648×
  • compact-normalized_mlp on fashion26.913×
  • compact-normalized_mlp on mnist8.937×
  • compact-resnet on digits2.491×
  • compact-resnet on fashion3.555×
  • compact-resnet on mnist3.008×
  • appliance-gru on appliance-energy768.215×
  • mitbih-gru on mitbih888.262×
  • mnist-rows-lstm on mnist-rows725.975×
  • speech-commands-gru on speech-commands684.124×
  • speech-commands-lstm-gru on speech-commands858.979×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1102.061×
  • tiny-shakespeare-lstm on tiny-shakespeare828.036×
  • uci-har-gru on uci-har842.565×
  • uci-har-lstm on uci-har1131.592×
  • uci-har-mixer on UCI-HAR167.006×
  • uci-har-mlp on UCI-HAR1.008×
  • uci-har-mobilenet on UCI-HAR0.886×
  • uci-har-normalized_mlp on UCI-HAR20.202×
  • uci-har-resnet on UCI-HAR4.771×
MLP2 runs

RunsGenerated 1.020× · real models 0.905×

MachineTime
  1. prod-amd64Seed 34091522991.038×Done1 attempt0.015× of the baseline kernels' time
  2. prod-arm64Seed 34091522990.889×Done1 attempt0.016× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.929×
  • Batch 1, sparse0.961×
  • Batch 1, wide1.171×

Real models

  • compact-mlp on digits0.552×
  • compact-mlp on fashion1.193×
  • compact-mlp on mnist1.010×
  • uci-har-mlp on UCI-HAR1.007×
ResNet2 runs

RunsGenerated 3.752× · real models 3.699×

MachineTime
  1. prod-amd64Seed 34091522993.839×Done1 attempt0.031× of the baseline kernels' time
  2. prod-arm64Seed 34091522993.614×Done1 attempt0.033× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal3.753×
  • Batch 1, sparse4.035×
  • Batch 1, wide3.545×

Real models

  • compact-resnet on digits2.470×
  • compact-resnet on fashion3.756×
  • compact-resnet on mnist3.840×
  • uci-har-resnet on UCI-HAR5.253×
MobileNet2 runs

RunsGenerated 0.981× · real models 0.948×

MachineTime
  1. prod-amd64Seed 34091522990.943×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 34091522990.985×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.003×
  • Batch 1, sparse0.987×
  • Batch 1, wide0.984×

Real models

  • compact-mobilenet on digits0.884×
  • compact-mobilenet on fashion0.983×
  • compact-mobilenet on mnist0.965×
  • uci-har-mobilenet on UCI-HAR0.960×
Normalized MLP2 runs

RunsGenerated 62.690× · real models 32.469×

MachineTime
  1. prod-amd64Seed 340915229942.450×Done1 attempt0.533× of the baseline kernels' time
  2. prod-arm64Seed 340915229947.952×Done1 attempt0.537× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal61.959×
  • Batch 1, sparse64.169×
  • Batch 1, wide64.013×

Real models

  • compact-normalized_mlp on digits83.344×
  • compact-normalized_mlp on fashion22.658×
  • compact-normalized_mlp on mnist24.592×
  • uci-har-normalized_mlp on UCI-HAR23.933×
Mixer2 runs

RunsGenerated 194.174× · real models 128.057×

MachineTime
  1. prod-amd64Seed 3409152299164.207×Done1 attempt0.852× of the baseline kernels' time
  2. prod-arm64Seed 3409152299151.427×Done1 attempt0.854× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal186.420×
  • Batch 1, sparse192.771×
  • Batch 1, wide222.226×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction227.009×
  • compact-mixer on digits44.282×
  • compact-mixer on fashion156.350×
  • compact-mixer on mnist151.124×
  • uci-har-mixer on UCI-HAR144.984×
Recurrent2 runs

RunsGenerated 709.614× · real models 952.678×

MachineTime
  1. prod-amd64Seed 3409152299878.766×Done1 attempt8.143× of the baseline kernels' time
  2. prod-arm64Seed 3409152299769.299×Done1 attempt8.220× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal720.551×
  • Batch 1, sparse718.474×
  • Batch 1, wide698.353×

Real models

  • appliance-gru on appliance-energy768.340×
  • mitbih-gru on mitbih881.179×
  • mnist-rows-lstm on mnist-rows1027.681×
  • speech-commands-gru on speech-commands664.859×
  • speech-commands-lstm-gru on speech-commands861.100×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1257.389×
  • tiny-shakespeare-lstm on tiny-shakespeare1234.128×
  • uci-har-gru on uci-har822.996×
  • uci-har-lstm on uci-har1270.654×