Conjectures.io
Tensor kernels

Entry #127

Completed
Hotkey
5G17vb…MHPNEh
Sent
12 Nov 2026, 23:19 UTC
Source
e896fd81b948

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 #61Done2 of 2 runs81.084×Generated 85.806×, real models 76.622×Above the barGain lower bound -8520.5%Guard passed
MLPvs #44Done2 of 2 runs1.059×Generated 1.015×, real models 1.105×Above the barGain lower bound -28.5%Guard passed
ResNetvs #59Done2 of 2 runs5508.142×Generated 6101.825×, real models 4972.221×Above the barGain lower bound -642268.8%Guard passed
MobileNetvs #53Done2 of 2 runs2943.139×Generated 3702.914×, real models 2339.256×Above the barGain lower bound -357024.8%Guard passed
Normalized MLPvs #48Done2 of 2 runs40.553×Generated 51.273×, real models 32.074×Above the barGain lower bound -5374.1%Guard passed
Mixervs #58Done2 of 2 runs624.494×Generated 594.655×, real models 655.830×Above the barGain lower bound -78028.2%Guard passed
Recurrentvs #62Done2 of 2 runs0.992×Generated 1.018×, real models 0.967×Above the barGain lower bound -4.3%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 85.806× · real models 76.622×

MachineTime
  1. prod-amd64Seed 344222924884.692×Done1 attempt0.851× of the baseline kernels' time
  2. prod-arm64Seed 344222924877.630×Done1 attempt0.864× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.102×
  • Batch 1, sparse1.035×
  • Batch 1, wide1.051×

Mixer

  • Batch 1, normal635.974×
  • Batch 1, sparse600.443×
  • Batch 1, wide663.532×

ResNet

  • Batch 1, normal6053.780×
  • Batch 1, sparse6538.930×
  • Batch 1, wide6085.928×

MobileNet

  • Batch 1, normal2672.732×
  • Batch 1, sparse2510.761×
  • Batch 1, wide2610.753×

Recurrent

  • Batch 1, normal0.863×
  • Batch 1, sparse0.823×
  • Batch 1, wide0.847×

Normalized MLP

  • Batch 1, normal50.090×
  • Batch 1, sparse52.148×
  • Batch 1, wide47.206×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction922.182×
  • compact-mixer on digits401.344×
  • compact-mixer on fashion659.502×
  • compact-mixer on mnist621.822×
  • compact-mlp on digits0.867×
  • compact-mlp on fashion0.594×
  • compact-mlp on mnist2.174×
  • compact-mobilenet on digits1960.394×
  • compact-mobilenet on fashion2552.004×
  • compact-mobilenet on mnist2519.530×
  • compact-normalized_mlp on digits84.166×
  • compact-normalized_mlp on fashion18.786×
  • compact-normalized_mlp on mnist25.007×
  • compact-resnet on digits5243.549×
  • compact-resnet on fashion7721.322×
  • compact-resnet on mnist7744.452×
  • appliance-gru on appliance-energy0.853×
  • mitbih-gru on mitbih0.950×
  • mnist-rows-lstm on mnist-rows0.814×
  • speech-commands-gru on speech-commands0.853×
  • speech-commands-lstm-gru on speech-commands0.872×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.842×
  • tiny-shakespeare-lstm on tiny-shakespeare0.992×
  • uci-har-gru on uci-har0.917×
  • uci-har-lstm on uci-har0.852×
  • uci-har-mixer on UCI-HAR623.679×
  • uci-har-mlp on UCI-HAR1.001×
  • uci-har-mobilenet on UCI-HAR1945.688×
  • uci-har-normalized_mlp on UCI-HAR22.033×
  • uci-har-resnet on UCI-HAR2443.732×
MLP2 runs

RunsGenerated 1.015× · real models 1.105×

MachineTime
  1. prod-amd64Seed 34422292480.985×Done1 attempt0.017× of the baseline kernels' time
  2. prod-arm64Seed 34422292481.137×Done1 attempt0.018× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.041×
  • Batch 1, sparse1.044×
  • Batch 1, wide1.013×

Real models

  • compact-mlp on digits1.951×
  • compact-mlp on fashion0.993×
  • compact-mlp on mnist0.748×
  • uci-har-mlp on UCI-HAR1.027×
ResNet2 runs

RunsGenerated 6101.825× · real models 4972.221×

MachineTime
  1. prod-amd64Seed 34422292485545.767×Done1 attempt51.748× of the baseline kernels' time
  2. prod-arm64Seed 34422292485470.771×Done1 attempt52.195× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal6062.664×
  • Batch 1, sparse6440.448×
  • Batch 1, wide6162.018×

Real models

  • compact-resnet on digits4880.974×
  • compact-resnet on fashion7642.267×
  • compact-resnet on mnist7495.112×
  • uci-har-resnet on UCI-HAR2186.225×
MobileNet2 runs

RunsGenerated 3702.914× · real models 2339.256×

MachineTime
  1. prod-amd64Seed 34422292482569.098×Done1 attempt33.062× of the baseline kernels' time
  2. prod-arm64Seed 34422292483371.637×Done1 attempt33.292× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal3757.608×
  • Batch 1, sparse3521.690×
  • Batch 1, wide3548.036×

Real models

  • compact-mobilenet on digits1787.766×
  • compact-mobilenet on fashion2639.571×
  • compact-mobilenet on mnist2669.940×
  • uci-har-mobilenet on UCI-HAR2376.653×
Normalized MLP2 runs

RunsGenerated 51.273× · real models 32.074×

MachineTime
  1. prod-amd64Seed 344222924840.132×Done1 attempt0.533× of the baseline kernels' time
  2. prod-arm64Seed 344222924840.978×Done1 attempt0.530× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal52.359×
  • Batch 1, sparse55.224×
  • Batch 1, wide51.943×

Real models

  • compact-normalized_mlp on digits74.591×
  • compact-normalized_mlp on fashion15.959×
  • compact-normalized_mlp on mnist36.334×
  • uci-har-normalized_mlp on UCI-HAR24.468×
Mixer2 runs

RunsGenerated 594.655× · real models 655.830×

MachineTime
  1. prod-amd64Seed 3442229248630.630×Done1 attempt2.801× of the baseline kernels' time
  2. prod-arm64Seed 3442229248618.417×Done1 attempt2.801× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal656.148×
  • Batch 1, sparse627.807×
  • Batch 1, wide646.572×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1014.819×
  • compact-mixer on digits502.977×
  • compact-mixer on fashion618.227×
  • compact-mixer on mnist629.586×
  • uci-har-mixer on UCI-HAR610.683×
Recurrent2 runs

RunsGenerated 1.018× · real models 0.967×

MachineTime
  1. prod-amd64Seed 34422292481.032×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 34422292480.954×Done1 attempt0.009× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal1.022×
  • Batch 1, sparse1.053×
  • Batch 1, wide0.976×

Real models

  • appliance-gru on appliance-energy1.005×
  • mitbih-gru on mitbih0.977×
  • mnist-rows-lstm on mnist-rows0.990×
  • speech-commands-gru on speech-commands0.939×
  • speech-commands-lstm-gru on speech-commands0.962×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.937×
  • tiny-shakespeare-lstm on tiny-shakespeare0.974×
  • uci-har-gru on uci-har0.968×
  • uci-har-lstm on uci-har0.955×