Conjectures.io
Tensor kernels

Entry #67

Completed
Hotkey
5HNDsE…hsRRP3
Sent
11 Nov 2026, 20:43 UTC
Source
5397f405e921

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 runs23.400×Generated 24.639×, real models 22.223×Above the barGain lower bound -2480.5%Guard passed
MLPvs #44Done2 of 2 runs0.954×Generated 1.109×, real models 0.821×Not significantGain lower bound -31.6%Guard passed
ResNetvs #51Done2 of 2 runs102.678×Generated 113.959×, real models 92.514×Above the barGain lower bound -11083.6%Guard passed
MobileNetvs #53Done2 of 2 runs97.289×Generated 114.103×, real models 82.952×Above the barGain lower bound -11259.6%Guard passed
Normalized MLPvs #48Done2 of 2 runs25.654×Generated 32.772×, real models 20.082×Above the barGain lower bound -3185.7%Guard passed
Mixervs #47Done2 of 2 runs0.933×Generated 1.008×, real models 0.864×Not significantGain lower bound -9.5%Guard passed
Recurrentvs #52Done2 of 2 runs764.600×Generated 667.192×, real models 876.228×Above the barGain lower bound -79314.8%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 24.639× · real models 22.223×

MachineTime
  1. prod-amd64Seed 163860868223.705×Done1 attempt0.241× of the baseline kernels' time
  2. prod-arm64Seed 163860868223.098×Done1 attempt0.246× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.986×
  • Batch 1, sparse1.141×
  • Batch 1, wide1.040×

Mixer

  • Batch 1, normal1.009×
  • Batch 1, sparse1.055×
  • Batch 1, wide1.046×

ResNet

  • Batch 1, normal108.973×
  • Batch 1, sparse131.129×
  • Batch 1, wide112.022×

MobileNet

  • Batch 1, normal94.893×
  • Batch 1, sparse96.983×
  • Batch 1, wide94.818×

Recurrent

  • Batch 1, normal710.355×
  • Batch 1, sparse646.558×
  • Batch 1, wide655.947×

Normalized MLP

  • Batch 1, normal28.950×
  • Batch 1, sparse26.895×
  • Batch 1, wide27.592×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.903×
  • compact-mixer on digits1.204×
  • compact-mixer on fashion0.931×
  • compact-mixer on mnist1.015×
  • compact-mlp on digits0.896×
  • compact-mlp on fashion1.297×
  • compact-mlp on mnist1.398×
  • compact-mobilenet on digits64.290×
  • compact-mobilenet on fashion79.755×
  • compact-mobilenet on mnist78.738×
  • compact-normalized_mlp on digits38.676×
  • compact-normalized_mlp on fashion17.960×
  • compact-normalized_mlp on mnist15.044×
  • compact-resnet on digits71.438×
  • compact-resnet on fashion120.763×
  • compact-resnet on mnist116.046×
  • appliance-gru on appliance-energy750.263×
  • mitbih-gru on mitbih909.729×
  • mnist-rows-lstm on mnist-rows726.567×
  • speech-commands-gru on speech-commands697.802×
  • speech-commands-lstm-gru on speech-commands872.581×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1127.680×
  • tiny-shakespeare-lstm on tiny-shakespeare824.847×
  • uci-har-gru on uci-har829.799×
  • uci-har-lstm on uci-har1149.840×
  • uci-har-mixer on UCI-HAR0.972×
  • uci-har-mlp on UCI-HAR1.010×
  • uci-har-mobilenet on UCI-HAR78.416×
  • uci-har-normalized_mlp on UCI-HAR9.353×
  • uci-har-resnet on UCI-HAR74.412×
MLP2 runs

RunsGenerated 1.109× · real models 0.821×

MachineTime
  1. prod-amd64Seed 16386086820.773×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 16386086821.178×Done1 attempt0.017× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.046×
  • Batch 1, sparse0.974×
  • Batch 1, wide1.071×

Real models

  • compact-mlp on digits0.634×
  • compact-mlp on fashion1.026×
  • compact-mlp on mnist0.778×
  • uci-har-mlp on UCI-HAR0.899×
ResNet2 runs

RunsGenerated 113.959× · real models 92.514×

MachineTime
  1. prod-amd64Seed 1638608682105.779×Done1 attempt0.961× of the baseline kernels' time
  2. prod-arm64Seed 163860868299.668×Done1 attempt0.983× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal105.578×
  • Batch 1, sparse129.975×
  • Batch 1, wide111.896×

Real models

  • compact-resnet on digits72.003×
  • compact-resnet on fashion121.068×
  • compact-resnet on mnist115.579×
  • uci-har-resnet on UCI-HAR72.706×
MobileNet2 runs

RunsGenerated 114.103× · real models 82.952×

MachineTime
  1. prod-amd64Seed 163860868291.824×Done1 attempt1.186× of the baseline kernels' time
  2. prod-arm64Seed 1638608682103.079×Done1 attempt1.171× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal136.814×
  • Batch 1, sparse98.994×
  • Batch 1, wide93.097×

Real models

  • compact-mobilenet on digits66.438×
  • compact-mobilenet on fashion95.917×
  • compact-mobilenet on mnist94.506×
  • uci-har-mobilenet on UCI-HAR78.621×
Normalized MLP2 runs

RunsGenerated 32.772× · real models 20.082×

MachineTime
  1. prod-amd64Seed 163860868232.705×Done1 attempt0.271× of the baseline kernels' time
  2. prod-arm64Seed 163860868220.123×Done1 attempt0.283× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal30.405×
  • Batch 1, sparse31.578×
  • Batch 1, wide29.447×

Real models

  • compact-normalized_mlp on digits39.555×
  • compact-normalized_mlp on fashion20.657×
  • compact-normalized_mlp on mnist21.407×
  • uci-har-normalized_mlp on UCI-HAR9.297×
Mixer2 runs

RunsGenerated 1.008× · real models 0.864×

MachineTime
  1. prod-amd64Seed 16386086820.987×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 16386086820.882×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal1.011×
  • Batch 1, sparse1.016×
  • Batch 1, wide1.053×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.939×
  • compact-mixer on digits0.660×
  • compact-mixer on fashion0.934×
  • compact-mixer on mnist0.988×
  • uci-har-mixer on UCI-HAR0.839×
Recurrent2 runs

RunsGenerated 667.192× · real models 876.228×

MachineTime
  1. prod-amd64Seed 1638608682786.190×Done1 attempt7.943× of the baseline kernels' time
  2. prod-arm64Seed 1638608682743.602×Done1 attempt8.097× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal719.928×
  • Batch 1, sparse652.793×
  • Batch 1, wide663.086×

Real models

  • appliance-gru on appliance-energy757.024×
  • mitbih-gru on mitbih915.384×
  • mnist-rows-lstm on mnist-rows735.899×
  • speech-commands-gru on speech-commands706.249×
  • speech-commands-lstm-gru on speech-commands880.175×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1164.752×
  • tiny-shakespeare-lstm on tiny-shakespeare834.249×
  • uci-har-gru on uci-har851.212×
  • uci-har-lstm on uci-har1161.264×