Conjectures.io
Tensor kernels

Entry #92

Completed
Hotkey
5HYsKB…nBa4r9
Sent
12 Nov 2026, 07:49 UTC
Source
2990bf9d3402

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 runs0.989×Generated 0.965×, real models 1.013×Above the barGain lower bound -3.4%Guard passed
MLPvs #44Done2 of 2 runs1.051×Generated 0.965×, real models 1.144×Above the barGain lower bound -26.3%Guard passed
ResNetvs #51Done2 of 2 runs1.257×Generated 1.606×, real models 0.984×Above the barGain lower bound -67.9%Guard passed
MobileNetvs #53Done2 of 2 runs1.154×Generated 1.219×, real models 1.093×Above the barGain lower bound -67.0%Guard passed
Normalized MLPvs #48Done2 of 2 runs0.927×Generated 1.028×, real models 0.835×Not significantGain lower bound -3.6%Guard passed
Mixervs #47Done2 of 2 runs0.984×Generated 0.914×, real models 1.059×Above the barGain lower bound -11.8%Guard passed
Recurrentvs #55Done2 of 2 runs0.990×Generated 0.989×, real models 0.990×Above the barGain lower bound -0.9%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 0.965× · real models 1.013×

MachineTime
  1. prod-amd64Seed 25362553000.976×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 25362553001.002×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.017×
  • Batch 1, sparse0.984×
  • Batch 1, wide1.000×

Mixer

  • Batch 1, normal0.972×
  • Batch 1, sparse0.863×
  • Batch 1, wide0.904×

ResNet

  • Batch 1, normal0.994×
  • Batch 1, sparse1.048×
  • Batch 1, wide1.000×

MobileNet

  • Batch 1, normal1.052×
  • Batch 1, sparse1.082×
  • Batch 1, wide1.059×

Recurrent

  • Batch 1, normal0.871×
  • Batch 1, sparse0.922×
  • Batch 1, wide0.895×

Normalized MLP

  • Batch 1, normal1.032×
  • Batch 1, sparse1.047×
  • Batch 1, wide1.128×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.237×
  • compact-mixer on digits1.616×
  • compact-mixer on fashion1.019×
  • compact-mixer on mnist1.003×
  • compact-mlp on digits1.247×
  • compact-mlp on fashion0.998×
  • compact-mlp on mnist0.758×
  • compact-mobilenet on digits0.889×
  • compact-mobilenet on fashion0.925×
  • compact-mobilenet on mnist0.918×
  • compact-normalized_mlp on digits0.983×
  • compact-normalized_mlp on fashion1.068×
  • compact-normalized_mlp on mnist1.013×
  • compact-resnet on digits0.911×
  • compact-resnet on fashion0.944×
  • compact-resnet on mnist0.972×
  • appliance-gru on appliance-energy0.947×
  • mitbih-gru on mitbih0.911×
  • mnist-rows-lstm on mnist-rows0.947×
  • speech-commands-gru on speech-commands0.899×
  • speech-commands-lstm-gru on speech-commands0.876×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.866×
  • tiny-shakespeare-lstm on tiny-shakespeare0.911×
  • uci-har-gru on uci-har0.927×
  • uci-har-lstm on uci-har0.843×
  • uci-har-mixer on UCI-HAR1.087×
  • uci-har-mlp on UCI-HAR1.037×
  • uci-har-mobilenet on UCI-HAR1.222×
  • uci-har-normalized_mlp on UCI-HAR1.225×
  • uci-har-resnet on UCI-HAR1.106×
MLP2 runs

RunsGenerated 0.965× · real models 1.144×

MachineTime
  1. prod-amd64Seed 25362553000.986×Done1 attempt0.016× of the baseline kernels' time
  2. prod-arm64Seed 25362553001.120×Done1 attempt0.018× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.985×
  • Batch 1, sparse0.896×
  • Batch 1, wide0.941×

Real models

  • compact-mlp on digits0.974×
  • compact-mlp on fashion1.350×
  • compact-mlp on mnist1.240×
  • uci-har-mlp on UCI-HAR1.051×
ResNet2 runs

RunsGenerated 1.606× · real models 0.984×

MachineTime
  1. prod-amd64Seed 25362553001.017×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 25362553001.554×Done1 attempt0.009× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal1.442×
  • Batch 1, sparse1.387×
  • Batch 1, wide1.383×

Real models

  • compact-resnet on digits0.915×
  • compact-resnet on fashion0.941×
  • compact-resnet on mnist0.980×
  • uci-har-resnet on UCI-HAR1.112×
MobileNet2 runs

RunsGenerated 1.219× · real models 1.093×

MachineTime
  1. prod-amd64Seed 25362553001.050×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 25362553001.269×Done1 attempt0.014× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.431×
  • Batch 1, sparse1.108×
  • Batch 1, wide1.317×

Real models

  • compact-mobilenet on digits0.976×
  • compact-mobilenet on fashion1.068×
  • compact-mobilenet on mnist1.099×
  • uci-har-mobilenet on UCI-HAR1.246×
Normalized MLP2 runs

RunsGenerated 1.028× · real models 0.835×

MachineTime
  1. prod-amd64Seed 25362553000.948×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 25362553000.905×Done1 attempt0.015× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.043×
  • Batch 1, sparse1.068×
  • Batch 1, wide1.008×

Real models

  • compact-normalized_mlp on digits0.565×
  • compact-normalized_mlp on fashion0.900×
  • compact-normalized_mlp on mnist0.961×
  • uci-har-normalized_mlp on UCI-HAR0.998×
Mixer2 runs

RunsGenerated 0.914× · real models 1.059×

MachineTime
  1. prod-amd64Seed 25362553000.898×Done1 attempt0.004× of the baseline kernels' time
  2. prod-arm64Seed 25362553001.078×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.943×
  • Batch 1, sparse0.944×
  • Batch 1, wide1.000×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.142×
  • compact-mixer on digits1.265×
  • compact-mixer on fashion1.050×
  • compact-mixer on mnist0.989×
  • uci-har-mixer on UCI-HAR0.887×
Recurrent2 runs

RunsGenerated 0.989× · real models 0.990×

MachineTime
  1. prod-amd64Seed 25362553000.983×Done1 attempt0.010× of the baseline kernels' time
  2. prod-arm64Seed 25362553000.996×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.961×
  • Batch 1, sparse0.990×
  • Batch 1, wide0.983×

Real models

  • appliance-gru on appliance-energy1.020×
  • mitbih-gru on mitbih1.004×
  • mnist-rows-lstm on mnist-rows0.946×
  • speech-commands-gru on speech-commands0.994×
  • speech-commands-lstm-gru on speech-commands0.982×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.996×
  • tiny-shakespeare-lstm on tiny-shakespeare0.992×
  • uci-har-gru on uci-har0.994×
  • uci-har-lstm on uci-har0.985×