Conjectures.io
Tensor kernels

Entry #116

Completed
Hotkey
5FCgYk…JApLBK
Sent
12 Nov 2026, 16:34 UTC
Source
9c9201174af2

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 runs84.480×Generated 92.004×, real models 77.570×Above the barGain lower bound -9026.8%Guard passed
MLPvs #44Done2 of 2 runs1.026×Generated 1.077×, real models 0.978×Above the barGain lower bound -11.4%Guard passed
ResNetvs #59Done2 of 2 runs5584.869×Generated 6022.796×, real models 5178.784×Above the barGain lower bound -640360.6%Guard passed
MobileNetvs #53Done2 of 2 runs2531.391×Generated 2734.378×, real models 2343.474×Above the barGain lower bound -288994.6%Guard passed
Normalized MLPvs #48Done2 of 2 runs55.711×Generated 78.637×, real models 39.469×Above the barGain lower bound -6704.1%Guard passed
Mixervs #58Done2 of 2 runs667.948×Generated 665.055×, real models 670.853×Above the barGain lower bound -74126.1%Guard passed
Recurrentvs #60Done2 of 2 runs0.868×Generated 0.823×, real models 0.915×QualifiesGain lower bound 2.9%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 92.004× · real models 77.570×

MachineTime
  1. prod-amd64Seed 312506236187.229×Done1 attempt0.877× of the baseline kernels' time
  2. prod-arm64Seed 312506236181.817×Done1 attempt0.942× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.140×
  • Batch 1, sparse1.404×
  • Batch 1, wide1.039×

Mixer

  • Batch 1, normal703.278×
  • Batch 1, sparse663.023×
  • Batch 1, wide709.654×

ResNet

  • Batch 1, normal5941.466×
  • Batch 1, sparse6458.426×
  • Batch 1, wide5894.715×

MobileNet

  • Batch 1, normal2531.690×
  • Batch 1, sparse2415.710×
  • Batch 1, wide2587.860×

Recurrent

  • Batch 1, normal0.857×
  • Batch 1, sparse0.849×
  • Batch 1, wide0.846×

Normalized MLP

  • Batch 1, normal67.142×
  • Batch 1, sparse62.093×
  • Batch 1, wide65.635×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction855.846×
  • compact-mixer on digits560.094×
  • compact-mixer on fashion601.544×
  • compact-mixer on mnist643.771×
  • compact-mlp on digits0.774×
  • compact-mlp on fashion1.046×
  • compact-mlp on mnist0.767×
  • compact-mobilenet on digits2035.378×
  • compact-mobilenet on fashion2545.341×
  • compact-mobilenet on mnist2531.013×
  • compact-normalized_mlp on digits67.308×
  • compact-normalized_mlp on fashion31.000×
  • compact-normalized_mlp on mnist26.785×
  • compact-resnet on digits5725.168×
  • compact-resnet on fashion7694.942×
  • compact-resnet on mnist7608.180×
  • appliance-gru on appliance-energy0.848×
  • mitbih-gru on mitbih0.954×
  • mnist-rows-lstm on mnist-rows0.858×
  • speech-commands-gru on speech-commands0.925×
  • speech-commands-lstm-gru on speech-commands0.894×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.881×
  • tiny-shakespeare-lstm on tiny-shakespeare1.063×
  • uci-har-gru on uci-har0.907×
  • uci-har-lstm on uci-har0.886×
  • uci-har-mixer on UCI-HAR586.184×
  • uci-har-mlp on UCI-HAR1.246×
  • uci-har-mobilenet on UCI-HAR1955.897×
  • uci-har-normalized_mlp on UCI-HAR21.211×
  • uci-har-resnet on UCI-HAR2447.254×
MLP2 runs

RunsGenerated 1.077× · real models 0.978×

MachineTime
  1. prod-amd64Seed 31250623611.098×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 31250623610.959×Done1 attempt0.025× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.929×
  • Batch 1, sparse1.052×
  • Batch 1, wide1.001×

Real models

  • compact-mlp on digits0.933×
  • compact-mlp on fashion0.953×
  • compact-mlp on mnist0.959×
  • uci-har-mlp on UCI-HAR1.072×
ResNet2 runs

RunsGenerated 6022.796× · real models 5178.784×

MachineTime
  1. prod-amd64Seed 31250623615713.305×Done1 attempt49.736× of the baseline kernels' time
  2. prod-arm64Seed 31250623615459.321×Done1 attempt52.375× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal5703.548×
  • Batch 1, sparse6168.975×
  • Batch 1, wide6141.489×

Real models

  • compact-resnet on digits5707.845×
  • compact-resnet on fashion7498.371×
  • compact-resnet on mnist7416.037×
  • uci-har-resnet on UCI-HAR2266.211×
MobileNet2 runs

RunsGenerated 2734.378× · real models 2343.474×

MachineTime
  1. prod-amd64Seed 31250623612496.005×Done1 attempt32.733× of the baseline kernels' time
  2. prod-arm64Seed 31250623612567.279×Done1 attempt33.351× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal2620.481×
  • Batch 1, sparse2511.550×
  • Batch 1, wide2490.839×

Real models

  • compact-mobilenet on digits1883.067×
  • compact-mobilenet on fashion2622.598×
  • compact-mobilenet on mnist2635.739×
  • uci-har-mobilenet on UCI-HAR2317.078×
Normalized MLP2 runs

RunsGenerated 78.637× · real models 39.469×

MachineTime
  1. prod-amd64Seed 312506236164.168×Done1 attempt0.538× of the baseline kernels' time
  2. prod-arm64Seed 312506236148.369×Done1 attempt0.544× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal68.574×
  • Batch 1, sparse87.650×
  • Batch 1, wide84.186×

Real models

  • compact-normalized_mlp on digits69.991×
  • compact-normalized_mlp on fashion40.292×
  • compact-normalized_mlp on mnist41.228×
  • uci-har-normalized_mlp on UCI-HAR20.873×
Mixer2 runs

RunsGenerated 665.055× · real models 670.853×

MachineTime
  1. prod-amd64Seed 3125062361690.540×Done1 attempt3.130× of the baseline kernels' time
  2. prod-arm64Seed 3125062361646.094×Done1 attempt3.205× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal724.336×
  • Batch 1, sparse714.652×
  • Batch 1, wide792.358×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1018.958×
  • compact-mixer on digits528.126×
  • compact-mixer on fashion659.383×
  • compact-mixer on mnist627.886×
  • uci-har-mixer on UCI-HAR609.853×
Recurrent2 runs

RunsGenerated 0.823× · real models 0.915×

MachineTime
  1. prod-amd64Seed 31250623610.843×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 31250623610.894×Done1 attempt0.009× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.820×
  • Batch 1, sparse0.830×
  • Batch 1, wide0.796×

Real models

  • appliance-gru on appliance-energy0.890×
  • mitbih-gru on mitbih0.949×
  • mnist-rows-lstm on mnist-rows0.897×
  • speech-commands-gru on speech-commands0.893×
  • speech-commands-lstm-gru on speech-commands0.885×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.879×
  • tiny-shakespeare-lstm on tiny-shakespeare1.011×
  • uci-har-gru on uci-har0.973×
  • uci-har-lstm on uci-har0.870×