Conjectures.io
Tensor kernels

Entry #10

Completed
Hotkey
5He6Yx…YrSi72
Sent
10 Nov 2026, 20:02 UTC
Source
bf49e2de1fa3

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 #17Done2 of 2 runs4.001×Generated 4.547×, real models 3.520×Above the barGain lower bound -340.2%Guard passed
MLPvs #1Done2 of 2 runs0.973×Generated 0.978×, real models 0.968×Above the barGain lower bound -22.1%Guard passed
ResNetvs #15Done2 of 2 runs2.049×Generated 2.017×, real models 2.081×Above the barGain lower bound -115.9%Guard passed
MobileNetvs #16Done2 of 2 runs2.089×Generated 2.177×, real models 2.005×Above the barGain lower bound -114.8%Guard passed
Normalized MLPvs #6Done2 of 2 runs41.062×Generated 54.788×, real models 30.775×Above the barGain lower bound -4997.0%Guard passed
Mixervs #7Done2 of 2 runs121.902×Generated 167.111×, real models 88.924×Above the barGain lower bound -14879.2%Guard passed
Recurrentvs #14Done2 of 2 runs0.838×Generated 0.880×, real models 0.798×QualifiesGain lower bound 14.3%Guard passed

Download Kernel.lean

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 4.547× · real models 3.520×

MachineTime
  1. prod-amd64Seed 28047965334.218×Done1 attempt0.256× of the baseline kernels' time
  2. prod-arm64Seed 28047965333.795×Done1 attempt0.302× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.109×
  • Batch 1, sparse1.018×
  • Batch 1, wide0.784×

Mixer

  • Batch 1, normal165.698×
  • Batch 1, sparse156.171×
  • Batch 1, wide173.658×

ResNet

  • Batch 1, normal1.002×
  • Batch 1, sparse1.005×
  • Batch 1, wide1.003×

MobileNet

  • Batch 1, normal1.004×
  • Batch 1, sparse1.004×
  • Batch 1, wide1.006×

Recurrent

  • Batch 1, normal0.882×
  • Batch 1, sparse0.899×
  • Batch 1, wide0.880×

Normalized MLP

  • Batch 1, normal70.559×
  • Batch 1, sparse71.486×
  • Batch 1, wide68.266×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction205.277×
  • compact-mixer on digits67.413×
  • compact-mixer on fashion24.209×
  • compact-mixer on mnist32.647×
  • compact-mlp on digits0.894×
  • compact-mlp on fashion1.516×
  • compact-mlp on mnist0.997×
  • compact-mobilenet on digits1.016×
  • compact-mobilenet on fashion1.048×
  • compact-mobilenet on mnist1.059×
  • compact-normalized_mlp on digits79.925×
  • compact-normalized_mlp on fashion22.101×
  • compact-normalized_mlp on mnist24.737×
  • compact-resnet on digits1.015×
  • compact-resnet on fashion1.036×
  • compact-resnet on mnist1.039×
  • appliance-gru on appliance-energy0.748×
  • mitbih-gru on mitbih0.686×
  • mnist-rows-lstm on mnist-rows0.886×
  • speech-commands-gru on speech-commands0.920×
  • speech-commands-lstm-gru on speech-commands0.886×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.785×
  • tiny-shakespeare-lstm on tiny-shakespeare0.886×
  • uci-har-gru on uci-har0.746×
  • uci-har-lstm on uci-har0.743×
  • uci-har-mixer on UCI-HAR126.219×
  • uci-har-mlp on UCI-HAR1.064×
  • uci-har-mobilenet on UCI-HAR1.048×
  • uci-har-normalized_mlp on UCI-HAR18.630×
  • uci-har-resnet on UCI-HAR1.005×
MLP2 runs

RunsGenerated 0.978× · real models 0.968×

MachineTime
  1. prod-amd64Seed 28047965330.962×Done1 attempt0.015× of the baseline kernels' time
  2. prod-arm64Seed 28047965330.984×Done1 attempt0.038× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.994×
  • Batch 1, sparse0.993×
  • Batch 1, wide0.995×

Real models

  • compact-mlp on digits0.892×
  • compact-mlp on fashion0.898×
  • compact-mlp on mnist1.087×
  • uci-har-mlp on UCI-HAR1.007×
ResNet2 runs

RunsGenerated 2.017× · real models 2.081×

MachineTime
  1. prod-amd64Seed 28047965332.039×Done1 attempt0.268× of the baseline kernels' time
  2. prod-arm64Seed 28047965332.059×Done1 attempt0.290× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal2.001×
  • Batch 1, sparse2.031×
  • Batch 1, wide2.031×

Real models

  • compact-resnet on digits1.994×
  • compact-resnet on fashion1.867×
  • compact-resnet on mnist1.820×
  • uci-har-resnet on UCI-HAR2.768×
MobileNet2 runs

RunsGenerated 2.177× · real models 2.005×

MachineTime
  1. prod-amd64Seed 28047965332.130×Done1 attempt0.498× of the baseline kernels' time
  2. prod-arm64Seed 28047965332.049×Done1 attempt0.517× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal2.173×
  • Batch 1, sparse2.179×
  • Batch 1, wide2.184×

Real models

  • compact-mobilenet on digits2.039×
  • compact-mobilenet on fashion1.920×
  • compact-mobilenet on mnist1.858×
  • uci-har-mobilenet on UCI-HAR2.223×
Normalized MLP2 runs

RunsGenerated 54.788× · real models 30.775×

MachineTime
  1. prod-amd64Seed 280479653344.024×Done1 attempt0.613× of the baseline kernels' time
  2. prod-arm64Seed 280479653338.300×Done1 attempt0.624× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal55.939×
  • Batch 1, sparse56.859×
  • Batch 1, wide60.408×

Real models

  • compact-normalized_mlp on digits83.832×
  • compact-normalized_mlp on fashion22.664×
  • compact-normalized_mlp on mnist26.984×
  • uci-har-normalized_mlp on UCI-HAR17.496×
Mixer2 runs

RunsGenerated 167.111× · real models 88.924×

MachineTime
  1. prod-amd64Seed 2804796533139.778×Done1 attempt0.792× of the baseline kernels' time
  2. prod-arm64Seed 2804796533106.313×Done1 attempt0.796× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal142.818×
  • Batch 1, sparse161.469×
  • Batch 1, wide183.854×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction255.584×
  • compact-mixer on digits87.053×
  • compact-mixer on fashion69.394×
  • compact-mixer on mnist27.681×
  • uci-har-mixer on UCI-HAR130.096×
Recurrent2 runs

RunsGenerated 0.880× · real models 0.798×

MachineTime
  1. prod-amd64Seed 28047965330.840×Done1 attempt0.309× of the baseline kernels' time
  2. prod-arm64Seed 28047965330.836×Done1 attempt0.307× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.877×
  • Batch 1, sparse0.892×
  • Batch 1, wide0.877×

Real models

  • appliance-gru on appliance-energy0.744×
  • mitbih-gru on mitbih0.676×
  • mnist-rows-lstm on mnist-rows0.883×
  • speech-commands-gru on speech-commands0.916×
  • speech-commands-lstm-gru on speech-commands0.879×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.773×
  • tiny-shakespeare-lstm on tiny-shakespeare0.877×
  • uci-har-gru on uci-har0.744×
  • uci-har-lstm on uci-har0.726×