Conjectures.io
Tensor kernels

Entry #52

Completed
Hotkey
5Cz1sJ…FSuchu
Sent
11 Nov 2026, 13:17 UTC
Source
45d8b3330e90

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 #43Done2 of 2 runs0.993×Generated 0.978×, real models 1.008×Above the barGain lower bound -6.6%Guard passed
MLPvs #44Done2 of 2 runs1.058×Generated 1.074×, real models 1.042×Above the barGain lower bound -31.0%Guard passed
ResNetvs #45Done2 of 2 runs1.019×Generated 1.018×, real models 1.020×Above the barGain lower bound -17.4%Guard passed
MobileNetvs #42Done2 of 2 runs1.184×Generated 1.363×, real models 1.028×Above the barGain lower bound -33.9%Guard passed
Normalized MLPvs #48Done2 of 2 runs1.048×Generated 1.053×, real models 1.042×Above the barGain lower bound -20.2%Guard passed
Mixervs #47Done2 of 2 runs1.010×Generated 1.034×, real models 0.987×Above the barGain lower bound -8.0%Guard passed
Recurrentvs #46Done2 of 2 runs0.946×Generated 0.925×, real models 0.968×QualifiesGain lower bound 2.7%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 0.978× · real models 1.008×

MachineTime
  1. prod-amd64Seed 35320099461.023×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 35320099460.963×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.989×
  • Batch 1, sparse1.017×
  • Batch 1, wide0.950×

Mixer

  • Batch 1, normal0.891×
  • Batch 1, sparse0.904×
  • Batch 1, wide0.884×

ResNet

  • Batch 1, normal1.013×
  • Batch 1, sparse0.948×
  • Batch 1, wide0.989×

MobileNet

  • Batch 1, normal1.059×
  • Batch 1, sparse1.091×
  • Batch 1, wide1.057×

Recurrent

  • Batch 1, normal0.875×
  • Batch 1, sparse0.909×
  • Batch 1, wide0.921×

Normalized MLP

  • Batch 1, normal1.099×
  • Batch 1, sparse1.098×
  • Batch 1, wide0.996×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.959×
  • compact-mixer on digits1.569×
  • compact-mixer on fashion0.919×
  • compact-mixer on mnist0.928×
  • compact-mlp on digits0.919×
  • compact-mlp on fashion1.412×
  • compact-mlp on mnist1.483×
  • compact-mobilenet on digits1.013×
  • compact-mobilenet on fashion0.885×
  • compact-mobilenet on mnist0.926×
  • compact-normalized_mlp on digits1.024×
  • compact-normalized_mlp on fashion0.948×
  • compact-normalized_mlp on mnist0.772×
  • compact-resnet on digits0.990×
  • compact-resnet on fashion0.949×
  • compact-resnet on mnist0.975×
  • appliance-gru on appliance-energy0.942×
  • mitbih-gru on mitbih0.898×
  • mnist-rows-lstm on mnist-rows1.031×
  • speech-commands-gru on speech-commands0.934×
  • speech-commands-lstm-gru on speech-commands0.903×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.849×
  • tiny-shakespeare-lstm on tiny-shakespeare0.940×
  • uci-har-gru on uci-har0.891×
  • uci-har-lstm on uci-har0.847×
  • uci-har-mixer on UCI-HAR0.997×
  • uci-har-mlp on UCI-HAR1.025×
  • uci-har-mobilenet on UCI-HAR1.159×
  • uci-har-normalized_mlp on UCI-HAR0.981×
  • uci-har-resnet on UCI-HAR1.103×
MLP2 runs

RunsGenerated 1.074× · real models 1.042×

MachineTime
  1. prod-amd64Seed 35320099461.141×Done1 attempt0.016× of the baseline kernels' time
  2. prod-arm64Seed 35320099460.981×Done1 attempt0.018× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.081×
  • Batch 1, sparse1.148×
  • Batch 1, wide1.142×

Real models

  • compact-mlp on digits1.020×
  • compact-mlp on fashion0.991×
  • compact-mlp on mnist1.154×
  • uci-har-mlp on UCI-HAR1.011×
ResNet2 runs

RunsGenerated 1.018× · real models 1.020×

MachineTime
  1. prod-amd64Seed 35320099461.032×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 35320099461.007×Done1 attempt0.009× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal1.043×
  • Batch 1, sparse1.041×
  • Batch 1, wide0.987×

Real models

  • compact-resnet on digits1.041×
  • compact-resnet on fashion0.965×
  • compact-resnet on mnist0.983×
  • uci-har-resnet on UCI-HAR1.096×
MobileNet2 runs

RunsGenerated 1.363× · real models 1.028×

MachineTime
  1. prod-amd64Seed 35320099461.108×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 35320099461.265×Done1 attempt0.014× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.618×
  • Batch 1, sparse1.141×
  • Batch 1, wide1.082×

Real models

  • compact-mobilenet on digits1.154×
  • compact-mobilenet on fashion0.893×
  • compact-mobilenet on mnist0.924×
  • uci-har-mobilenet on UCI-HAR1.172×
Normalized MLP2 runs

RunsGenerated 1.053× · real models 1.042×

MachineTime
  1. prod-amd64Seed 35320099461.094×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 35320099461.004×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.103×
  • Batch 1, sparse1.107×
  • Batch 1, wide1.135×

Real models

  • compact-normalized_mlp on digits1.042×
  • compact-normalized_mlp on fashion0.971×
  • compact-normalized_mlp on mnist1.019×
  • uci-har-normalized_mlp on UCI-HAR1.144×
Mixer2 runs

RunsGenerated 1.034× · real models 0.987×

MachineTime
  1. prod-amd64Seed 35320099460.980×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 35320099461.040×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal1.023×
  • Batch 1, sparse1.079×
  • Batch 1, wide1.011×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.101×
  • compact-mixer on digits0.860×
  • compact-mixer on fashion0.948×
  • compact-mixer on mnist1.024×
  • uci-har-mixer on UCI-HAR1.018×
Recurrent2 runs

RunsGenerated 0.925× · real models 0.968×

MachineTime
  1. prod-amd64Seed 35320099460.973×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 35320099460.920×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.920×
  • Batch 1, sparse0.936×
  • Batch 1, wide0.905×

Real models

  • appliance-gru on appliance-energy0.993×
  • mitbih-gru on mitbih0.957×
  • mnist-rows-lstm on mnist-rows1.097×
  • speech-commands-gru on speech-commands1.057×
  • speech-commands-lstm-gru on speech-commands0.969×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.902×
  • tiny-shakespeare-lstm on tiny-shakespeare0.978×
  • uci-har-gru on uci-har0.945×
  • uci-har-lstm on uci-har0.835×