Conjectures.io
Tensor kernels

Entry #8

Completed
Hotkey
5EqEdM…nJPmMs
Sent
10 Nov 2026, 19:58 UTC
Source
7bc6d019f852

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 #11Done2 of 2 runs0.482×Generated 0.561×, real models 0.415×QualifiesGain lower bound 48.8%Guard passed
MLPvs #1Done2 of 2 runs0.985×Generated 0.976×, real models 0.993×Above the barGain lower bound -16.7%Guard passed
ResNetvs #15Done2 of 2 runs2.042×Generated 2.014×, real models 2.071×Above the barGain lower bound -115.1%Guard passed
MobileNetvs #16Done2 of 2 runs2.060×Generated 2.161×, real models 1.964×Above the barGain lower bound -111.4%Guard passed
Normalized MLPvs #6Done2 of 2 runs0.977×Generated 0.930×, real models 1.025×Above the barGain lower bound -11.6%Guard passed
Mixervs #7Done2 of 2 runs1.060×Generated 1.088×, real models 1.033×Above the barGain lower bound -18.5%Guard passed
Recurrentvs #14Done2 of 2 runs1.001×Generated 1.001×, real models 1.001×Above the barGain lower bound -4.5%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 0.561× · real models 0.415×

MachineTime
  1. prod-amd64Seed 38937124940.472×Done1 attempt0.061× of the baseline kernels' time
  2. prod-arm64Seed 38937124940.493×Done1 attempt0.066× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.933×
  • Batch 1, sparse0.864×
  • Batch 1, wide0.900×

Mixer

  • Batch 1, normal0.995×
  • Batch 1, sparse0.971×
  • Batch 1, wide1.055×

ResNet

  • Batch 1, normal0.999×
  • Batch 1, sparse0.993×
  • Batch 1, wide0.998×

MobileNet

  • Batch 1, normal0.831×
  • Batch 1, sparse0.838×
  • Batch 1, wide0.836×

Recurrent

  • Batch 1, normal0.044×
  • Batch 1, sparse0.045×
  • Batch 1, wide0.045×

Normalized MLP

  • Batch 1, normal0.938×
  • Batch 1, sparse1.061×
  • Batch 1, wide1.016×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.988×
  • compact-mixer on digits0.914×
  • compact-mixer on fashion1.021×
  • compact-mixer on mnist1.007×
  • compact-mlp on digits1.057×
  • compact-mlp on fashion1.207×
  • compact-mlp on mnist0.875×
  • compact-mobilenet on digits0.529×
  • compact-mobilenet on fashion0.701×
  • compact-mobilenet on mnist0.218×
  • compact-normalized_mlp on digits0.847×
  • compact-normalized_mlp on fashion0.688×
  • compact-normalized_mlp on mnist0.597×
  • compact-resnet on digits0.462×
  • compact-resnet on fashion0.310×
  • compact-resnet on mnist0.123×
  • appliance-gru on appliance-energy0.045×
  • mitbih-gru on mitbih0.040×
  • mnist-rows-lstm on mnist-rows0.031×
  • speech-commands-gru on speech-commands0.055×
  • speech-commands-lstm-gru on speech-commands0.039×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.033×
  • tiny-shakespeare-lstm on tiny-shakespeare0.030×
  • uci-har-gru on uci-har0.044×
  • uci-har-lstm on uci-har0.034×
  • uci-har-mixer on UCI-HAR1.047×
  • uci-har-mlp on UCI-HAR0.806×
  • uci-har-mobilenet on UCI-HAR0.814×
  • uci-har-normalized_mlp on UCI-HAR0.893×
  • uci-har-resnet on UCI-HAR1.002×
MLP2 runs

RunsGenerated 0.976× · real models 0.993×

MachineTime
  1. prod-amd64Seed 38937124940.977×Done1 attempt0.019× of the baseline kernels' time
  2. prod-arm64Seed 38937124940.993×Done1 attempt0.017× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.953×
  • Batch 1, sparse1.004×
  • Batch 1, wide0.951×

Real models

  • compact-mlp on digits0.980×
  • compact-mlp on fashion0.990×
  • compact-mlp on mnist1.090×
  • uci-har-mlp on UCI-HAR0.922×
ResNet2 runs

RunsGenerated 2.014× · real models 2.071×

MachineTime
  1. prod-amd64Seed 38937124942.023×Done1 attempt0.298× of the baseline kernels' time
  2. prod-arm64Seed 38937124942.062×Done1 attempt0.328× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal2.022×
  • Batch 1, sparse2.021×
  • Batch 1, wide2.017×

Real models

  • compact-resnet on digits1.979×
  • compact-resnet on fashion1.933×
  • compact-resnet on mnist1.741×
  • uci-har-resnet on UCI-HAR2.762×
MobileNet2 runs

RunsGenerated 2.161× · real models 1.964×

MachineTime
  1. prod-amd64Seed 38937124942.085×Done1 attempt0.521× of the baseline kernels' time
  2. prod-arm64Seed 38937124942.036×Done1 attempt0.542× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal2.162×
  • Batch 1, sparse2.163×
  • Batch 1, wide2.157×

Real models

  • compact-mobilenet on digits2.025×
  • compact-mobilenet on fashion1.846×
  • compact-mobilenet on mnist1.782×
  • uci-har-mobilenet on UCI-HAR2.232×
Normalized MLP2 runs

RunsGenerated 0.930× · real models 1.025×

MachineTime
  1. prod-amd64Seed 38937124940.989×Done1 attempt0.007× of the baseline kernels' time
  2. prod-arm64Seed 38937124940.965×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal0.956×
  • Batch 1, sparse1.002×
  • Batch 1, wide0.987×

Real models

  • compact-normalized_mlp on digits1.019×
  • compact-normalized_mlp on fashion1.013×
  • compact-normalized_mlp on mnist1.114×
  • uci-har-normalized_mlp on UCI-HAR0.960×
Mixer2 runs

RunsGenerated 1.088× · real models 1.033×

MachineTime
  1. prod-amd64Seed 38937124941.111×Done1 attempt0.004× of the baseline kernels' time
  2. prod-arm64Seed 38937124941.012×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal1.003×
  • Batch 1, sparse0.971×
  • Batch 1, wide1.249×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.039×
  • compact-mixer on digits1.116×
  • compact-mixer on fashion0.995×
  • compact-mixer on mnist1.005×
  • uci-har-mixer on UCI-HAR1.016×
Recurrent2 runs

RunsGenerated 1.001× · real models 1.001×

MachineTime
  1. prod-amd64Seed 38937124941.001×Done1 attempt0.359× of the baseline kernels' time
  2. prod-arm64Seed 38937124941.001×Done1 attempt0.357× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal1.000×
  • Batch 1, sparse1.000×
  • Batch 1, wide1.004×

Real models

  • appliance-gru on appliance-energy0.995×
  • mitbih-gru on mitbih0.995×
  • mnist-rows-lstm on mnist-rows1.009×
  • speech-commands-gru on speech-commands1.002×
  • speech-commands-lstm-gru on speech-commands1.002×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.006×
  • tiny-shakespeare-lstm on tiny-shakespeare1.002×
  • uci-har-gru on uci-har0.996×
  • uci-har-lstm on uci-har1.001×