Conjectures.io
Tensor kernels

Entry #115

Completed
Hotkey
5G6hSu…3m8ASL
Sent
12 Nov 2026, 16:12 UTC
Source
bbf3feca43e4

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 runs0.928×Generated 0.943×, real models 0.913×QualifiesGain lower bound 0.9%Guard passed
MLPvs #44Done2 of 2 runs1.006×Generated 1.005×, real models 1.006×Above the barGain lower bound -26.3%Guard passed
ResNetvs #59Done2 of 2 runs0.994×Generated 0.972×, real models 1.016×Above the barGain lower bound -12.9%Guard passed
MobileNetvs #53Done2 of 2 runs1.117×Generated 1.227×, real models 1.017×Above the barGain lower bound -27.9%Guard passed
Normalized MLPvs #48Done2 of 2 runs1.016×Generated 1.000×, real models 1.032×Above the barGain lower bound -13.4%Guard passed
Mixervs #58Done2 of 2 runs1.003×Generated 0.988×, real models 1.019×Above the barGain lower bound -26.3%Guard passed
Recurrentvs #60Done2 of 2 runs0.988×Generated 0.986×, real models 0.989×Above the barGain lower bound -2.3%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 0.943× · real models 0.913×

MachineTime
  1. prod-amd64Seed 35135903780.900×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 35135903780.957×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.095×
  • Batch 1, sparse1.058×
  • Batch 1, wide0.926×

Mixer

  • Batch 1, normal0.920×
  • Batch 1, sparse0.932×
  • Batch 1, wide0.910×

ResNet

  • Batch 1, normal0.943×
  • Batch 1, sparse0.992×
  • Batch 1, wide0.975×

MobileNet

  • Batch 1, normal0.986×
  • Batch 1, sparse1.032×
  • Batch 1, wide0.991×

Recurrent

  • Batch 1, normal0.868×
  • Batch 1, sparse0.827×
  • Batch 1, wide0.836×

Normalized MLP

  • Batch 1, normal1.038×
  • Batch 1, sparse0.962×
  • Batch 1, wide0.893×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.997×
  • compact-mixer on digits0.866×
  • compact-mixer on fashion1.003×
  • compact-mixer on mnist0.971×
  • compact-mlp on digits1.008×
  • compact-mlp on fashion1.341×
  • compact-mlp on mnist0.747×
  • compact-mobilenet on digits0.896×
  • compact-mobilenet on fashion0.871×
  • compact-mobilenet on mnist0.870×
  • compact-normalized_mlp on digits1.235×
  • compact-normalized_mlp on fashion0.864×
  • compact-normalized_mlp on mnist0.636×
  • compact-resnet on digits0.784×
  • compact-resnet on fashion0.934×
  • compact-resnet on mnist0.914×
  • appliance-gru on appliance-energy0.911×
  • mitbih-gru on mitbih0.756×
  • mnist-rows-lstm on mnist-rows0.559×
  • speech-commands-gru on speech-commands0.811×
  • speech-commands-lstm-gru on speech-commands0.854×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.878×
  • tiny-shakespeare-lstm on tiny-shakespeare0.598×
  • uci-har-gru on uci-har0.793×
  • uci-har-lstm on uci-har0.869×
  • uci-har-mixer on UCI-HAR1.037×
  • uci-har-mlp on UCI-HAR0.994×
  • uci-har-mobilenet on UCI-HAR1.115×
  • uci-har-normalized_mlp on UCI-HAR0.955×
  • uci-har-resnet on UCI-HAR1.078×
MLP2 runs

RunsGenerated 1.005× · real models 1.006×

MachineTime
  1. prod-amd64Seed 35135903781.009×Done1 attempt0.019× of the baseline kernels' time
  2. prod-arm64Seed 35135903781.002×Done1 attempt0.022× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.051×
  • Batch 1, sparse1.010×
  • Batch 1, wide0.999×

Real models

  • compact-mlp on digits1.043×
  • compact-mlp on fashion1.594×
  • compact-mlp on mnist0.635×
  • uci-har-mlp on UCI-HAR0.971×
ResNet2 runs

RunsGenerated 0.972× · real models 1.016×

MachineTime
  1. prod-amd64Seed 35135903780.991×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 35135903780.996×Done1 attempt0.008× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal0.987×
  • Batch 1, sparse1.000×
  • Batch 1, wide0.982×

Real models

  • compact-resnet on digits1.094×
  • compact-resnet on fashion0.972×
  • compact-resnet on mnist0.981×
  • uci-har-resnet on UCI-HAR1.020×
MobileNet2 runs

RunsGenerated 1.227× · real models 1.017×

MachineTime
  1. prod-amd64Seed 35135903781.024×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 35135903781.218×Done1 attempt0.013× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.448×
  • Batch 1, sparse1.032×
  • Batch 1, wide0.983×

Real models

  • compact-mobilenet on digits0.837×
  • compact-mobilenet on fashion1.011×
  • compact-mobilenet on mnist1.052×
  • uci-har-mobilenet on UCI-HAR1.202×
Normalized MLP2 runs

RunsGenerated 1.000× · real models 1.032×

MachineTime
  1. prod-amd64Seed 35135903780.998×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 35135903781.034×Done1 attempt0.015× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.017×
  • Batch 1, sparse0.962×
  • Batch 1, wide1.054×

Real models

  • compact-normalized_mlp on digits1.050×
  • compact-normalized_mlp on fashion1.041×
  • compact-normalized_mlp on mnist0.979×
  • uci-har-normalized_mlp on UCI-HAR1.059×
Mixer2 runs

RunsGenerated 0.988× · real models 1.019×

MachineTime
  1. prod-amd64Seed 35135903781.002×Done1 attempt0.004× of the baseline kernels' time
  2. prod-arm64Seed 35135903781.004×Done1 attempt0.004× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.997×
  • Batch 1, sparse0.975×
  • Batch 1, wide0.992×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.974×
  • compact-mixer on digits1.158×
  • compact-mixer on fashion1.000×
  • compact-mixer on mnist0.966×
  • uci-har-mixer on UCI-HAR1.006×
Recurrent2 runs

RunsGenerated 0.986× · real models 0.989×

MachineTime
  1. prod-amd64Seed 35135903780.970×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 35135903781.006×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.989×
  • Batch 1, sparse0.977×
  • Batch 1, wide0.966×

Real models

  • appliance-gru on appliance-energy0.973×
  • mitbih-gru on mitbih0.974×
  • mnist-rows-lstm on mnist-rows1.005×
  • speech-commands-gru on speech-commands1.020×
  • speech-commands-lstm-gru on speech-commands0.988×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.016×
  • tiny-shakespeare-lstm on tiny-shakespeare0.961×
  • uci-har-gru on uci-har0.968×
  • uci-har-lstm on uci-har0.996×