Conjectures.io
Tensor kernels

Entry #20

Completed
Hotkey
5CG78Z…rHFbrv
Sent
11 Nov 2026, 00:26 UTC
Source
0aff9511c31b

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 #26Done2 of 2 runs0.932×Generated 1.088×, real models 0.799×Not significantGain lower bound -8.1%Guard passed
MLPvs #1Done2 of 2 runs0.957×Generated 0.966×, real models 0.947×Not significantGain lower bound -21.2%Guard passed
ResNetvs #20Done2 of 2 runs0.683×Generated 1.044×, real models 0.446×QualifiesGain lower bound 3.4%Guard passed
MobileNetvs #21Done2 of 2 runs0.738×Generated 1.205×, real models 0.452×QualifiesGain lower bound 4.0%Guard passed
Normalized MLPvs #6Done2 of 2 runs1.025×Generated 0.877×, real models 1.197×Above the barGain lower bound -28.4%Guard passed
Mixervs #24Done2 of 2 runs1.077×Generated 1.058×, real models 1.095×Above the barGain lower bound -17.6%Guard passed
Recurrentvs #25Done2 of 2 runs1.353×Generated 1.589×, real models 1.152×Above the barGain lower bound -42.5%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 1.088× · real models 0.799×

MachineTime
  1. prod-amd64Seed 7490612310.951×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 7490612310.914×Done1 attempt0.017× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.986×
  • Batch 1, sparse1.041×
  • Batch 1, wide1.121×

Mixer

  • Batch 1, normal1.031×
  • Batch 1, sparse0.987×
  • Batch 1, wide1.010×

ResNet

  • Batch 1, normal0.963×
  • Batch 1, sparse1.029×
  • Batch 1, wide0.954×

MobileNet

  • Batch 1, normal0.931×
  • Batch 1, sparse0.972×
  • Batch 1, wide0.944×

Recurrent

  • Batch 1, normal1.607×
  • Batch 1, sparse1.612×
  • Batch 1, wide1.599×

Normalized MLP

  • Batch 1, normal1.015×
  • Batch 1, sparse1.180×
  • Batch 1, wide1.057×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.289×
  • compact-mixer on digits4.521×
  • compact-mixer on fashion1.184×
  • compact-mixer on mnist0.918×
  • compact-mlp on digits1.158×
  • compact-mlp on fashion0.471×
  • compact-mlp on mnist0.978×
  • compact-mobilenet on digits0.798×
  • compact-mobilenet on fashion0.740×
  • compact-mobilenet on mnist0.709×
  • compact-normalized_mlp on digits1.056×
  • compact-normalized_mlp on fashion0.891×
  • compact-normalized_mlp on mnist0.722×
  • compact-resnet on digits0.986×
  • compact-resnet on fashion0.839×
  • compact-resnet on mnist0.745×
  • appliance-gru on appliance-energy1.198×
  • mitbih-gru on mitbih0.985×
  • mnist-rows-lstm on mnist-rows1.441×
  • speech-commands-gru on speech-commands1.363×
  • speech-commands-lstm-gru on speech-commands1.360×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.231×
  • tiny-shakespeare-lstm on tiny-shakespeare1.352×
  • uci-har-gru on uci-har1.156×
  • uci-har-lstm on uci-har0.551×
  • uci-har-mixer on UCI-HAR1.030×
  • uci-har-mlp on UCI-HAR1.023×
  • uci-har-mobilenet on UCI-HAR0.102×
  • uci-har-normalized_mlp on UCI-HAR0.972×
  • uci-har-resnet on UCI-HAR0.064×
MLP2 runs

RunsGenerated 0.966× · real models 0.947×

MachineTime
  1. prod-amd64Seed 7490612310.889×Done1 attempt0.015× of the baseline kernels' time
  2. prod-arm64Seed 7490612311.029×Done1 attempt0.029× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.922×
  • Batch 1, sparse1.075×
  • Batch 1, wide0.937×

Real models

  • compact-mlp on digits1.078×
  • compact-mlp on fashion0.889×
  • compact-mlp on mnist0.968×
  • uci-har-mlp on UCI-HAR0.868×
ResNet2 runs

RunsGenerated 1.044× · real models 0.446×

MachineTime
  1. prod-amd64Seed 7490612310.686×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 7490612310.679×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal1.008×
  • Batch 1, sparse1.061×
  • Batch 1, wide1.041×

Real models

  • compact-resnet on digits0.975×
  • compact-resnet on fashion0.836×
  • compact-resnet on mnist0.741×
  • uci-har-resnet on UCI-HAR0.066×
MobileNet2 runs

RunsGenerated 1.205× · real models 0.452×

MachineTime
  1. prod-amd64Seed 7490612310.660×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 7490612310.825×Done1 attempt0.014× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.440×
  • Batch 1, sparse1.007×
  • Batch 1, wide0.961×

Real models

  • compact-mobilenet on digits0.823×
  • compact-mobilenet on fashion0.727×
  • compact-mobilenet on mnist0.700×
  • uci-har-mobilenet on UCI-HAR0.099×
Normalized MLP2 runs

RunsGenerated 0.877× · real models 1.197×

MachineTime
  1. prod-amd64Seed 7490612311.262×Done1 attempt0.007× of the baseline kernels' time
  2. prod-arm64Seed 7490612310.832×Done1 attempt0.014× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal0.936×
  • Batch 1, sparse0.859×
  • Batch 1, wide0.979×

Real models

  • compact-normalized_mlp on digits2.272×
  • compact-normalized_mlp on fashion1.004×
  • compact-normalized_mlp on mnist0.918×
  • uci-har-normalized_mlp on UCI-HAR0.981×
Mixer2 runs

RunsGenerated 1.058× · real models 1.095×

MachineTime
  1. prod-amd64Seed 7490612311.158×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 7490612311.001×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.991×
  • Batch 1, sparse1.009×
  • Batch 1, wide1.050×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.229×
  • compact-mixer on digits1.298×
  • compact-mixer on fashion0.988×
  • compact-mixer on mnist0.938×
  • uci-har-mixer on UCI-HAR1.066×
Recurrent2 runs

RunsGenerated 1.589× · real models 1.152×

MachineTime
  1. prod-amd64Seed 7490612311.360×Done1 attempt0.064× of the baseline kernels' time
  2. prod-arm64Seed 7490612311.346×Done1 attempt0.069× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal1.594×
  • Batch 1, sparse1.617×
  • Batch 1, wide1.600×

Real models

  • appliance-gru on appliance-energy1.256×
  • mitbih-gru on mitbih0.993×
  • mnist-rows-lstm on mnist-rows1.447×
  • speech-commands-gru on speech-commands1.353×
  • speech-commands-lstm-gru on speech-commands1.382×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.217×
  • tiny-shakespeare-lstm on tiny-shakespeare1.359×
  • uci-har-gru on uci-har1.145×
  • uci-har-lstm on uci-har0.558×