Conjectures.io
Tensor kernels

Entry #13

Completed
Hotkey
5D4tGP…Jdt8VZ
Sent
10 Nov 2026, 21:48 UTC
Source
fb655645f709

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 #23Done2 of 2 runs0.713×Generated 0.762×, real models 0.668×QualifiesGain lower bound 23.7%Guard passed
MLPvs #1Done2 of 2 runs0.896×Generated 0.938×, real models 0.857×Not significantGain lower bound -3.2%Guard passed
ResNetvs #20Done2 of 2 runs1.008×Generated 1.008×, real models 1.009×Above the barGain lower bound -3.0%Guard passed
MobileNetvs #21Done2 of 2 runs0.989×Generated 0.988×, real models 0.991×Above the barGain lower bound -1.8%Guard passed
Normalized MLPvs #6Done2 of 2 runs0.956×Generated 0.966×, real models 0.945×Not significantGain lower bound -8.1%Guard passed
Mixervs #7Done2 of 2 runs0.775×Generated 1.072×, real models 0.560×QualifiesGain lower bound 6.0%Guard passed
Recurrentvs #22Done2 of 2 runs0.179×Generated 0.185×, real models 0.173×QualifiesGain lower bound 80.9%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 0.762× · real models 0.668×

MachineTime
  1. prod-amd64Seed 7925628850.702×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 7925628850.725×Done1 attempt0.013× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.992×
  • Batch 1, sparse1.007×
  • Batch 1, wide1.038×

Mixer

  • Batch 1, normal1.059×
  • Batch 1, sparse1.045×
  • Batch 1, wide1.007×

ResNet

  • Batch 1, normal0.986×
  • Batch 1, sparse1.034×
  • Batch 1, wide1.015×

MobileNet

  • Batch 1, normal1.022×
  • Batch 1, sparse0.994×
  • Batch 1, wide1.010×

Recurrent

  • Batch 1, normal0.182×
  • Batch 1, sparse0.184×
  • Batch 1, wide0.188×

Normalized MLP

  • Batch 1, normal1.020×
  • Batch 1, sparse1.028×
  • Batch 1, wide0.945×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.043×
  • compact-mixer on digits0.686×
  • compact-mixer on fashion0.687×
  • compact-mixer on mnist0.185×
  • compact-mlp on digits0.504×
  • compact-mlp on fashion1.120×
  • compact-mlp on mnist1.187×
  • compact-mobilenet on digits1.029×
  • compact-mobilenet on fashion0.987×
  • compact-mobilenet on mnist1.019×
  • compact-normalized_mlp on digits0.719×
  • compact-normalized_mlp on fashion0.980×
  • compact-normalized_mlp on mnist0.966×
  • compact-resnet on digits1.115×
  • compact-resnet on fashion1.015×
  • compact-resnet on mnist0.984×
  • appliance-gru on appliance-energy0.198×
  • mitbih-gru on mitbih0.172×
  • mnist-rows-lstm on mnist-rows0.129×
  • speech-commands-gru on speech-commands0.301×
  • speech-commands-lstm-gru on speech-commands0.177×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.107×
  • tiny-shakespeare-lstm on tiny-shakespeare0.124×
  • uci-har-gru on uci-har0.201×
  • uci-har-lstm on uci-har0.224×
  • uci-har-mixer on UCI-HAR0.984×
  • uci-har-mlp on UCI-HAR0.871×
  • uci-har-mobilenet on UCI-HAR0.991×
  • uci-har-normalized_mlp on UCI-HAR1.052×
  • uci-har-resnet on UCI-HAR0.983×
MLP2 runs

RunsGenerated 0.938× · real models 0.857×

MachineTime
  1. prod-amd64Seed 7925628850.833×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 7925628850.965×Done1 attempt0.016× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.908×
  • Batch 1, sparse0.935×
  • Batch 1, wide0.918×

Real models

  • compact-mlp on digits0.914×
  • compact-mlp on fashion0.737×
  • compact-mlp on mnist0.889×
  • uci-har-mlp on UCI-HAR0.899×
ResNet2 runs

RunsGenerated 1.008× · real models 1.009×

MachineTime
  1. prod-amd64Seed 7925628851.012×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 7925628851.005×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal1.020×
  • Batch 1, sparse1.008×
  • Batch 1, wide0.995×

Real models

  • compact-resnet on digits1.068×
  • compact-resnet on fashion0.993×
  • compact-resnet on mnist0.994×
  • uci-har-resnet on UCI-HAR0.984×
MobileNet2 runs

RunsGenerated 0.988× · real models 0.991×

MachineTime
  1. prod-amd64Seed 7925628850.987×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 7925628850.992×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal0.988×
  • Batch 1, sparse1.002×
  • Batch 1, wide1.013×

Real models

  • compact-mobilenet on digits1.002×
  • compact-mobilenet on fashion0.984×
  • compact-mobilenet on mnist0.980×
  • uci-har-mobilenet on UCI-HAR0.997×
Normalized MLP2 runs

RunsGenerated 0.966× · real models 0.945×

MachineTime
  1. prod-amd64Seed 7925628850.897×Done1 attempt0.010× of the baseline kernels' time
  2. prod-arm64Seed 7925628851.018×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal0.899×
  • Batch 1, sparse0.981×
  • Batch 1, wide0.965×

Real models

  • compact-normalized_mlp on digits0.911×
  • compact-normalized_mlp on fashion1.000×
  • compact-normalized_mlp on mnist0.937×
  • uci-har-normalized_mlp on UCI-HAR0.936×
Mixer2 runs

RunsGenerated 1.072× · real models 0.560×

MachineTime
  1. prod-amd64Seed 7925628850.783×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 7925628850.767×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal1.012×
  • Batch 1, sparse1.070×
  • Batch 1, wide1.142×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.933×
  • compact-mixer on digits0.360×
  • compact-mixer on fashion0.871×
  • compact-mixer on mnist0.189×
  • uci-har-mixer on UCI-HAR0.994×
Recurrent2 runs

RunsGenerated 0.185× · real models 0.173×

MachineTime
  1. prod-amd64Seed 7925628850.174×Done1 attempt0.041× of the baseline kernels' time
  2. prod-arm64Seed 7925628850.184×Done1 attempt0.043× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.186×
  • Batch 1, sparse0.186×
  • Batch 1, wide0.189×

Real models

  • appliance-gru on appliance-energy0.198×
  • mitbih-gru on mitbih0.173×
  • mnist-rows-lstm on mnist-rows0.128×
  • speech-commands-gru on speech-commands0.302×
  • speech-commands-lstm-gru on speech-commands0.179×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.109×
  • tiny-shakespeare-lstm on tiny-shakespeare0.123×
  • uci-har-gru on uci-har0.200×
  • uci-har-lstm on uci-har0.224×