Conjectures.io
Tensor kernels

Entry #2

Completed
Hotkey
5CwnNj…teZPSM
Sent
10 Nov 2026, 17:52 UTC
Source
d4c2fadf8fe7

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 #5Done2 of 2 runs0.370×Generated 0.325×, real models 0.420×QualifiesGain lower bound 60.4%Guard passed
MLPvs #1Done2 of 2 runs0.988×Generated 0.991×, real models 0.986×Above the barGain lower bound -13.7%Guard passed
ResNetvs #2Done2 of 2 runs1.015×Generated 0.994×, real models 1.037×Above the barGain lower bound -6.0%Guard passed
MobileNetvs the shipped referenceDone2 of 2 runs1.205×Generated 1.191×, real models 1.219×Above the barGain lower bound -21.0%Guard passed
Normalized MLPvs #3Done2 of 2 runs0.021×Generated 0.016×, real models 0.027×QualifiesGain lower bound 97.4%Guard passed
Mixervs #4Done2 of 2 runs0.009×Generated 0.007×, real models 0.013×QualifiesGain lower bound 98.8%Guard passed
Recurrentvs the shipped referenceDone2 of 2 runs8.534×Generated 8.149×, real models 8.936×Above the barGain lower bound -783.6%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 0.325× · real models 0.420×

MachineTime
  1. prod-amd64Seed 26674945050.371×Done1 attempt0.142× of the baseline kernels' time
  2. prod-arm64Seed 26674945050.368×Done1 attempt0.148× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.046×
  • Batch 1, sparse1.028×
  • Batch 1, wide1.044×

Mixer

  • Batch 1, normal0.007×
  • Batch 1, sparse0.007×
  • Batch 1, wide0.006×

ResNet

  • Batch 1, normal0.996×
  • Batch 1, sparse0.994×
  • Batch 1, wide0.996×

MobileNet

  • Batch 1, normal1.182×
  • Batch 1, sparse1.179×
  • Batch 1, wide1.181×

Recurrent

  • Batch 1, normal8.040×
  • Batch 1, sparse8.126×
  • Batch 1, wide7.803×

Normalized MLP

  • Batch 1, normal0.019×
  • Batch 1, sparse0.021×
  • Batch 1, wide0.020×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.006×
  • compact-mixer on digits0.016×
  • compact-mixer on fashion0.024×
  • compact-mixer on mnist0.034×
  • compact-mlp on digits0.995×
  • compact-mlp on fashion1.012×
  • compact-mlp on mnist1.279×
  • compact-mobilenet on digits1.195×
  • compact-mobilenet on fashion1.220×
  • compact-mobilenet on mnist1.220×
  • compact-normalized_mlp on digits0.013×
  • compact-normalized_mlp on fashion0.040×
  • compact-normalized_mlp on mnist0.042×
  • compact-resnet on digits0.995×
  • compact-resnet on fashion0.995×
  • compact-resnet on mnist0.993×
  • appliance-gru on appliance-energy8.449×
  • mitbih-gru on mitbih9.924×
  • mnist-rows-lstm on mnist-rows9.619×
  • speech-commands-gru on speech-commands6.837×
  • speech-commands-lstm-gru on speech-commands7.947×
  • spoken-arabic-digit-lstm on spoken-arabic-digit10.814×
  • tiny-shakespeare-lstm on tiny-shakespeare8.023×
  • uci-har-gru on uci-har8.905×
  • uci-har-lstm on uci-har10.838×
  • uci-har-mixer on UCI-HAR0.008×
  • uci-har-mlp on UCI-HAR0.976×
  • uci-har-mobilenet on UCI-HAR1.242×
  • uci-har-normalized_mlp on UCI-HAR0.050×
  • uci-har-resnet on UCI-HAR1.233×
MLP2 runs

RunsGenerated 0.991× · real models 0.986×

MachineTime
  1. prod-amd64Seed 26674945050.992×Done1 attempt0.015× of the baseline kernels' time
  2. prod-arm64Seed 26674945050.985×Done1 attempt0.024× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.058×
  • Batch 1, sparse0.926×
  • Batch 1, wide1.148×

Real models

  • compact-mlp on digits0.979×
  • compact-mlp on fashion0.891×
  • compact-mlp on mnist1.064×
  • uci-har-mlp on UCI-HAR1.018×
ResNet2 runs

RunsGenerated 0.994× · real models 1.037×

MachineTime
  1. prod-amd64Seed 26674945051.010×Done1 attempt0.957× of the baseline kernels' time
  2. prod-arm64Seed 26674945051.021×Done1 attempt0.973× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal0.995×
  • Batch 1, sparse0.992×
  • Batch 1, wide0.995×

Real models

  • compact-resnet on digits0.997×
  • compact-resnet on fashion0.941×
  • compact-resnet on mnist0.996×
  • uci-har-resnet on UCI-HAR1.237×
MobileNet2 runs

RunsGenerated 1.191× · real models 1.219×

MachineTime
  1. prod-amd64Seed 26674945051.225×Done1 attempt1.200× of the baseline kernels' time
  2. prod-arm64Seed 26674945051.186×Done1 attempt1.172× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.185×
  • Batch 1, sparse1.182×
  • Batch 1, wide1.182×

Real models

  • compact-mobilenet on digits1.202×
  • compact-mobilenet on fashion1.226×
  • compact-mobilenet on mnist1.221×
  • uci-har-mobilenet on UCI-HAR1.228×
Normalized MLP2 runs

RunsGenerated 0.016× · real models 0.027×

MachineTime
  1. prod-amd64Seed 26674945050.017×Done1 attempt0.006× of the baseline kernels' time
  2. prod-arm64Seed 26674945050.024×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal0.015×
  • Batch 1, sparse0.014×
  • Batch 1, wide0.014×

Real models

  • compact-normalized_mlp on digits0.013×
  • compact-normalized_mlp on fashion0.025×
  • compact-normalized_mlp on mnist0.029×
  • uci-har-normalized_mlp on UCI-HAR0.058×
Mixer2 runs

RunsGenerated 0.007× · real models 0.013×

MachineTime
  1. prod-amd64Seed 26674945050.010×Done1 attempt0.006× of the baseline kernels' time
  2. prod-arm64Seed 26674945050.009×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.007×
  • Batch 1, sparse0.007×
  • Batch 1, wide0.006×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.005×
  • compact-mixer on digits0.020×
  • compact-mixer on fashion0.012×
  • compact-mixer on mnist0.033×
  • uci-har-mixer on UCI-HAR0.008×
Recurrent2 runs

RunsGenerated 8.149× · real models 8.936×

MachineTime
  1. prod-amd64Seed 26674945058.433×Done1 attempt7.919× of the baseline kernels' time
  2. prod-arm64Seed 26674945058.636×Done1 attempt8.054× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal8.003×
  • Batch 1, sparse8.125×
  • Batch 1, wide7.777×

Real models

  • appliance-gru on appliance-energy8.498×
  • mitbih-gru on mitbih9.928×
  • mnist-rows-lstm on mnist-rows9.584×
  • speech-commands-gru on speech-commands6.917×
  • speech-commands-lstm-gru on speech-commands7.908×
  • spoken-arabic-digit-lstm on spoken-arabic-digit10.689×
  • tiny-shakespeare-lstm on tiny-shakespeare7.990×
  • uci-har-gru on uci-har8.921×
  • uci-har-lstm on uci-har10.780×