Conjectures.io
Tensor kernels

Entry #35

Completed
Hotkey
5CyHJK…zFgsHQ
Sent
11 Nov 2026, 07:08 UTC
Source
e440fffce4a9

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 #39Done2 of 2 runs0.925×Generated 0.926×, real models 0.925×QualifiesGain lower bound 1.7%Guard passed
MLPvs #1Done2 of 2 runs1.099×Generated 0.981×, real models 1.230×Above the barGain lower bound -42.0%Guard passed
ResNetvs #40Done2 of 2 runs0.967×Generated 1.009×, real models 0.928×Not significantGain lower bound -7.9%Guard passed
MobileNetvs #37Done2 of 2 runs0.824×Generated 0.756×, real models 0.898×QualifiesGain lower bound 4.8%Guard passed
Normalized MLPvs #6Done2 of 2 runs0.927×Generated 0.915×, real models 0.938×Not significantGain lower bound -4.6%Guard passed
Mixervs #24Done2 of 2 runs1.067×Generated 1.122×, real models 1.014×Above the barGain lower bound -28.0%Guard passed
Recurrentvs #41Done2 of 2 runs0.991×Generated 1.001×, real models 0.982×Above the barGain lower bound -9.4%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 0.926× · real models 0.925×

MachineTime
  1. prod-amd64Seed 28054140990.919×Done1 attempt0.010× of the baseline kernels' time
  2. prod-arm64Seed 28054140990.932×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.988×
  • Batch 1, sparse0.766×
  • Batch 1, wide0.954×

Mixer

  • Batch 1, normal0.998×
  • Batch 1, sparse1.029×
  • Batch 1, wide1.133×

ResNet

  • Batch 1, normal0.890×
  • Batch 1, sparse0.898×
  • Batch 1, wide0.917×

MobileNet

  • Batch 1, normal0.908×
  • Batch 1, sparse0.861×
  • Batch 1, wide0.944×

Recurrent

  • Batch 1, normal0.825×
  • Batch 1, sparse0.795×
  • Batch 1, wide0.806×

Normalized MLP

  • Batch 1, normal0.956×
  • Batch 1, sparse0.954×
  • Batch 1, wide1.215×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.004×
  • compact-mixer on digits0.908×
  • compact-mixer on fashion1.097×
  • compact-mixer on mnist1.208×
  • compact-mlp on digits0.660×
  • compact-mlp on fashion0.833×
  • compact-mlp on mnist0.984×
  • compact-mobilenet on digits0.800×
  • compact-mobilenet on fashion1.059×
  • compact-mobilenet on mnist1.014×
  • compact-normalized_mlp on digits1.224×
  • compact-normalized_mlp on fashion1.252×
  • compact-normalized_mlp on mnist1.087×
  • compact-resnet on digits0.854×
  • compact-resnet on fashion0.949×
  • compact-resnet on mnist0.942×
  • appliance-gru on appliance-energy0.726×
  • mitbih-gru on mitbih0.747×
  • mnist-rows-lstm on mnist-rows0.831×
  • speech-commands-gru on speech-commands0.701×
  • speech-commands-lstm-gru on speech-commands0.788×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.878×
  • tiny-shakespeare-lstm on tiny-shakespeare0.856×
  • uci-har-gru on uci-har0.744×
  • uci-har-lstm on uci-har0.850×
  • uci-har-mixer on UCI-HAR0.956×
  • uci-har-mlp on UCI-HAR1.164×
  • uci-har-mobilenet on UCI-HAR0.729×
  • uci-har-normalized_mlp on UCI-HAR0.952×
  • uci-har-resnet on UCI-HAR0.735×
MLP2 runs

RunsGenerated 0.981× · real models 1.230×

MachineTime
  1. prod-amd64Seed 28054140990.923×Done1 attempt0.015× of the baseline kernels' time
  2. prod-arm64Seed 28054140991.308×Done1 attempt0.017× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.190×
  • Batch 1, sparse0.905×
  • Batch 1, wide0.789×

Real models

  • compact-mlp on digits1.027×
  • compact-mlp on fashion2.655×
  • compact-mlp on mnist0.876×
  • uci-har-mlp on UCI-HAR0.960×
ResNet2 runs

RunsGenerated 1.009× · real models 0.928×

MachineTime
  1. prod-amd64Seed 28054140990.938×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 28054140990.998×Done1 attempt0.008× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal1.002×
  • Batch 1, sparse1.006×
  • Batch 1, wide0.982×

Real models

  • compact-resnet on digits0.745×
  • compact-resnet on fashion1.008×
  • compact-resnet on mnist0.993×
  • uci-har-resnet on UCI-HAR0.992×
MobileNet2 runs

RunsGenerated 0.756× · real models 0.898×

MachineTime
  1. prod-amd64Seed 28054140990.883×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 28054140990.769×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal0.650×
  • Batch 1, sparse0.885×
  • Batch 1, wide0.904×

Real models

  • compact-mobilenet on digits0.872×
  • compact-mobilenet on fashion1.072×
  • compact-mobilenet on mnist1.015×
  • uci-har-mobilenet on UCI-HAR0.684×
Normalized MLP2 runs

RunsGenerated 0.915× · real models 0.938×

MachineTime
  1. prod-amd64Seed 28054140990.891×Done1 attempt0.007× of the baseline kernels' time
  2. prod-arm64Seed 28054140990.964×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal0.924×
  • Batch 1, sparse0.905×
  • Batch 1, wide0.927×

Real models

  • compact-normalized_mlp on digits1.000×
  • compact-normalized_mlp on fashion0.995×
  • compact-normalized_mlp on mnist0.825×
  • uci-har-normalized_mlp on UCI-HAR0.944×
Mixer2 runs

RunsGenerated 1.122× · real models 1.014×

MachineTime
  1. prod-amd64Seed 28054140991.109×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 28054140991.026×Done1 attempt0.006× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal1.092×
  • Batch 1, sparse1.160×
  • Batch 1, wide1.037×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.134×
  • compact-mixer on digits0.924×
  • compact-mixer on fashion0.990×
  • compact-mixer on mnist1.023×
  • uci-har-mixer on UCI-HAR1.012×
Recurrent2 runs

RunsGenerated 1.001× · real models 0.982×

MachineTime
  1. prod-amd64Seed 28054140990.991×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 28054140990.991×Done1 attempt0.014× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal1.015×
  • Batch 1, sparse0.994×
  • Batch 1, wide0.998×

Real models

  • appliance-gru on appliance-energy1.019×
  • mitbih-gru on mitbih0.991×
  • mnist-rows-lstm on mnist-rows0.844×
  • speech-commands-gru on speech-commands1.006×
  • speech-commands-lstm-gru on speech-commands1.002×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.008×
  • tiny-shakespeare-lstm on tiny-shakespeare0.978×
  • uci-har-gru on uci-har0.995×
  • uci-har-lstm on uci-har1.006×