Conjectures.io
Tensor kernels

Entry #9

Completed
Hotkey
5E9zeD…d9ttNh
Sent
10 Nov 2026, 19:59 UTC
Source
89483e35fdd3

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 #17Done2 of 2 runs0.974×Generated 0.368×, real models 2.580×Above the barGain lower bound -8.0%Guard passed
MLPvs #1Done2 of 2 runs0.962×Generated 1.026×, real models 0.903×Not significantGain lower bound -31.9%Guard passed
ResNetvs #15Done2 of 2 runs0.974×Generated 0.059×, real models 15.988×Above the barGain lower bound -57.2%Guard passed
MobileNetvs #16Done2 of 2 runs0.853×Generated 0.046×, real models 15.739×Not significantGain lower bound -16.2%Guard passed
Normalized MLPvs #6Done2 of 2 runs0.974×Generated 0.975×, real models 0.974×Above the barGain lower bound -5.3%Guard passed
Mixervs #7Done2 of 2 runs1.003×Generated 0.975×, real models 1.031×Above the barGain lower bound -13.2%Guard passed
Recurrentvs #14Done2 of 2 runs3.039×Generated 2.820×, real models 3.275×Above the barGain lower bound -223.3%Guard passed

The source stays private while it holds a board, and for entries that never held one.

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 0.368× · real models 2.580×

MachineTime
  1. prod-amd64Seed 33625290960.985×Done1 attempt0.021× of the baseline kernels' time
  2. prod-arm64Seed 33625290960.963×Done1 attempt0.025× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.116×
  • Batch 1, sparse1.104×
  • Batch 1, wide1.089×

Mixer

  • Batch 1, normal0.987×
  • Batch 1, sparse1.013×
  • Batch 1, wide0.947×

ResNet

  • Batch 1, normal0.033×
  • Batch 1, sparse0.027×
  • Batch 1, wide0.033×

MobileNet

  • Batch 1, normal0.027×
  • Batch 1, sparse0.026×
  • Batch 1, wide0.027×

Recurrent

  • Batch 1, normal2.796×
  • Batch 1, sparse2.920×
  • Batch 1, wide2.806×

Normalized MLP

  • Batch 1, normal1.266×
  • Batch 1, sparse1.158×
  • Batch 1, wide1.139×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.019×
  • compact-mixer on digits1.928×
  • compact-mixer on fashion0.971×
  • compact-mixer on mnist1.000×
  • compact-mlp on digits0.862×
  • compact-mlp on fashion1.608×
  • compact-mlp on mnist2.346×
  • compact-mobilenet on digits2.727×
  • compact-mobilenet on fashion5.990×
  • compact-mobilenet on mnist8.299×
  • compact-normalized_mlp on digits0.860×
  • compact-normalized_mlp on fashion0.841×
  • compact-normalized_mlp on mnist0.913×
  • compact-resnet on digits1.713×
  • compact-resnet on fashion4.094×
  • compact-resnet on mnist7.213×
  • appliance-gru on appliance-energy2.448×
  • mitbih-gru on mitbih2.308×
  • mnist-rows-lstm on mnist-rows4.182×
  • speech-commands-gru on speech-commands2.937×
  • speech-commands-lstm-gru on speech-commands3.881×
  • spoken-arabic-digit-lstm on spoken-arabic-digit3.473×
  • tiny-shakespeare-lstm on tiny-shakespeare5.888×
  • uci-har-gru on uci-har2.495×
  • uci-har-lstm on uci-har3.239×
  • uci-har-mixer on UCI-HAR0.974×
  • uci-har-mlp on UCI-HAR1.116×
  • uci-har-mobilenet on UCI-HAR33.839×
  • uci-har-normalized_mlp on UCI-HAR0.959×
  • uci-har-resnet on UCI-HAR74.224×
MLP2 runs

RunsGenerated 1.026× · real models 0.903×

MachineTime
  1. prod-amd64Seed 33625290960.942×Done1 attempt0.015× of the baseline kernels' time
  2. prod-arm64Seed 33625290960.983×Done1 attempt0.022× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.018×
  • Batch 1, sparse0.961×
  • Batch 1, wide1.022×

Real models

  • compact-mlp on digits1.013×
  • compact-mlp on fashion0.989×
  • compact-mlp on mnist0.695×
  • uci-har-mlp on UCI-HAR0.954×
ResNet2 runs

RunsGenerated 0.059× · real models 15.988×

MachineTime
  1. prod-amd64Seed 33625290960.971×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 33625290960.978×Done1 attempt0.009× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal0.062×
  • Batch 1, sparse0.054×
  • Batch 1, wide0.066×

Real models

  • compact-resnet on digits3.390×
  • compact-resnet on fashion7.562×
  • compact-resnet on mnist12.357×
  • uci-har-resnet on UCI-HAR206.265×
MobileNet2 runs

RunsGenerated 0.046× · real models 15.739×

MachineTime
  1. prod-amd64Seed 33625290960.970×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 33625290960.749×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal0.038×
  • Batch 1, sparse0.054×
  • Batch 1, wide0.058×

Real models

  • compact-mobilenet on digits5.400×
  • compact-mobilenet on fashion10.973×
  • compact-mobilenet on mnist13.973×
  • uci-har-mobilenet on UCI-HAR74.109×
Normalized MLP2 runs

RunsGenerated 0.975× · real models 0.974×

MachineTime
  1. prod-amd64Seed 33625290960.978×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 33625290960.971×Done1 attempt0.013× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.110×
  • Batch 1, sparse0.936×
  • Batch 1, wide0.836×

Real models

  • compact-normalized_mlp on digits1.017×
  • compact-normalized_mlp on fashion1.133×
  • compact-normalized_mlp on mnist0.827×
  • uci-har-normalized_mlp on UCI-HAR0.945×
Mixer2 runs

RunsGenerated 0.975× · real models 1.031×

MachineTime
  1. prod-amd64Seed 33625290961.006×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 33625290960.999×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal1.006×
  • Batch 1, sparse0.972×
  • Batch 1, wide0.896×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.031×
  • compact-mixer on digits1.085×
  • compact-mixer on fashion1.033×
  • compact-mixer on mnist1.003×
  • uci-har-mixer on UCI-HAR1.008×
Recurrent2 runs

RunsGenerated 2.820× · real models 3.275×

MachineTime
  1. prod-amd64Seed 33625290963.040×Done1 attempt1.118× of the baseline kernels' time
  2. prod-arm64Seed 33625290963.039×Done1 attempt1.093× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal2.817×
  • Batch 1, sparse2.940×
  • Batch 1, wide2.821×

Real models

  • appliance-gru on appliance-energy2.416×
  • mitbih-gru on mitbih2.289×
  • mnist-rows-lstm on mnist-rows4.200×
  • speech-commands-gru on speech-commands2.914×
  • speech-commands-lstm-gru on speech-commands3.877×
  • spoken-arabic-digit-lstm on spoken-arabic-digit3.472×
  • tiny-shakespeare-lstm on tiny-shakespeare5.898×
  • uci-har-gru on uci-har2.486×
  • uci-har-lstm on uci-har3.248×