Conjectures.io
Tensor kernels

Entry #81

Completed
Hotkey
5H9gTK…cEBhRH
Sent
12 Nov 2026, 02:43 UTC
Source
97e1dbd59e92

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 #50Done2 of 2 runs20.309×Generated 22.314×, real models 18.485×Above the barGain lower bound -2034.8%Guard passed
MLPvs #44Done2 of 2 runs0.999×Generated 0.813×, real models 1.229×Above the barGain lower bound -28.7%Guard passed
ResNetvs #51Done2 of 2 runs112.223×Generated 131.176×, real models 96.009×Above the barGain lower bound -11889.8%Guard passed
MobileNetvs #53Done2 of 2 runs100.707×Generated 121.858×, real models 83.226×Above the barGain lower bound -11422.8%Guard passed
Normalized MLPvs #48Done2 of 2 runs43.613×Generated 58.153×, real models 32.709×Above the barGain lower bound -5199.8%Guard passed
Mixervs #47Done2 of 2 runs163.129×Generated 172.086×, real models 154.638×Above the barGain lower bound -17956.7%Guard passed
Recurrentvs #54Done2 of 2 runs0.988×Generated 0.966×, real models 1.010×Above the barGain lower bound -6.6%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 22.314× · real models 18.485×

MachineTime
  1. prod-amd64Seed 205816187820.165×Done1 attempt0.205× of the baseline kernels' time
  2. prod-arm64Seed 205816187820.454×Done1 attempt0.210× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.992×
  • Batch 1, sparse0.992×
  • Batch 1, wide0.998×

Mixer

  • Batch 1, normal159.182×
  • Batch 1, sparse166.022×
  • Batch 1, wide189.239×

ResNet

  • Batch 1, normal121.098×
  • Batch 1, sparse141.334×
  • Batch 1, wide123.000×

MobileNet

  • Batch 1, normal96.680×
  • Batch 1, sparse99.270×
  • Batch 1, wide98.444×

Recurrent

  • Batch 1, normal0.937×
  • Batch 1, sparse0.907×
  • Batch 1, wide0.907×

Normalized MLP

  • Batch 1, normal58.496×
  • Batch 1, sparse54.628×
  • Batch 1, wide55.812×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction191.632×
  • compact-mixer on digits169.532×
  • compact-mixer on fashion154.886×
  • compact-mixer on mnist150.605×
  • compact-mlp on digits1.070×
  • compact-mlp on fashion1.839×
  • compact-mlp on mnist1.607×
  • compact-mobilenet on digits61.768×
  • compact-mobilenet on fashion83.956×
  • compact-mobilenet on mnist78.065×
  • compact-normalized_mlp on digits69.482×
  • compact-normalized_mlp on fashion18.759×
  • compact-normalized_mlp on mnist19.799×
  • compact-resnet on digits78.559×
  • compact-resnet on fashion121.935×
  • compact-resnet on mnist118.785×
  • appliance-gru on appliance-energy0.921×
  • mitbih-gru on mitbih0.962×
  • mnist-rows-lstm on mnist-rows0.774×
  • speech-commands-gru on speech-commands0.954×
  • speech-commands-lstm-gru on speech-commands0.945×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.937×
  • tiny-shakespeare-lstm on tiny-shakespeare0.796×
  • uci-har-gru on uci-har0.955×
  • uci-har-lstm on uci-har0.929×
  • uci-har-mixer on UCI-HAR168.675×
  • uci-har-mlp on UCI-HAR1.027×
  • uci-har-mobilenet on UCI-HAR78.893×
  • uci-har-normalized_mlp on UCI-HAR21.519×
  • uci-har-resnet on UCI-HAR74.925×
MLP2 runs

RunsGenerated 0.813× · real models 1.229×

MachineTime
  1. prod-amd64Seed 20581618781.026×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 20581618780.973×Done1 attempt0.015× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.735×
  • Batch 1, sparse0.729×
  • Batch 1, wide0.975×

Real models

  • compact-mlp on digits0.730×
  • compact-mlp on fashion2.060×
  • compact-mlp on mnist1.578×
  • uci-har-mlp on UCI-HAR0.960×
ResNet2 runs

RunsGenerated 131.176× · real models 96.009×

MachineTime
  1. prod-amd64Seed 2058161878117.356×Done1 attempt0.991× of the baseline kernels' time
  2. prod-arm64Seed 2058161878107.315×Done1 attempt0.998× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal119.911×
  • Batch 1, sparse140.720×
  • Batch 1, wide123.191×

Real models

  • compact-resnet on digits75.794×
  • compact-resnet on fashion119.588×
  • compact-resnet on mnist117.842×
  • uci-har-resnet on UCI-HAR79.546×
MobileNet2 runs

RunsGenerated 121.858× · real models 83.226×

MachineTime
  1. prod-amd64Seed 205816187895.205×Done1 attempt1.199× of the baseline kernels' time
  2. prod-arm64Seed 2058161878106.526×Done1 attempt1.154× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal144.889×
  • Batch 1, sparse99.580×
  • Batch 1, wide99.161×

Real models

  • compact-mobilenet on digits65.633×
  • compact-mobilenet on fashion96.609×
  • compact-mobilenet on mnist93.134×
  • uci-har-mobilenet on UCI-HAR81.245×
Normalized MLP2 runs

RunsGenerated 58.153× · real models 32.709×

MachineTime
  1. prod-amd64Seed 205816187842.327×Done1 attempt0.511× of the baseline kernels' time
  2. prod-arm64Seed 205816187844.938×Done1 attempt0.515× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal58.667×
  • Batch 1, sparse57.954×
  • Batch 1, wide57.989×

Real models

  • compact-normalized_mlp on digits79.322×
  • compact-normalized_mlp on fashion24.674×
  • compact-normalized_mlp on mnist26.722×
  • uci-har-normalized_mlp on UCI-HAR21.885×
Mixer2 runs

RunsGenerated 172.086× · real models 154.638×

MachineTime
  1. prod-amd64Seed 2058161878162.055×Done1 attempt0.858× of the baseline kernels' time
  2. prod-arm64Seed 2058161878164.209×Done1 attempt0.858× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal171.404×
  • Batch 1, sparse177.376×
  • Batch 1, wide196.458×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction259.602×
  • compact-mixer on digits94.554×
  • compact-mixer on fashion161.560×
  • compact-mixer on mnist155.232×
  • uci-har-mixer on UCI-HAR143.639×
Recurrent2 runs

RunsGenerated 0.966× · real models 1.010×

MachineTime
  1. prod-amd64Seed 20581618781.014×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 20581618780.962×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.970×
  • Batch 1, sparse0.983×
  • Batch 1, wide0.966×

Real models

  • appliance-gru on appliance-energy0.998×
  • mitbih-gru on mitbih0.974×
  • mnist-rows-lstm on mnist-rows1.065×
  • speech-commands-gru on speech-commands0.950×
  • speech-commands-lstm-gru on speech-commands0.954×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.029×
  • tiny-shakespeare-lstm on tiny-shakespeare1.146×
  • uci-har-gru on uci-har0.938×
  • uci-har-lstm on uci-har1.050×