Conjectures.io
Tensor kernels

Entry #90

Completed
Hotkey
5DfgoW…bpo7zX
Sent
12 Nov 2026, 07:10 UTC
Source
611dde49a897

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 runs20.062×Generated 21.107×, real models 19.068×Above the barGain lower bound -2057.5%Guard passed
MLPvs #44Done2 of 2 runs1.113×Generated 0.970×, real models 1.278×Above the barGain lower bound -29.3%Guard passed
ResNetvs #51Done2 of 2 runs108.613×Generated 127.716×, real models 92.366×Above the barGain lower bound -13509.2%Guard passed
MobileNetvs #53Done2 of 2 runs96.946×Generated 111.404×, real models 84.364×Above the barGain lower bound -10654.8%Guard passed
Normalized MLPvs #48Done2 of 2 runs42.664×Generated 60.367×, real models 30.153×Above the barGain lower bound -5431.0%Guard passed
Mixervs #47Done2 of 2 runs174.230×Generated 182.888×, real models 165.981×Above the barGain lower bound -18646.2%Guard passed
Recurrentvs #54Done2 of 2 runs0.961×Generated 0.984×, real models 0.939×Not significantGain lower bound -0.8%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 21.107× · real models 19.068×

MachineTime
  1. prod-amd64Seed 139711942420.193×Done1 attempt0.200× of the baseline kernels' time
  2. prod-arm64Seed 139711942419.931×Done1 attempt0.217× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.056×
  • Batch 1, sparse1.012×
  • Batch 1, wide0.969×

Mixer

  • Batch 1, normal157.630×
  • Batch 1, sparse159.255×
  • Batch 1, wide202.611×

ResNet

  • Batch 1, normal112.682×
  • Batch 1, sparse129.225×
  • Batch 1, wide114.070×

MobileNet

  • Batch 1, normal90.455×
  • Batch 1, sparse92.624×
  • Batch 1, wide92.253×

Recurrent

  • Batch 1, normal0.875×
  • Batch 1, sparse0.884×
  • Batch 1, wide0.874×

Normalized MLP

  • Batch 1, normal60.354×
  • Batch 1, sparse60.316×
  • Batch 1, wide60.559×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction212.627×
  • compact-mixer on digits133.448×
  • compact-mixer on fashion160.609×
  • compact-mixer on mnist156.729×
  • compact-mlp on digits1.651×
  • compact-mlp on fashion1.405×
  • compact-mlp on mnist1.032×
  • compact-mobilenet on digits65.886×
  • compact-mobilenet on fashion82.580×
  • compact-mobilenet on mnist77.810×
  • compact-normalized_mlp on digits83.602×
  • compact-normalized_mlp on fashion32.593×
  • compact-normalized_mlp on mnist42.669×
  • compact-resnet on digits68.810×
  • compact-resnet on fashion120.215×
  • compact-resnet on mnist116.904×
  • appliance-gru on appliance-energy0.940×
  • mitbih-gru on mitbih0.888×
  • mnist-rows-lstm on mnist-rows0.700×
  • speech-commands-gru on speech-commands0.917×
  • speech-commands-lstm-gru on speech-commands0.901×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.870×
  • tiny-shakespeare-lstm on tiny-shakespeare0.734×
  • uci-har-gru on uci-har0.891×
  • uci-har-lstm on uci-har0.877×
  • uci-har-mixer on UCI-HAR160.743×
  • uci-har-mlp on UCI-HAR1.034×
  • uci-har-mobilenet on UCI-HAR76.810×
  • uci-har-normalized_mlp on UCI-HAR19.183×
  • uci-har-resnet on UCI-HAR82.170×
MLP2 runs

RunsGenerated 0.970× · real models 1.278×

MachineTime
  1. prod-amd64Seed 13971194241.005×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 13971194241.233×Done1 attempt0.019× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.931×
  • Batch 1, sparse0.924×
  • Batch 1, wide0.954×

Real models

  • compact-mlp on digits0.985×
  • compact-mlp on fashion1.326×
  • compact-mlp on mnist2.097×
  • uci-har-mlp on UCI-HAR0.973×
ResNet2 runs

RunsGenerated 127.716× · real models 92.366×

MachineTime
  1. prod-amd64Seed 1397119424105.494×Done1 attempt0.990× of the baseline kernels' time
  2. prod-arm64Seed 1397119424111.824×Done1 attempt0.997× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal150.968×
  • Batch 1, sparse130.239×
  • Batch 1, wide115.360×

Real models

  • compact-resnet on digits70.851×
  • compact-resnet on fashion119.014×
  • compact-resnet on mnist118.393×
  • uci-har-resnet on UCI-HAR72.909×
MobileNet2 runs

RunsGenerated 111.404× · real models 84.364×

MachineTime
  1. prod-amd64Seed 139711942494.019×Done1 attempt1.204× of the baseline kernels' time
  2. prod-arm64Seed 139711942499.964×Done1 attempt1.167× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal128.823×
  • Batch 1, sparse96.391×
  • Batch 1, wide94.280×

Real models

  • compact-mobilenet on digits67.710×
  • compact-mobilenet on fashion95.739×
  • compact-mobilenet on mnist93.574×
  • uci-har-mobilenet on UCI-HAR83.510×
Normalized MLP2 runs

RunsGenerated 60.367× · real models 30.153×

MachineTime
  1. prod-amd64Seed 139711942440.628×Done1 attempt0.495× of the baseline kernels' time
  2. prod-arm64Seed 139711942444.802×Done1 attempt0.502× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal59.597×
  • Batch 1, sparse65.862×
  • Batch 1, wide62.560×

Real models

  • compact-normalized_mlp on digits86.145×
  • compact-normalized_mlp on fashion22.911×
  • compact-normalized_mlp on mnist21.077×
  • uci-har-normalized_mlp on UCI-HAR19.871×
Mixer2 runs

RunsGenerated 182.888× · real models 165.981×

MachineTime
  1. prod-amd64Seed 1397119424173.556×Done1 attempt0.858× of the baseline kernels' time
  2. prod-arm64Seed 1397119424174.906×Done1 attempt0.859× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal181.799×
  • Batch 1, sparse176.005×
  • Batch 1, wide193.778×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction255.270×
  • compact-mixer on digits141.029×
  • compact-mixer on fashion157.538×
  • compact-mixer on mnist153.396×
  • uci-har-mixer on UCI-HAR144.807×
Recurrent2 runs

RunsGenerated 0.984× · real models 0.939×

MachineTime
  1. prod-amd64Seed 13971194240.976×Done1 attempt0.010× of the baseline kernels' time
  2. prod-arm64Seed 13971194240.946×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal1.023×
  • Batch 1, sparse0.967×
  • Batch 1, wide0.942×

Real models

  • appliance-gru on appliance-energy0.920×
  • mitbih-gru on mitbih0.883×
  • mnist-rows-lstm on mnist-rows1.012×
  • speech-commands-gru on speech-commands0.896×
  • speech-commands-lstm-gru on speech-commands0.919×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.951×
  • tiny-shakespeare-lstm on tiny-shakespeare1.055×
  • uci-har-gru on uci-har0.868×
  • uci-har-lstm on uci-har0.961×