Conjectures.io
Tensor kernels

Entry #73

Completed
Hotkey
5FpuC9…bc7LiY
Sent
12 Nov 2026, 00:23 UTC
Source
b657b827c664

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 runs17.210×Generated 17.481×, real models 16.943×Above the barGain lower bound -1729.1%Guard passed
MLPvs #44Done2 of 2 runs0.953×Generated 0.986×, real models 0.922×Not significantGain lower bound -28.4%Guard passed
ResNetvs #51Done2 of 2 runs3.797×Generated 3.865×, real models 3.730×Above the barGain lower bound -331.3%Guard passed
MobileNetvs #53Done2 of 2 runs0.985×Generated 0.940×, real models 1.032×Above the barGain lower bound -3.9%Guard passed
Normalized MLPvs #48Done2 of 2 runs39.402×Generated 54.349×, real models 28.566×Above the barGain lower bound -4841.5%Guard passed
Mixervs #47Done2 of 2 runs167.607×Generated 199.106×, real models 141.092×Above the barGain lower bound -19080.5%Guard passed
Recurrentvs #52Done2 of 2 runs780.156×Generated 691.594×, real models 880.059×Above the barGain lower bound -81811.4%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 17.481× · real models 16.943×

MachineTime
  1. prod-amd64Seed 365029495818.616×Done1 attempt0.174× of the baseline kernels' time
  2. prod-arm64Seed 365029495815.910×Done1 attempt0.178× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.145×
  • Batch 1, sparse1.242×
  • Batch 1, wide1.269×

Mixer

  • Batch 1, normal166.451×
  • Batch 1, sparse172.113×
  • Batch 1, wide197.028×

ResNet

  • Batch 1, normal3.655×
  • Batch 1, sparse4.150×
  • Batch 1, wide3.703×

MobileNet

  • Batch 1, normal0.997×
  • Batch 1, sparse1.011×
  • Batch 1, wide1.014×

Recurrent

  • Batch 1, normal709.004×
  • Batch 1, sparse638.675×
  • Batch 1, wide664.935×

Normalized MLP

  • Batch 1, normal50.923×
  • Batch 1, sparse53.638×
  • Batch 1, wide51.971×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction227.894×
  • compact-mixer on digits191.473×
  • compact-mixer on fashion159.114×
  • compact-mixer on mnist157.642×
  • compact-mlp on digits1.088×
  • compact-mlp on fashion2.157×
  • compact-mlp on mnist0.996×
  • compact-mobilenet on digits1.060×
  • compact-mobilenet on fashion0.860×
  • compact-mobilenet on mnist0.870×
  • compact-normalized_mlp on digits73.055×
  • compact-normalized_mlp on fashion32.913×
  • compact-normalized_mlp on mnist25.044×
  • compact-resnet on digits2.918×
  • compact-resnet on fashion3.697×
  • compact-resnet on mnist3.813×
  • appliance-gru on appliance-energy702.545×
  • mitbih-gru on mitbih890.245×
  • mnist-rows-lstm on mnist-rows729.365×
  • speech-commands-gru on speech-commands685.841×
  • speech-commands-lstm-gru on speech-commands868.799×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1121.522×
  • tiny-shakespeare-lstm on tiny-shakespeare818.756×
  • uci-har-gru on uci-har839.511×
  • uci-har-lstm on uci-har1147.915×
  • uci-har-mixer on UCI-HAR166.972×
  • uci-har-mlp on UCI-HAR1.048×
  • uci-har-mobilenet on UCI-HAR0.967×
  • uci-har-normalized_mlp on UCI-HAR21.218×
  • uci-har-resnet on UCI-HAR5.844×
MLP2 runs

RunsGenerated 0.986× · real models 0.922×

MachineTime
  1. prod-amd64Seed 36502949580.798×Done1 attempt0.016× of the baseline kernels' time
  2. prod-arm64Seed 36502949581.138×Done1 attempt0.016× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.982×
  • Batch 1, sparse0.981×
  • Batch 1, wide1.077×

Real models

  • compact-mlp on digits0.936×
  • compact-mlp on fashion0.924×
  • compact-mlp on mnist0.978×
  • uci-har-mlp on UCI-HAR0.854×
ResNet2 runs

RunsGenerated 3.865× · real models 3.730×

MachineTime
  1. prod-amd64Seed 36502949583.765×Done1 attempt0.031× of the baseline kernels' time
  2. prod-arm64Seed 36502949583.829×Done1 attempt0.035× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal3.945×
  • Batch 1, sparse4.160×
  • Batch 1, wide3.609×

Real models

  • compact-resnet on digits2.738×
  • compact-resnet on fashion3.695×
  • compact-resnet on mnist3.856×
  • uci-har-resnet on UCI-HAR4.962×
MobileNet2 runs

RunsGenerated 0.940× · real models 1.032×

MachineTime
  1. prod-amd64Seed 36502949581.000×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 36502949580.971×Done1 attempt0.009× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal0.974×
  • Batch 1, sparse1.000×
  • Batch 1, wide0.933×

Real models

  • compact-mobilenet on digits1.062×
  • compact-mobilenet on fashion1.023×
  • compact-mobilenet on mnist1.026×
  • uci-har-mobilenet on UCI-HAR1.019×
Normalized MLP2 runs

RunsGenerated 54.349× · real models 28.566×

MachineTime
  1. prod-amd64Seed 365029495839.820×Done1 attempt0.518× of the baseline kernels' time
  2. prod-arm64Seed 365029495838.989×Done1 attempt0.527× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal49.618×
  • Batch 1, sparse52.817×
  • Batch 1, wide50.106×

Real models

  • compact-normalized_mlp on digits83.288×
  • compact-normalized_mlp on fashion19.589×
  • compact-normalized_mlp on mnist18.322×
  • uci-har-normalized_mlp on UCI-HAR22.276×
Mixer2 runs

RunsGenerated 199.106× · real models 141.092×

MachineTime
  1. prod-amd64Seed 3650294958181.002×Done1 attempt0.860× of the baseline kernels' time
  2. prod-arm64Seed 3650294958155.204×Done1 attempt0.862× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal181.018×
  • Batch 1, sparse182.633×
  • Batch 1, wide241.224×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction241.701×
  • compact-mixer on digits68.296×
  • compact-mixer on fashion157.310×
  • compact-mixer on mnist155.314×
  • uci-har-mixer on UCI-HAR138.631×
Recurrent2 runs

RunsGenerated 691.594× · real models 880.059×

MachineTime
  1. prod-amd64Seed 3650294958813.503×Done1 attempt8.223× of the baseline kernels' time
  2. prod-arm64Seed 3650294958748.176×Done1 attempt8.616× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal742.967×
  • Batch 1, sparse670.115×
  • Batch 1, wide678.115×

Real models

  • appliance-gru on appliance-energy772.441×
  • mitbih-gru on mitbih908.915×
  • mnist-rows-lstm on mnist-rows732.118×
  • speech-commands-gru on speech-commands694.205×
  • speech-commands-lstm-gru on speech-commands887.132×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1154.758×
  • tiny-shakespeare-lstm on tiny-shakespeare832.411×
  • uci-har-gru on uci-har862.862×
  • uci-har-lstm on uci-har1206.120×