Conjectures.io
Tensor kernels

Entry #88

Completed
Hotkey
5E7sEx…z1UAx7
Sent
12 Nov 2026, 06:15 UTC
Source
ab9cebfd598b

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 runs16.809×Generated 18.221×, real models 15.507×Above the barGain lower bound -1787.4%Guard passed
MLPvs #44Done2 of 2 runs1.009×Generated 1.155×, real models 0.882×Above the barGain lower bound -33.8%Guard passed
ResNetvs #51Done2 of 2 runs4.166×Generated 4.839×, real models 3.586×Above the barGain lower bound -554.1%Guard passed
MobileNetvs #53Done2 of 2 runs1.003×Generated 1.047×, real models 0.961×Above the barGain lower bound -11.8%Guard passed
Normalized MLPvs #48Done2 of 2 runs50.105×Generated 67.175×, real models 37.372×Above the barGain lower bound -5946.1%Guard passed
Mixervs #47Done2 of 2 runs169.854×Generated 176.104×, real models 163.826×Above the barGain lower bound -18348.0%Guard passed
Recurrentvs #54Done2 of 2 runs810.532×Generated 691.556×, real models 949.977×Above the barGain lower bound -84759.7%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 18.221× · real models 15.507×

MachineTime
  1. prod-amd64Seed 240492713016.798×Done1 attempt0.168× of the baseline kernels' time
  2. prod-arm64Seed 240492713016.820×Done1 attempt0.182× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.302×
  • Batch 1, sparse1.144×
  • Batch 1, wide0.945×

Mixer

  • Batch 1, normal165.390×
  • Batch 1, sparse170.010×
  • Batch 1, wide196.446×

ResNet

  • Batch 1, normal3.868×
  • Batch 1, sparse4.466×
  • Batch 1, wide3.882×

MobileNet

  • Batch 1, normal0.992×
  • Batch 1, sparse0.997×
  • Batch 1, wide1.039×

Recurrent

  • Batch 1, normal705.078×
  • Batch 1, sparse663.049×
  • Batch 1, wide664.276×

Normalized MLP

  • Batch 1, normal66.894×
  • Batch 1, sparse74.019×
  • Batch 1, wide71.766×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction198.470×
  • compact-mixer on digits144.210×
  • compact-mixer on fashion156.721×
  • compact-mixer on mnist138.976×
  • compact-mlp on digits0.883×
  • compact-mlp on fashion1.694×
  • compact-mlp on mnist0.972×
  • compact-mobilenet on digits0.833×
  • compact-mobilenet on fashion0.822×
  • compact-mobilenet on mnist0.810×
  • compact-normalized_mlp on digits87.605×
  • compact-normalized_mlp on fashion22.926×
  • compact-normalized_mlp on mnist18.376×
  • compact-resnet on digits2.477×
  • compact-resnet on fashion3.668×
  • compact-resnet on mnist3.661×
  • appliance-gru on appliance-energy760.911×
  • mitbih-gru on mitbih891.517×
  • mnist-rows-lstm on mnist-rows699.853×
  • speech-commands-gru on speech-commands700.033×
  • speech-commands-lstm-gru on speech-commands862.338×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1111.958×
  • tiny-shakespeare-lstm on tiny-shakespeare827.987×
  • uci-har-gru on uci-har830.176×
  • uci-har-lstm on uci-har1137.241×
  • uci-har-mixer on UCI-HAR165.260×
  • uci-har-mlp on UCI-HAR1.063×
  • uci-har-mobilenet on UCI-HAR0.943×
  • uci-har-normalized_mlp on UCI-HAR20.661×
  • uci-har-resnet on UCI-HAR5.230×
MLP2 runs

RunsGenerated 1.155× · real models 0.882×

MachineTime
  1. prod-amd64Seed 24049271301.007×Done1 attempt0.016× of the baseline kernels' time
  2. prod-arm64Seed 24049271301.011×Done1 attempt0.021× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.136×
  • Batch 1, sparse1.123×
  • Batch 1, wide1.193×

Real models

  • compact-mlp on digits0.604×
  • compact-mlp on fashion1.039×
  • compact-mlp on mnist1.050×
  • uci-har-mlp on UCI-HAR0.920×
ResNet2 runs

RunsGenerated 4.839× · real models 3.586×

MachineTime
  1. prod-amd64Seed 24049271303.802×Done1 attempt0.029× of the baseline kernels' time
  2. prod-arm64Seed 24049271304.564×Done1 attempt0.033× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal5.516×
  • Batch 1, sparse4.382×
  • Batch 1, wide3.929×

Real models

  • compact-resnet on digits2.439×
  • compact-resnet on fashion3.576×
  • compact-resnet on mnist3.606×
  • uci-har-resnet on UCI-HAR5.259×
MobileNet2 runs

RunsGenerated 1.047× · real models 0.961×

MachineTime
  1. prod-amd64Seed 24049271301.000×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 24049271301.006×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.017×
  • Batch 1, sparse0.996×
  • Batch 1, wide0.998×

Real models

  • compact-mobilenet on digits0.940×
  • compact-mobilenet on fashion0.964×
  • compact-mobilenet on mnist0.965×
  • uci-har-mobilenet on UCI-HAR0.977×
Normalized MLP2 runs

RunsGenerated 67.175× · real models 37.372×

MachineTime
  1. prod-amd64Seed 240492713053.197×Done1 attempt0.563× of the baseline kernels' time
  2. prod-arm64Seed 240492713047.192×Done1 attempt0.564× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal76.808×
  • Batch 1, sparse78.528×
  • Batch 1, wide80.871×

Real models

  • compact-normalized_mlp on digits70.779×
  • compact-normalized_mlp on fashion38.911×
  • compact-normalized_mlp on mnist34.088×
  • uci-har-normalized_mlp on UCI-HAR20.779×
Mixer2 runs

RunsGenerated 176.104× · real models 163.826×

MachineTime
  1. prod-amd64Seed 2404927130170.370×Done1 attempt0.843× of the baseline kernels' time
  2. prod-arm64Seed 2404927130169.339×Done1 attempt0.838× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal172.624×
  • Batch 1, sparse164.639×
  • Batch 1, wide207.287×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction262.533×
  • compact-mixer on digits133.448×
  • compact-mixer on fashion146.350×
  • compact-mixer on mnist157.357×
  • uci-har-mixer on UCI-HAR146.263×
Recurrent2 runs

RunsGenerated 691.556× · real models 949.977×

MachineTime
  1. prod-amd64Seed 2404927130842.294×Done1 attempt8.625× of the baseline kernels' time
  2. prod-arm64Seed 2404927130779.968×Done1 attempt8.849× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal708.028×
  • Batch 1, sparse695.357×
  • Batch 1, wide689.303×

Real models

  • appliance-gru on appliance-energy748.005×
  • mitbih-gru on mitbih898.781×
  • mnist-rows-lstm on mnist-rows1032.845×
  • speech-commands-gru on speech-commands669.789×
  • speech-commands-lstm-gru on speech-commands879.874×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1251.327×
  • tiny-shakespeare-lstm on tiny-shakespeare1200.168×
  • uci-har-gru on uci-har806.492×
  • uci-har-lstm on uci-har1271.309×