Conjectures.io
Tensor kernels

Entry #83

Completed
Hotkey
5DemNC…7yxv4i
Sent
12 Nov 2026, 03:31 UTC
Source
5ca2bf130eb7

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 runs16.196×Generated 17.194×, real models 15.255×Above the barGain lower bound -1629.2%Guard passed
MLPvs #44Done2 of 2 runs0.937×Generated 0.982×, real models 0.894×Not significantGain lower bound -16.6%Guard passed
ResNetvs #51Done2 of 2 runs4.208×Generated 4.775×, real models 3.708×Above the barGain lower bound -525.5%Guard passed
MobileNetvs #53Done2 of 2 runs0.992×Generated 1.043×, real models 0.944×Above the barGain lower bound -46.7%Guard passed
Normalized MLPvs #48Done2 of 2 runs40.092×Generated 53.039×, real models 30.305×Above the barGain lower bound -4798.3%Guard passed
Mixervs #47Done2 of 2 runs175.293×Generated 197.571×, real models 155.527×Above the barGain lower bound -19169.3%Guard passed
Recurrentvs #54Done2 of 2 runs801.806×Generated 686.034×, real models 937.115×Above the barGain lower bound -84202.4%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 17.194× · real models 15.255×

MachineTime
  1. prod-amd64Seed 426237363516.905×Done1 attempt0.164× of the baseline kernels' time
  2. prod-arm64Seed 426237363515.516×Done1 attempt0.180× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.981×
  • Batch 1, sparse1.108×
  • Batch 1, wide1.071×

Mixer

  • Batch 1, normal165.657×
  • Batch 1, sparse166.644×
  • Batch 1, wide192.486×

ResNet

  • Batch 1, normal3.863×
  • Batch 1, sparse4.436×
  • Batch 1, wide3.927×

MobileNet

  • Batch 1, normal1.018×
  • Batch 1, sparse1.022×
  • Batch 1, wide1.035×

Recurrent

  • Batch 1, normal718.786×
  • Batch 1, sparse658.762×
  • Batch 1, wide664.534×

Normalized MLP

  • Batch 1, normal46.436×
  • Batch 1, sparse53.757×
  • Batch 1, wide44.340×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction221.393×
  • compact-mixer on digits154.902×
  • compact-mixer on fashion153.730×
  • compact-mixer on mnist149.149×
  • compact-mlp on digits0.871×
  • compact-mlp on fashion1.316×
  • compact-mlp on mnist0.881×
  • compact-mobilenet on digits0.844×
  • compact-mobilenet on fashion0.852×
  • compact-mobilenet on mnist0.804×
  • compact-normalized_mlp on digits59.958×
  • compact-normalized_mlp on fashion22.084×
  • compact-normalized_mlp on mnist21.364×
  • compact-resnet on digits2.752×
  • compact-resnet on fashion3.850×
  • compact-resnet on mnist3.764×
  • appliance-gru on appliance-energy718.683×
  • mitbih-gru on mitbih887.533×
  • mnist-rows-lstm on mnist-rows723.228×
  • speech-commands-gru on speech-commands710.030×
  • speech-commands-lstm-gru on speech-commands863.133×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1129.773×
  • tiny-shakespeare-lstm on tiny-shakespeare816.773×
  • uci-har-gru on uci-har831.851×
  • uci-har-lstm on uci-har1145.977×
  • uci-har-mixer on UCI-HAR165.611×
  • uci-har-mlp on UCI-HAR0.978×
  • uci-har-mobilenet on UCI-HAR0.925×
  • uci-har-normalized_mlp on UCI-HAR19.336×
  • uci-har-resnet on UCI-HAR5.217×
MLP2 runs

RunsGenerated 0.982× · real models 0.894×

MachineTime
  1. prod-amd64Seed 42623736350.963×Done1 attempt0.015× of the baseline kernels' time
  2. prod-arm64Seed 42623736350.911×Done1 attempt0.015× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.006×
  • Batch 1, sparse0.998×
  • Batch 1, wide0.963×

Real models

  • compact-mlp on digits0.699×
  • compact-mlp on fashion1.002×
  • compact-mlp on mnist0.815×
  • uci-har-mlp on UCI-HAR1.116×
ResNet2 runs

RunsGenerated 4.775× · real models 3.708×

MachineTime
  1. prod-amd64Seed 42623736353.946×Done1 attempt0.030× of the baseline kernels' time
  2. prod-arm64Seed 42623736354.487×Done1 attempt0.034× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal5.478×
  • Batch 1, sparse4.421×
  • Batch 1, wide3.951×

Real models

  • compact-resnet on digits3.026×
  • compact-resnet on fashion3.703×
  • compact-resnet on mnist3.568×
  • uci-har-resnet on UCI-HAR4.729×
MobileNet2 runs

RunsGenerated 1.043× · real models 0.944×

MachineTime
  1. prod-amd64Seed 42623736350.962×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 42623736351.024×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.018×
  • Batch 1, sparse1.024×
  • Batch 1, wide0.993×

Real models

  • compact-mobilenet on digits0.882×
  • compact-mobilenet on fashion0.972×
  • compact-mobilenet on mnist0.966×
  • uci-har-mobilenet on UCI-HAR0.958×
Normalized MLP2 runs

RunsGenerated 53.039× · real models 30.305×

MachineTime
  1. prod-amd64Seed 426237363539.780×Done1 attempt0.477× of the baseline kernels' time
  2. prod-arm64Seed 426237363540.406×Done1 attempt0.487× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal52.408×
  • Batch 1, sparse51.089×
  • Batch 1, wide59.320×

Real models

  • compact-normalized_mlp on digits67.246×
  • compact-normalized_mlp on fashion25.619×
  • compact-normalized_mlp on mnist22.387×
  • uci-har-normalized_mlp on UCI-HAR21.869×
Mixer2 runs

RunsGenerated 197.571× · real models 155.527×

MachineTime
  1. prod-amd64Seed 4262373635182.284×Done1 attempt0.859× of the baseline kernels' time
  2. prod-arm64Seed 4262373635168.570×Done1 attempt0.868× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal184.010×
  • Batch 1, sparse213.674×
  • Batch 1, wide218.114×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction207.320×
  • compact-mixer on digits135.705×
  • compact-mixer on fashion152.263×
  • compact-mixer on mnist151.381×
  • uci-har-mixer on UCI-HAR140.324×
Recurrent2 runs

RunsGenerated 686.034× · real models 937.115×

MachineTime
  1. prod-amd64Seed 4262373635833.420×Done1 attempt8.271× of the baseline kernels' time
  2. prod-arm64Seed 4262373635771.391×Done1 attempt8.514× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal690.971×
  • Batch 1, sparse673.787×
  • Batch 1, wide651.000×

Real models

  • appliance-gru on appliance-energy724.948×
  • mitbih-gru on mitbih879.804×
  • mnist-rows-lstm on mnist-rows1034.383×
  • speech-commands-gru on speech-commands682.376×
  • speech-commands-lstm-gru on speech-commands867.068×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1201.940×
  • tiny-shakespeare-lstm on tiny-shakespeare1186.517×
  • uci-har-gru on uci-har789.019×
  • uci-har-lstm on uci-har1268.935×