Conjectures.io
Tensor kernels

Entry #89

Completed
Hotkey
5DUGsM…ww9J6S
Sent
12 Nov 2026, 06:16 UTC
Source
d4d128808675

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 runs17.425×Generated 17.647×, real models 17.206×Above the barGain lower bound -1756.0%Guard passed
MLPvs #44Done2 of 2 runs1.090×Generated 0.832×, real models 1.427×Above the barGain lower bound -32.9%Guard passed
ResNetvs #51Done2 of 2 runs3.676×Generated 3.806×, real models 3.551×Above the barGain lower bound -295.0%Guard passed
MobileNetvs #53Done2 of 2 runs0.968×Generated 0.969×, real models 0.967×Not significantGain lower bound -7.3%Guard passed
Normalized MLPvs #48Done2 of 2 runs44.600×Generated 59.766×, real models 33.282×Above the barGain lower bound -5552.7%Guard passed
Mixervs #47Done2 of 2 runs179.355×Generated 190.336×, real models 169.007×Above the barGain lower bound -19992.2%Guard passed
Recurrentvs #54Done2 of 2 runs837.604×Generated 737.096×, real models 951.817×Above the barGain lower bound -87473.5%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 17.647× · real models 17.206×

MachineTime
  1. prod-amd64Seed 364105883617.260×Done1 attempt0.159× of the baseline kernels' time
  2. prod-arm64Seed 364105883617.592×Done1 attempt0.191× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.283×
  • Batch 1, sparse1.021×
  • Batch 1, wide1.309×

Mixer

  • Batch 1, normal203.693×
  • Batch 1, sparse167.804×
  • Batch 1, wide212.363×

ResNet

  • Batch 1, normal3.738×
  • Batch 1, sparse4.286×
  • Batch 1, wide3.732×

MobileNet

  • Batch 1, normal1.024×
  • Batch 1, sparse1.001×
  • Batch 1, wide1.017×

Recurrent

  • Batch 1, normal699.940×
  • Batch 1, sparse661.428×
  • Batch 1, wide665.782×

Normalized MLP

  • Batch 1, normal50.851×
  • Batch 1, sparse57.035×
  • Batch 1, wide57.191×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction230.534×
  • compact-mixer on digits183.681×
  • compact-mixer on fashion155.474×
  • compact-mixer on mnist147.688×
  • compact-mlp on digits1.634×
  • compact-mlp on fashion1.703×
  • compact-mlp on mnist1.648×
  • compact-mobilenet on digits0.807×
  • compact-mobilenet on fashion0.838×
  • compact-mobilenet on mnist0.836×
  • compact-normalized_mlp on digits68.611×
  • compact-normalized_mlp on fashion35.297×
  • compact-normalized_mlp on mnist38.036×
  • compact-resnet on digits2.729×
  • compact-resnet on fashion3.566×
  • compact-resnet on mnist3.595×
  • appliance-gru on appliance-energy701.696×
  • mitbih-gru on mitbih911.436×
  • mnist-rows-lstm on mnist-rows730.409×
  • speech-commands-gru on speech-commands694.656×
  • speech-commands-lstm-gru on speech-commands866.603×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1143.474×
  • tiny-shakespeare-lstm on tiny-shakespeare832.323×
  • uci-har-gru on uci-har847.005×
  • uci-har-lstm on uci-har1159.179×
  • uci-har-mixer on UCI-HAR158.181×
  • uci-har-mlp on UCI-HAR1.123×
  • uci-har-mobilenet on UCI-HAR0.903×
  • uci-har-normalized_mlp on UCI-HAR20.637×
  • uci-har-resnet on UCI-HAR5.275×
MLP2 runs

RunsGenerated 0.832× · real models 1.427×

MachineTime
  1. prod-amd64Seed 36410588361.033×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 36410588361.149×Done1 attempt0.023× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.952×
  • Batch 1, sparse0.738×
  • Batch 1, wide0.840×

Real models

  • compact-mlp on digits0.834×
  • compact-mlp on fashion2.292×
  • compact-mlp on mnist2.096×
  • uci-har-mlp on UCI-HAR1.034×
ResNet2 runs

RunsGenerated 3.806× · real models 3.551×

MachineTime
  1. prod-amd64Seed 36410588363.752×Done1 attempt0.029× of the baseline kernels' time
  2. prod-arm64Seed 36410588363.601×Done1 attempt0.032× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal3.640×
  • Batch 1, sparse4.172×
  • Batch 1, wide3.653×

Real models

  • compact-resnet on digits2.468×
  • compact-resnet on fashion3.685×
  • compact-resnet on mnist3.712×
  • uci-har-resnet on UCI-HAR4.709×
MobileNet2 runs

RunsGenerated 0.969× · real models 0.967×

MachineTime
  1. prod-amd64Seed 36410588360.960×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 36410588360.975×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal0.970×
  • Batch 1, sparse1.019×
  • Batch 1, wide1.037×

Real models

  • compact-mobilenet on digits0.925×
  • compact-mobilenet on fashion0.986×
  • compact-mobilenet on mnist0.985×
  • uci-har-mobilenet on UCI-HAR0.973×
Normalized MLP2 runs

RunsGenerated 59.766× · real models 33.282×

MachineTime
  1. prod-amd64Seed 364105883646.319×Done1 attempt0.532× of the baseline kernels' time
  2. prod-arm64Seed 364105883642.944×Done1 attempt0.536× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal57.429×
  • Batch 1, sparse60.436×
  • Batch 1, wide63.599×

Real models

  • compact-normalized_mlp on digits80.288×
  • compact-normalized_mlp on fashion19.545×
  • compact-normalized_mlp on mnist35.176×
  • uci-har-normalized_mlp on UCI-HAR22.228×
Mixer2 runs

RunsGenerated 190.336× · real models 169.007×

MachineTime
  1. prod-amd64Seed 3641058836187.610×Done1 attempt0.850× of the baseline kernels' time
  2. prod-arm64Seed 3641058836171.463×Done1 attempt0.852× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal173.744×
  • Batch 1, sparse180.746×
  • Batch 1, wide227.992×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction247.642×
  • compact-mixer on digits161.004×
  • compact-mixer on fashion158.724×
  • compact-mixer on mnist155.131×
  • uci-har-mixer on UCI-HAR140.451×
Recurrent2 runs

RunsGenerated 737.096× · real models 951.817×

MachineTime
  1. prod-amd64Seed 3641058836901.819×Done1 attempt7.568× of the baseline kernels' time
  2. prod-arm64Seed 3641058836777.962×Done1 attempt7.647× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal763.336×
  • Batch 1, sparse743.202×
  • Batch 1, wide713.138×

Real models

  • appliance-gru on appliance-energy743.693×
  • mitbih-gru on mitbih892.862×
  • mnist-rows-lstm on mnist-rows1042.972×
  • speech-commands-gru on speech-commands666.329×
  • speech-commands-lstm-gru on speech-commands884.821×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1245.866×
  • tiny-shakespeare-lstm on tiny-shakespeare1201.402×
  • uci-har-gru on uci-har814.971×
  • uci-har-lstm on uci-har1287.317×