Conjectures.io
Tensor kernels

Entry #14

Completed
Hotkey
5Ebgpz…CPCcJs
Sent
10 Nov 2026, 22:07 UTC
Source
6b6f0602d0b4

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 #26Done2 of 2 runs1.933×Generated 1.137×, real models 3.288×Above the barGain lower bound -127.1%Guard passed
MLPvs #1Done2 of 2 runs1.036×Generated 0.972×, real models 1.105×Above the barGain lower bound -25.2%Guard passed
ResNetvs #20Done2 of 2 runs3.371×Generated 1.161×, real models 9.790×Above the barGain lower bound -584.9%Guard passed
MobileNetvs #21Done2 of 2 runs4.519×Generated 1.143×, real models 17.861×Above the barGain lower bound -745.1%Guard passed
Normalized MLPvs #6Done2 of 2 runs1.059×Generated 1.114×, real models 1.006×Above the barGain lower bound -26.2%Guard passed
Mixervs #24Done2 of 2 runs1.886×Generated 0.974×, real models 3.655×Above the barGain lower bound -163.0%Guard passed
Recurrentvs #25Done2 of 2 runs1.384×Generated 1.560×, real models 1.229×Above the barGain lower bound -44.6%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 1.137× · real models 3.288×

MachineTime
  1. prod-amd64Seed 16271720092.051×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 16271720091.822×Done1 attempt0.016× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.266×
  • Batch 1, sparse1.154×
  • Batch 1, wide1.096×

Mixer

  • Batch 1, normal1.022×
  • Batch 1, sparse1.107×
  • Batch 1, wide1.055×

ResNet

  • Batch 1, normal0.980×
  • Batch 1, sparse1.014×
  • Batch 1, wide0.980×

MobileNet

  • Batch 1, normal0.960×
  • Batch 1, sparse0.987×
  • Batch 1, wide0.957×

Recurrent

  • Batch 1, normal1.564×
  • Batch 1, sparse1.570×
  • Batch 1, wide1.569×

Normalized MLP

  • Batch 1, normal1.105×
  • Batch 1, sparse1.279×
  • Batch 1, wide1.227×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.368×
  • compact-mixer on digits11.499×
  • compact-mixer on fashion11.097×
  • compact-mixer on mnist15.317×
  • compact-mlp on digits1.067×
  • compact-mlp on fashion0.466×
  • compact-mlp on mnist1.592×
  • compact-mobilenet on digits40.456×
  • compact-mobilenet on fashion77.342×
  • compact-mobilenet on mnist82.568×
  • compact-normalized_mlp on digits1.041×
  • compact-normalized_mlp on fashion2.011×
  • compact-normalized_mlp on mnist0.953×
  • compact-resnet on digits19.157×
  • compact-resnet on fashion48.185×
  • compact-resnet on mnist75.392×
  • appliance-gru on appliance-energy1.290×
  • mitbih-gru on mitbih1.029×
  • mnist-rows-lstm on mnist-rows1.461×
  • speech-commands-gru on speech-commands1.369×
  • speech-commands-lstm-gru on speech-commands1.389×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.281×
  • tiny-shakespeare-lstm on tiny-shakespeare1.361×
  • uci-har-gru on uci-har1.175×
  • uci-har-lstm on uci-har0.920×
  • uci-har-mixer on UCI-HAR1.073×
  • uci-har-mlp on UCI-HAR1.232×
  • uci-har-mobilenet on UCI-HAR0.384×
  • uci-har-normalized_mlp on UCI-HAR0.953×
  • uci-har-resnet on UCI-HAR0.144×
MLP2 runs

RunsGenerated 0.972× · real models 1.105×

MachineTime
  1. prod-amd64Seed 16271720090.909×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 16271720091.181×Done1 attempt0.019× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.033×
  • Batch 1, sparse0.981×
  • Batch 1, wide0.987×

Real models

  • compact-mlp on digits2.035×
  • compact-mlp on fashion1.037×
  • compact-mlp on mnist0.770×
  • uci-har-mlp on UCI-HAR0.917×
ResNet2 runs

RunsGenerated 1.161× · real models 9.790×

MachineTime
  1. prod-amd64Seed 16271720093.161×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 16271720093.594×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal1.142×
  • Batch 1, sparse1.034×
  • Batch 1, wide1.027×

Real models

  • compact-resnet on digits18.734×
  • compact-resnet on fashion47.667×
  • compact-resnet on mnist73.196×
  • uci-har-resnet on UCI-HAR0.141×
MobileNet2 runs

RunsGenerated 1.143× · real models 17.861×

MachineTime
  1. prod-amd64Seed 16271720094.285×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 16271720094.766×Done1 attempt0.014× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.396×
  • Batch 1, sparse1.015×
  • Batch 1, wide0.971×

Real models

  • compact-mobilenet on digits41.163×
  • compact-mobilenet on fashion77.177×
  • compact-mobilenet on mnist81.840×
  • uci-har-mobilenet on UCI-HAR0.391×
Normalized MLP2 runs

RunsGenerated 1.114× · real models 1.006×

MachineTime
  1. prod-amd64Seed 16271720091.063×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 16271720091.054×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.081×
  • Batch 1, sparse1.096×
  • Batch 1, wide1.122×

Real models

  • compact-normalized_mlp on digits1.134×
  • compact-normalized_mlp on fashion1.056×
  • compact-normalized_mlp on mnist0.945×
  • uci-har-normalized_mlp on UCI-HAR0.905×
Mixer2 runs

RunsGenerated 0.974× · real models 3.655×

MachineTime
  1. prod-amd64Seed 16271720092.041×Done1 attempt0.004× of the baseline kernels' time
  2. prod-arm64Seed 16271720091.743×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.871×
  • Batch 1, sparse0.956×
  • Batch 1, wide0.982×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.339×
  • compact-mixer on digits8.695×
  • compact-mixer on fashion6.067×
  • compact-mixer on mnist9.485×
  • uci-har-mixer on UCI-HAR0.974×
Recurrent2 runs

RunsGenerated 1.560× · real models 1.229×

MachineTime
  1. prod-amd64Seed 16271720091.381×Done1 attempt0.065× of the baseline kernels' time
  2. prod-arm64Seed 16271720091.388×Done1 attempt0.069× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal1.556×
  • Batch 1, sparse1.552×
  • Batch 1, wide1.565×

Real models

  • appliance-gru on appliance-energy1.271×
  • mitbih-gru on mitbih1.016×
  • mnist-rows-lstm on mnist-rows1.469×
  • speech-commands-gru on speech-commands1.362×
  • speech-commands-lstm-gru on speech-commands1.390×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.272×
  • tiny-shakespeare-lstm on tiny-shakespeare1.349×
  • uci-har-gru on uci-har1.151×
  • uci-har-lstm on uci-har0.902×