Conjectures.io
Tensor kernels

Entry #24

Completed
Hotkey
5FNxAE…drQNfZ
Sent
11 Nov 2026, 01:12 UTC
Source
410bf27e0113

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 #29Done2 of 2 runs1.062×Generated 1.104×, real models 1.020×Above the barGain lower bound -13.2%Guard passed
MLPvs #1Done2 of 2 runs1.168×Generated 1.082×, real models 1.262×Above the barGain lower bound -41.9%Guard passed
ResNetvs #27Done2 of 2 runs0.951×Generated 0.995×, real models 0.908×QualifiesGain lower bound 0.6%Guard passed
MobileNetvs #28Done2 of 2 runs0.928×Generated 1.008×, real models 0.854×QualifiesGain lower bound 3.9%Guard passed
Normalized MLPvs #6Done2 of 2 runs0.964×Generated 0.925×, real models 1.005×Not significantGain lower bound -36.8%Guard passed
Mixervs #24Done2 of 2 runs1.073×Generated 1.028×, real models 1.120×Above the barGain lower bound -16.3%Guard passed
Recurrentvs #25Done2 of 2 runs1.322×Generated 1.614×, real models 1.084×Above the barGain lower bound -41.7%Guard passed

Download Kernel.lean

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 1.104× · real models 1.020×

MachineTime
  1. prod-amd64Seed 14453319481.073×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 14453319481.050×Done1 attempt0.015× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.015×
  • Batch 1, sparse1.340×
  • Batch 1, wide1.028×

Mixer

  • Batch 1, normal0.993×
  • Batch 1, sparse1.123×
  • Batch 1, wide0.999×

ResNet

  • Batch 1, normal0.990×
  • Batch 1, sparse1.050×
  • Batch 1, wide0.988×

MobileNet

  • Batch 1, normal1.011×
  • Batch 1, sparse1.013×
  • Batch 1, wide0.970×

Recurrent

  • Batch 1, normal1.586×
  • Batch 1, sparse1.582×
  • Batch 1, wide1.579×

Normalized MLP

  • Batch 1, normal1.172×
  • Batch 1, sparse1.153×
  • Batch 1, wide1.035×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.321×
  • compact-mixer on digits2.921×
  • compact-mixer on fashion1.045×
  • compact-mixer on mnist0.991×
  • compact-mlp on digits1.096×
  • compact-mlp on fashion0.968×
  • compact-mlp on mnist0.758×
  • compact-mobilenet on digits0.902×
  • compact-mobilenet on fashion1.440×
  • compact-mobilenet on mnist1.562×
  • compact-normalized_mlp on digits1.081×
  • compact-normalized_mlp on fashion1.024×
  • compact-normalized_mlp on mnist0.715×
  • compact-resnet on digits1.097×
  • compact-resnet on fashion1.135×
  • compact-resnet on mnist1.358×
  • appliance-gru on appliance-energy1.266×
  • mitbih-gru on mitbih1.009×
  • mnist-rows-lstm on mnist-rows1.456×
  • speech-commands-gru on speech-commands1.382×
  • speech-commands-lstm-gru on speech-commands1.367×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.781×
  • tiny-shakespeare-lstm on tiny-shakespeare1.356×
  • uci-har-gru on uci-har1.157×
  • uci-har-lstm on uci-har0.529×
  • uci-har-mixer on UCI-HAR1.015×
  • uci-har-mlp on UCI-HAR1.010×
  • uci-har-mobilenet on UCI-HAR0.349×
  • uci-har-normalized_mlp on UCI-HAR1.024×
  • uci-har-resnet on UCI-HAR0.466×
MLP2 runs

RunsGenerated 1.082× · real models 1.262×

MachineTime
  1. prod-amd64Seed 14453319481.089×Done1 attempt0.018× of the baseline kernels' time
  2. prod-arm64Seed 14453319481.254×Done1 attempt0.020× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.058×
  • Batch 1, sparse1.109×
  • Batch 1, wide1.039×

Real models

  • compact-mlp on digits1.715×
  • compact-mlp on fashion1.004×
  • compact-mlp on mnist1.657×
  • uci-har-mlp on UCI-HAR0.890×
ResNet2 runs

RunsGenerated 0.995× · real models 0.908×

MachineTime
  1. prod-amd64Seed 14453319480.945×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 14453319480.956×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal1.023×
  • Batch 1, sparse0.994×
  • Batch 1, wide1.001×

Real models

  • compact-resnet on digits0.883×
  • compact-resnet on fashion0.902×
  • compact-resnet on mnist0.822×
  • uci-har-resnet on UCI-HAR1.037×
MobileNet2 runs

RunsGenerated 1.008× · real models 0.854×

MachineTime
  1. prod-amd64Seed 14453319480.917×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 14453319480.939×Done1 attempt0.015× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.000×
  • Batch 1, sparse1.013×
  • Batch 1, wide0.996×

Real models

  • compact-mobilenet on digits0.909×
  • compact-mobilenet on fashion0.801×
  • compact-mobilenet on mnist0.778×
  • uci-har-mobilenet on UCI-HAR0.938×
Normalized MLP2 runs

RunsGenerated 0.925× · real models 1.005×

MachineTime
  1. prod-amd64Seed 14453319480.920×Done1 attempt0.006× of the baseline kernels' time
  2. prod-arm64Seed 14453319481.011×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal0.998×
  • Batch 1, sparse0.957×
  • Batch 1, wide0.965×

Real models

  • compact-normalized_mlp on digits1.193×
  • compact-normalized_mlp on fashion1.006×
  • compact-normalized_mlp on mnist0.879×
  • uci-har-normalized_mlp on UCI-HAR0.967×
Mixer2 runs

RunsGenerated 1.028× · real models 1.120×

MachineTime
  1. prod-amd64Seed 14453319481.142×Done1 attempt0.006× of the baseline kernels' time
  2. prod-arm64Seed 14453319481.008×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.965×
  • Batch 1, sparse1.017×
  • Batch 1, wide1.036×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.074×
  • compact-mixer on digits0.977×
  • compact-mixer on fashion0.968×
  • compact-mixer on mnist1.679×
  • uci-har-mixer on UCI-HAR1.032×
Recurrent2 runs

RunsGenerated 1.614× · real models 1.084×

MachineTime
  1. prod-amd64Seed 14453319481.330×Done1 attempt0.068× of the baseline kernels' time
  2. prod-arm64Seed 14453319481.315×Done1 attempt0.069× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal1.598×
  • Batch 1, sparse1.601×
  • Batch 1, wide1.596×

Real models

  • appliance-gru on appliance-energy1.300×
  • mitbih-gru on mitbih1.006×
  • mnist-rows-lstm on mnist-rows1.459×
  • speech-commands-gru on speech-commands1.374×
  • speech-commands-lstm-gru on speech-commands1.374×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.713×
  • tiny-shakespeare-lstm on tiny-shakespeare1.348×
  • uci-har-gru on uci-har1.158×
  • uci-har-lstm on uci-har0.513×