Conjectures.io
Tensor kernels

Entry #21

Completed
Hotkey
5DaLxp…jqBZqa
Sent
11 Nov 2026, 00:40 UTC
Source
942bd275d9d5

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 runs0.851×Generated 0.997×, real models 0.727×QualifiesGain lower bound 5.4%Guard passed
MLPvs #1Done2 of 2 runs0.952×Generated 0.948×, real models 0.956×Not significantGain lower bound -20.0%Guard passed
ResNetvs #27Done2 of 2 runs0.998×Generated 1.015×, real models 0.981×Above the barGain lower bound -15.6%Guard passed
MobileNetvs #28Done2 of 2 runs0.868×Generated 0.814×, real models 0.925×Not significantGain lower bound -11.2%Guard passed
Normalized MLPvs #6Done2 of 2 runs1.007×Generated 1.043×, real models 0.972×Above the barGain lower bound -10.3%Guard passed
Mixervs #24Done2 of 2 runs1.022×Generated 0.985×, real models 1.060×Above the barGain lower bound -10.8%Guard passed
Recurrentvs #25Done2 of 2 runs0.999×Generated 0.999×, real models 0.999×Above the barGain lower bound -1.0%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 0.997× · real models 0.727×

MachineTime
  1. prod-amd64Seed 33501012340.864×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 33501012340.838×Done1 attempt0.013× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.042×
  • Batch 1, sparse0.994×
  • Batch 1, wide1.055×

Mixer

  • Batch 1, normal1.024×
  • Batch 1, sparse0.962×
  • Batch 1, wide0.991×

ResNet

  • Batch 1, normal0.992×
  • Batch 1, sparse0.953×
  • Batch 1, wide0.981×

MobileNet

  • Batch 1, normal0.977×
  • Batch 1, sparse0.993×
  • Batch 1, wide1.017×

Recurrent

  • Batch 1, normal0.998×
  • Batch 1, sparse1.000×
  • Batch 1, wide0.996×

Normalized MLP

  • Batch 1, normal0.935×
  • Batch 1, sparse1.052×
  • Batch 1, wide0.974×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.937×
  • compact-mixer on digits0.471×
  • compact-mixer on fashion1.017×
  • compact-mixer on mnist1.017×
  • compact-mlp on digits1.668×
  • compact-mlp on fashion0.696×
  • compact-mlp on mnist1.415×
  • compact-mobilenet on digits0.624×
  • compact-mobilenet on fashion0.420×
  • compact-mobilenet on mnist0.363×
  • compact-normalized_mlp on digits1.001×
  • compact-normalized_mlp on fashion0.786×
  • compact-normalized_mlp on mnist0.831×
  • compact-resnet on digits0.876×
  • compact-resnet on fashion0.653×
  • compact-resnet on mnist0.428×
  • appliance-gru on appliance-energy0.999×
  • mitbih-gru on mitbih0.995×
  • mnist-rows-lstm on mnist-rows1.004×
  • speech-commands-gru on speech-commands1.002×
  • speech-commands-lstm-gru on speech-commands1.004×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.000×
  • tiny-shakespeare-lstm on tiny-shakespeare0.995×
  • uci-har-gru on uci-har0.998×
  • uci-har-lstm on uci-har0.999×
  • uci-har-mixer on UCI-HAR0.995×
  • uci-har-mlp on UCI-HAR1.056×
  • uci-har-mobilenet on UCI-HAR0.262×
  • uci-har-normalized_mlp on UCI-HAR0.962×
  • uci-har-resnet on UCI-HAR0.133×
MLP2 runs

RunsGenerated 0.948× · real models 0.956×

MachineTime
  1. prod-amd64Seed 33501012340.826×Done1 attempt0.017× of the baseline kernels' time
  2. prod-arm64Seed 33501012341.098×Done1 attempt0.016× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.986×
  • Batch 1, sparse0.935×
  • Batch 1, wide0.930×

Real models

  • compact-mlp on digits0.500×
  • compact-mlp on fashion0.992×
  • compact-mlp on mnist1.964×
  • uci-har-mlp on UCI-HAR0.859×
ResNet2 runs

RunsGenerated 1.015× · real models 0.981×

MachineTime
  1. prod-amd64Seed 33501012341.015×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 33501012340.981×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal1.025×
  • Batch 1, sparse1.005×
  • Batch 1, wide0.999×

Real models

  • compact-resnet on digits0.992×
  • compact-resnet on fashion0.764×
  • compact-resnet on mnist0.578×
  • uci-har-resnet on UCI-HAR2.115×
MobileNet2 runs

RunsGenerated 0.814× · real models 0.925×

MachineTime
  1. prod-amd64Seed 33501012340.988×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 33501012340.762×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal0.672×
  • Batch 1, sparse0.996×
  • Batch 1, wide1.040×

Real models

  • compact-mobilenet on digits0.909×
  • compact-mobilenet on fashion0.579×
  • compact-mobilenet on mnist0.518×
  • uci-har-mobilenet on UCI-HAR2.687×
Normalized MLP2 runs

RunsGenerated 1.043× · real models 0.972×

MachineTime
  1. prod-amd64Seed 33501012341.068×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 33501012340.949×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.009×
  • Batch 1, sparse1.007×
  • Batch 1, wide1.041×

Real models

  • compact-normalized_mlp on digits1.112×
  • compact-normalized_mlp on fashion1.003×
  • compact-normalized_mlp on mnist0.825×
  • uci-har-normalized_mlp on UCI-HAR0.970×
Mixer2 runs

RunsGenerated 0.985× · real models 1.060×

MachineTime
  1. prod-amd64Seed 33501012341.019×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 33501012341.024×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.954×
  • Batch 1, sparse0.969×
  • Batch 1, wide1.062×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.941×
  • compact-mixer on digits1.402×
  • compact-mixer on fashion1.000×
  • compact-mixer on mnist1.005×
  • uci-har-mixer on UCI-HAR1.010×
Recurrent2 runs

RunsGenerated 0.999× · real models 0.999×

MachineTime
  1. prod-amd64Seed 33501012341.009×Done1 attempt0.041× of the baseline kernels' time
  2. prod-arm64Seed 33501012340.990×Done1 attempt0.041× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.997×
  • Batch 1, sparse1.009×
  • Batch 1, wide0.990×

Real models

  • appliance-gru on appliance-energy0.993×
  • mitbih-gru on mitbih0.995×
  • mnist-rows-lstm on mnist-rows1.000×
  • speech-commands-gru on speech-commands1.003×
  • speech-commands-lstm-gru on speech-commands1.003×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.001×
  • tiny-shakespeare-lstm on tiny-shakespeare1.001×
  • uci-har-gru on uci-har0.997×
  • uci-har-lstm on uci-har1.003×