Conjectures.io
Tensor kernels

Entry #101

Completed
Hotkey
5H9ZAV…4Sf7ix
Sent
12 Nov 2026, 10:31 UTC
Source
4973141b2fe5

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 runs0.989×Generated 0.972×, real models 1.007×Above the barGain lower bound -3.3%Guard passed
MLPvs #44Done2 of 2 runs0.915×Generated 0.981×, real models 0.853×Not significantGain lower bound -11.9%Guard passed
ResNetvs #51Done2 of 2 runs1.122×Generated 1.286×, real models 0.979×Above the barGain lower bound -74.6%Guard passed
MobileNetvs #53Done2 of 2 runs1.142×Generated 1.132×, real models 1.152×Above the barGain lower bound -29.4%Guard passed
Normalized MLPvs #48Done2 of 2 runs0.994×Generated 0.993×, real models 0.995×Above the barGain lower bound -17.8%Guard passed
Mixervs #47Done2 of 2 runs0.950×Generated 1.042×, real models 0.866×Not significantGain lower bound -6.0%Guard passed
Recurrentvs #55Done2 of 2 runs0.951×Generated 0.949×, real models 0.952×QualifiesGain lower bound 1.4%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 0.972× · real models 1.007×

MachineTime
  1. prod-amd64Seed 4653775580.983×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 4653775580.995×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.927×
  • Batch 1, sparse0.979×
  • Batch 1, wide1.042×

Mixer

  • Batch 1, normal0.990×
  • Batch 1, sparse1.041×
  • Batch 1, wide1.031×

ResNet

  • Batch 1, normal0.974×
  • Batch 1, sparse0.991×
  • Batch 1, wide0.990×

MobileNet

  • Batch 1, normal0.962×
  • Batch 1, sparse0.993×
  • Batch 1, wide0.989×

Recurrent

  • Batch 1, normal0.875×
  • Batch 1, sparse0.843×
  • Batch 1, wide0.862×

Normalized MLP

  • Batch 1, normal1.039×
  • Batch 1, sparse0.977×
  • Batch 1, wide0.975×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.991×
  • compact-mixer on digits1.207×
  • compact-mixer on fashion1.070×
  • compact-mixer on mnist1.060×
  • compact-mlp on digits1.199×
  • compact-mlp on fashion1.573×
  • compact-mlp on mnist1.514×
  • compact-mobilenet on digits0.950×
  • compact-mobilenet on fashion1.061×
  • compact-mobilenet on mnist1.025×
  • compact-normalized_mlp on digits1.060×
  • compact-normalized_mlp on fashion0.970×
  • compact-normalized_mlp on mnist0.464×
  • compact-resnet on digits1.100×
  • compact-resnet on fashion0.965×
  • compact-resnet on mnist0.963×
  • appliance-gru on appliance-energy0.926×
  • mitbih-gru on mitbih0.929×
  • mnist-rows-lstm on mnist-rows0.569×
  • speech-commands-gru on speech-commands1.004×
  • speech-commands-lstm-gru on speech-commands0.942×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.894×
  • tiny-shakespeare-lstm on tiny-shakespeare0.698×
  • uci-har-gru on uci-har0.963×
  • uci-har-lstm on uci-har0.899×
  • uci-har-mixer on UCI-HAR1.117×
  • uci-har-mlp on UCI-HAR1.034×
  • uci-har-mobilenet on UCI-HAR1.080×
  • uci-har-normalized_mlp on UCI-HAR1.093×
  • uci-har-resnet on UCI-HAR0.887×
MLP2 runs

RunsGenerated 0.981× · real models 0.853×

MachineTime
  1. prod-amd64Seed 4653775580.947×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 4653775580.884×Done1 attempt0.016× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.002×
  • Batch 1, sparse0.975×
  • Batch 1, wide0.886×

Real models

  • compact-mlp on digits0.659×
  • compact-mlp on fashion0.832×
  • compact-mlp on mnist0.933×
  • uci-har-mlp on UCI-HAR1.035×
ResNet2 runs

RunsGenerated 1.286× · real models 0.979×

MachineTime
  1. prod-amd64Seed 4653775581.003×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 4653775581.254×Done1 attempt0.009× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal1.451×
  • Batch 1, sparse0.976×
  • Batch 1, wide1.174×

Real models

  • compact-resnet on digits0.965×
  • compact-resnet on fashion0.971×
  • compact-resnet on mnist0.970×
  • uci-har-resnet on UCI-HAR1.010×
MobileNet2 runs

RunsGenerated 1.132× · real models 1.152×

MachineTime
  1. prod-amd64Seed 4653775581.092×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 4653775581.194×Done1 attempt0.013× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.395×
  • Batch 1, sparse0.985×
  • Batch 1, wide0.988×

Real models

  • compact-mobilenet on digits1.111×
  • compact-mobilenet on fashion1.222×
  • compact-mobilenet on mnist1.233×
  • uci-har-mobilenet on UCI-HAR1.051×
Normalized MLP2 runs

RunsGenerated 0.993× · real models 0.995×

MachineTime
  1. prod-amd64Seed 4653775581.004×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 4653775580.985×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.002×
  • Batch 1, sparse1.031×
  • Batch 1, wide0.991×

Real models

  • compact-normalized_mlp on digits1.003×
  • compact-normalized_mlp on fashion1.027×
  • compact-normalized_mlp on mnist1.065×
  • uci-har-normalized_mlp on UCI-HAR0.895×
Mixer2 runs

RunsGenerated 1.042× · real models 0.866×

MachineTime
  1. prod-amd64Seed 4653775580.900×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 4653775581.002×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal1.097×
  • Batch 1, sparse1.026×
  • Batch 1, wide1.001×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.126×
  • compact-mixer on digits0.442×
  • compact-mixer on fashion1.037×
  • compact-mixer on mnist1.004×
  • uci-har-mixer on UCI-HAR0.940×
Recurrent2 runs

RunsGenerated 0.949× · real models 0.952×

MachineTime
  1. prod-amd64Seed 4653775580.965×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 4653775580.937×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.981×
  • Batch 1, sparse0.914×
  • Batch 1, wide0.923×

Real models

  • appliance-gru on appliance-energy1.015×
  • mitbih-gru on mitbih1.016×
  • mnist-rows-lstm on mnist-rows0.612×
  • speech-commands-gru on speech-commands1.104×
  • speech-commands-lstm-gru on speech-commands1.076×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.039×
  • tiny-shakespeare-lstm on tiny-shakespeare0.755×
  • uci-har-gru on uci-har1.046×
  • uci-har-lstm on uci-har1.046×