Conjectures.io
Tensor kernels

Entry #74

Completed
Hotkey
5FR9Z5…Spuq4M
Sent
12 Nov 2026, 00:33 UTC
Source
796b9bcb075f

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 runs0.994×Generated 1.020×, real models 0.969×Above the barGain lower bound -4.5%Guard passed
MLPvs #44Done2 of 2 runs0.982×Generated 1.085×, real models 0.889×Above the barGain lower bound -19.1%Guard passed
ResNetvs #51Done2 of 2 runs1.031×Generated 1.022×, real models 1.041×Above the barGain lower bound -21.1%Guard passed
MobileNetvs #53Done2 of 2 runs0.985×Generated 0.950×, real models 1.021×Above the barGain lower bound -5.2%Guard passed
Normalized MLPvs #48Done2 of 2 runs0.977×Generated 0.922×, real models 1.036×Above the barGain lower bound -25.5%Guard passed
Mixervs #47Done2 of 2 runs1.069×Generated 1.094×, real models 1.044×Above the barGain lower bound -17.7%Guard passed
Recurrentvs #52Done2 of 2 runs0.938×Generated 0.953×, real models 0.924×QualifiesGain lower bound 3.1%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 1.020× · real models 0.969×

MachineTime
  1. prod-amd64Seed 5303045701.013×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 5303045700.977×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.979×
  • Batch 1, sparse1.064×
  • Batch 1, wide1.131×

Mixer

  • Batch 1, normal1.039×
  • Batch 1, sparse1.065×
  • Batch 1, wide1.049×

ResNet

  • Batch 1, normal1.052×
  • Batch 1, sparse1.023×
  • Batch 1, wide1.064×

MobileNet

  • Batch 1, normal1.049×
  • Batch 1, sparse1.043×
  • Batch 1, wide1.033×

Recurrent

  • Batch 1, normal0.960×
  • Batch 1, sparse0.925×
  • Batch 1, wide0.937×

Normalized MLP

  • Batch 1, normal0.823×
  • Batch 1, sparse0.933×
  • Batch 1, wide0.823×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.587×
  • compact-mixer on digits0.824×
  • compact-mixer on fashion1.032×
  • compact-mixer on mnist1.005×
  • compact-mlp on digits1.549×
  • compact-mlp on fashion0.683×
  • compact-mlp on mnist0.474×
  • compact-mobilenet on digits0.907×
  • compact-mobilenet on fashion0.891×
  • compact-mobilenet on mnist0.854×
  • compact-normalized_mlp on digits1.240×
  • compact-normalized_mlp on fashion1.074×
  • compact-normalized_mlp on mnist1.136×
  • compact-resnet on digits0.962×
  • compact-resnet on fashion1.005×
  • compact-resnet on mnist0.995×
  • appliance-gru on appliance-energy0.956×
  • mitbih-gru on mitbih1.013×
  • mnist-rows-lstm on mnist-rows0.692×
  • speech-commands-gru on speech-commands1.041×
  • speech-commands-lstm-gru on speech-commands0.969×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.921×
  • tiny-shakespeare-lstm on tiny-shakespeare0.684×
  • uci-har-gru on uci-har1.015×
  • uci-har-lstm on uci-har0.896×
  • uci-har-mixer on UCI-HAR1.109×
  • uci-har-mlp on UCI-HAR1.049×
  • uci-har-mobilenet on UCI-HAR0.935×
  • uci-har-normalized_mlp on UCI-HAR1.018×
  • uci-har-resnet on UCI-HAR1.037×
MLP2 runs

RunsGenerated 1.085× · real models 0.889×

MachineTime
  1. prod-amd64Seed 5303045701.101×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 5303045700.876×Done1 attempt0.019× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.086×
  • Batch 1, sparse1.183×
  • Batch 1, wide0.920×

Real models

  • compact-mlp on digits0.644×
  • compact-mlp on fashion1.018×
  • compact-mlp on mnist0.988×
  • uci-har-mlp on UCI-HAR0.963×
ResNet2 runs

RunsGenerated 1.022× · real models 1.041×

MachineTime
  1. prod-amd64Seed 5303045700.982×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 5303045701.084×Done1 attempt0.009× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal1.063×
  • Batch 1, sparse0.993×
  • Batch 1, wide1.065×

Real models

  • compact-resnet on digits1.150×
  • compact-resnet on fashion1.044×
  • compact-resnet on mnist0.997×
  • uci-har-resnet on UCI-HAR0.982×
MobileNet2 runs

RunsGenerated 0.950× · real models 1.021×

MachineTime
  1. prod-amd64Seed 5303045701.037×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 5303045700.936×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.003×
  • Batch 1, sparse1.019×
  • Batch 1, wide1.003×

Real models

  • compact-mobilenet on digits1.036×
  • compact-mobilenet on fashion1.017×
  • compact-mobilenet on mnist1.011×
  • uci-har-mobilenet on UCI-HAR1.021×
Normalized MLP2 runs

RunsGenerated 0.922× · real models 1.036×

MachineTime
  1. prod-amd64Seed 5303045700.951×Done1 attempt0.007× of the baseline kernels' time
  2. prod-arm64Seed 5303045701.004×Done1 attempt0.015× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal0.841×
  • Batch 1, sparse0.932×
  • Batch 1, wide0.824×

Real models

  • compact-normalized_mlp on digits0.964×
  • compact-normalized_mlp on fashion0.997×
  • compact-normalized_mlp on mnist1.051×
  • uci-har-normalized_mlp on UCI-HAR1.141×
Mixer2 runs

RunsGenerated 1.094× · real models 1.044×

MachineTime
  1. prod-amd64Seed 5303045701.078×Done1 attempt0.004× of the baseline kernels' time
  2. prod-arm64Seed 5303045701.061×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal1.089×
  • Batch 1, sparse1.241×
  • Batch 1, wide1.105×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.385×
  • compact-mixer on digits0.849×
  • compact-mixer on fashion1.110×
  • compact-mixer on mnist1.043×
  • uci-har-mixer on UCI-HAR0.913×
Recurrent2 runs

RunsGenerated 0.953× · real models 0.924×

MachineTime
  1. prod-amd64Seed 5303045700.918×Done1 attempt0.010× of the baseline kernels' time
  2. prod-arm64Seed 5303045700.959×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.956×
  • Batch 1, sparse0.943×
  • Batch 1, wide0.947×

Real models

  • appliance-gru on appliance-energy1.041×
  • mitbih-gru on mitbih1.023×
  • mnist-rows-lstm on mnist-rows0.712×
  • speech-commands-gru on speech-commands1.054×
  • speech-commands-lstm-gru on speech-commands0.999×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.950×
  • tiny-shakespeare-lstm on tiny-shakespeare0.683×
  • uci-har-gru on uci-har1.019×
  • uci-har-lstm on uci-har0.930×