Conjectures.io
Tensor kernels

Entry #32

Completed
Hotkey
5Cop2t…b4axQb
Sent
11 Nov 2026, 04:44 UTC
Source
e8a645094f94

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 #39Done2 of 2 runs1.007×Generated 0.994×, real models 1.019×Above the barGain lower bound -4.1%Guard passed
MLPvs #1Done2 of 2 runs0.887×Generated 0.934×, real models 0.843×Not significantGain lower bound -11.7%Guard passed
ResNetvs #32Done2 of 2 runs0.804×Generated 0.709×, real models 0.912×QualifiesGain lower bound 4.9%Guard passed
MobileNetvs #37Done2 of 2 runs0.915×Generated 0.775×, real models 1.080×Not significantGain lower bound -4.8%Guard passed
Normalized MLPvs #6Done2 of 2 runs0.982×Generated 1.005×, real models 0.960×Above the barGain lower bound -18.0%Guard passed
Mixervs #24Done2 of 2 runs1.100×Generated 0.949×, real models 1.276×Above the barGain lower bound -35.2%Guard passed
Recurrentvs #38Done2 of 2 runs1.197×Generated 1.242×, real models 1.153×Above the barGain lower bound -23.7%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 0.994× · real models 1.019×

MachineTime
  1. prod-amd64Seed 4337553710.968×Done1 attempt0.010× of the baseline kernels' time
  2. prod-arm64Seed 4337553711.046×Done1 attempt0.013× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.059×
  • Batch 1, sparse0.969×
  • Batch 1, wide0.962×

Mixer

  • Batch 1, normal0.935×
  • Batch 1, sparse1.039×
  • Batch 1, wide1.013×

ResNet

  • Batch 1, normal0.942×
  • Batch 1, sparse0.906×
  • Batch 1, wide0.894×

MobileNet

  • Batch 1, normal0.926×
  • Batch 1, sparse0.889×
  • Batch 1, wide0.930×

Recurrent

  • Batch 1, normal1.207×
  • Batch 1, sparse1.212×
  • Batch 1, wide1.213×

Normalized MLP

  • Batch 1, normal1.074×
  • Batch 1, sparse0.941×
  • Batch 1, wide0.971×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.054×
  • compact-mixer on digits0.858×
  • compact-mixer on fashion0.976×
  • compact-mixer on mnist1.080×
  • compact-mlp on digits1.069×
  • compact-mlp on fashion1.009×
  • compact-mlp on mnist0.972×
  • compact-mobilenet on digits1.313×
  • compact-mobilenet on fashion1.223×
  • compact-mobilenet on mnist1.117×
  • compact-normalized_mlp on digits1.105×
  • compact-normalized_mlp on fashion0.975×
  • compact-normalized_mlp on mnist0.861×
  • compact-resnet on digits1.002×
  • compact-resnet on fashion0.953×
  • compact-resnet on mnist0.930×
  • appliance-gru on appliance-energy1.004×
  • mitbih-gru on mitbih0.956×
  • mnist-rows-lstm on mnist-rows1.331×
  • speech-commands-gru on speech-commands0.993×
  • speech-commands-lstm-gru on speech-commands1.156×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.312×
  • tiny-shakespeare-lstm on tiny-shakespeare1.406×
  • uci-har-gru on uci-har1.000×
  • uci-har-lstm on uci-har1.243×
  • uci-har-mixer on UCI-HAR0.948×
  • uci-har-mlp on UCI-HAR1.017×
  • uci-har-mobilenet on UCI-HAR0.842×
  • uci-har-normalized_mlp on UCI-HAR0.980×
  • uci-har-resnet on UCI-HAR0.761×
MLP2 runs

RunsGenerated 0.934× · real models 0.843×

MachineTime
  1. prod-amd64Seed 4337553710.828×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 4337553710.951×Done1 attempt0.024× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.928×
  • Batch 1, sparse0.885×
  • Batch 1, wide0.951×

Real models

  • compact-mlp on digits0.938×
  • compact-mlp on fashion0.772×
  • compact-mlp on mnist0.782×
  • uci-har-mlp on UCI-HAR0.890×
ResNet2 runs

RunsGenerated 0.709× · real models 0.912×

MachineTime
  1. prod-amd64Seed 4337553710.896×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 4337553710.721×Done1 attempt0.007× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal0.614×
  • Batch 1, sparse0.905×
  • Batch 1, wide0.872×

Real models

  • compact-resnet on digits1.133×
  • compact-resnet on fashion0.966×
  • compact-resnet on mnist0.881×
  • uci-har-resnet on UCI-HAR0.717×
MobileNet2 runs

RunsGenerated 0.775× · real models 1.080×

MachineTime
  1. prod-amd64Seed 4337553711.008×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 4337553710.831×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal0.678×
  • Batch 1, sparse0.866×
  • Batch 1, wide0.916×

Real models

  • compact-mobilenet on digits1.274×
  • compact-mobilenet on fashion1.187×
  • compact-mobilenet on mnist1.109×
  • uci-har-mobilenet on UCI-HAR0.810×
Normalized MLP2 runs

RunsGenerated 1.005× · real models 0.960×

MachineTime
  1. prod-amd64Seed 4337553710.988×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 4337553710.977×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.010×
  • Batch 1, sparse0.984×
  • Batch 1, wide1.040×

Real models

  • compact-normalized_mlp on digits0.977×
  • compact-normalized_mlp on fashion0.985×
  • compact-normalized_mlp on mnist0.904×
  • uci-har-normalized_mlp on UCI-HAR0.979×
Mixer2 runs

RunsGenerated 0.949× · real models 1.276×

MachineTime
  1. prod-amd64Seed 4337553711.101×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 4337553711.099×Done1 attempt0.006× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.923×
  • Batch 1, sparse1.028×
  • Batch 1, wide0.930×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.157×
  • compact-mixer on digits3.073×
  • compact-mixer on fashion0.967×
  • compact-mixer on mnist1.013×
  • uci-har-mixer on UCI-HAR0.970×
Recurrent2 runs

RunsGenerated 1.242× · real models 1.153×

MachineTime
  1. prod-amd64Seed 4337553711.094×Done1 attempt0.020× of the baseline kernels' time
  2. prod-arm64Seed 4337553711.309×Done1 attempt0.024× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal1.217×
  • Batch 1, sparse1.241×
  • Batch 1, wide1.232×

Real models

  • appliance-gru on appliance-energy0.988×
  • mitbih-gru on mitbih0.966×
  • mnist-rows-lstm on mnist-rows1.331×
  • speech-commands-gru on speech-commands0.985×
  • speech-commands-lstm-gru on speech-commands1.202×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.329×
  • tiny-shakespeare-lstm on tiny-shakespeare1.440×
  • uci-har-gru on uci-har0.996×
  • uci-har-lstm on uci-har1.256×