Conjectures.io
Tensor kernels

Entry #11

Completed
Hotkey
5HeLNg…828pfH
Sent
10 Nov 2026, 20:18 UTC
Source
89cf57b8a595

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 #17Done2 of 2 runs0.906×Generated 0.924×, real models 0.889×QualifiesGain lower bound 6.5%Guard passed
MLPvs #1Done2 of 2 runs0.982×Generated 1.013×, real models 0.953×Above the barGain lower bound -23.6%Guard passed
ResNetvs #15Done2 of 2 runs1.410×Generated 1.466×, real models 1.357×Above the barGain lower bound -44.1%Guard passed
MobileNetvs #16Done2 of 2 runs1.711×Generated 1.855×, real models 1.577×Above the barGain lower bound -76.2%Guard passed
Normalized MLPvs #6Done2 of 2 runs1.108×Generated 1.030×, real models 1.191×Above the barGain lower bound -29.5%Guard passed
Mixervs #7Done2 of 2 runs1.094×Generated 1.187×, real models 1.008×Above the barGain lower bound -25.7%Guard passed
Recurrentvs #18Done2 of 2 runs1.100×Generated 1.078×, real models 1.122×Above the barGain lower bound -12.4%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 0.924× · real models 0.889×

MachineTime
  1. prod-amd64Seed 30556772330.912×Done1 attempt0.054× of the baseline kernels' time
  2. prod-arm64Seed 30556772330.901×Done1 attempt0.059× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.084×
  • Batch 1, sparse1.089×
  • Batch 1, wide1.016×

Mixer

  • Batch 1, normal0.990×
  • Batch 1, sparse1.005×
  • Batch 1, wide0.992×

ResNet

  • Batch 1, normal0.722×
  • Batch 1, sparse0.722×
  • Batch 1, wide0.721×

MobileNet

  • Batch 1, normal0.862×
  • Batch 1, sparse0.851×
  • Batch 1, wide0.860×

Recurrent

  • Batch 1, normal0.937×
  • Batch 1, sparse0.936×
  • Batch 1, wide0.935×

Normalized MLP

  • Batch 1, normal1.033×
  • Batch 1, sparse1.083×
  • Batch 1, wide0.981×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.090×
  • compact-mixer on digits1.192×
  • compact-mixer on fashion1.012×
  • compact-mixer on mnist0.992×
  • compact-mlp on digits1.008×
  • compact-mlp on fashion1.045×
  • compact-mlp on mnist1.138×
  • compact-mobilenet on digits0.864×
  • compact-mobilenet on fashion0.798×
  • compact-mobilenet on mnist0.772×
  • compact-normalized_mlp on digits0.965×
  • compact-normalized_mlp on fashion0.899×
  • compact-normalized_mlp on mnist0.926×
  • compact-resnet on digits0.719×
  • compact-resnet on fashion0.710×
  • compact-resnet on mnist0.727×
  • appliance-gru on appliance-energy0.951×
  • mitbih-gru on mitbih0.947×
  • mnist-rows-lstm on mnist-rows0.920×
  • speech-commands-gru on speech-commands0.959×
  • speech-commands-lstm-gru on speech-commands0.938×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.853×
  • tiny-shakespeare-lstm on tiny-shakespeare0.919×
  • uci-har-gru on uci-har0.951×
  • uci-har-lstm on uci-har0.868×
  • uci-har-mixer on UCI-HAR1.016×
  • uci-har-mlp on UCI-HAR1.027×
  • uci-har-mobilenet on UCI-HAR0.799×
  • uci-har-normalized_mlp on UCI-HAR0.850×
  • uci-har-resnet on UCI-HAR0.495×
MLP2 runs

RunsGenerated 1.013× · real models 0.953×

MachineTime
  1. prod-amd64Seed 30556772330.962×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 30556772331.003×Done1 attempt0.017× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.021×
  • Batch 1, sparse1.003×
  • Batch 1, wide1.003×

Real models

  • compact-mlp on digits0.882×
  • compact-mlp on fashion0.993×
  • compact-mlp on mnist0.927×
  • uci-har-mlp on UCI-HAR1.016×
ResNet2 runs

RunsGenerated 1.466× · real models 1.357×

MachineTime
  1. prod-amd64Seed 30556772331.390×Done1 attempt0.200× of the baseline kernels' time
  2. prod-arm64Seed 30556772331.430×Done1 attempt0.220× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal1.464×
  • Batch 1, sparse1.468×
  • Batch 1, wide1.463×

Real models

  • compact-resnet on digits1.439×
  • compact-resnet on fashion1.369×
  • compact-resnet on mnist1.246×
  • uci-har-resnet on UCI-HAR1.380×
MobileNet2 runs

RunsGenerated 1.855× · real models 1.577×

MachineTime
  1. prod-amd64Seed 30556772331.774×Done1 attempt0.443× of the baseline kernels' time
  2. prod-arm64Seed 30556772331.649×Done1 attempt0.439× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.859×
  • Batch 1, sparse1.846×
  • Batch 1, wide1.865×

Real models

  • compact-mobilenet on digits1.755×
  • compact-mobilenet on fashion1.443×
  • compact-mobilenet on mnist1.381×
  • uci-har-mobilenet on UCI-HAR1.769×
Normalized MLP2 runs

RunsGenerated 1.030× · real models 1.191×

MachineTime
  1. prod-amd64Seed 30556772330.985×Done1 attempt0.010× of the baseline kernels' time
  2. prod-arm64Seed 30556772331.245×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.010×
  • Batch 1, sparse1.052×
  • Batch 1, wide0.995×

Real models

  • compact-normalized_mlp on digits1.986×
  • compact-normalized_mlp on fashion1.082×
  • compact-normalized_mlp on mnist1.010×
  • uci-har-normalized_mlp on UCI-HAR0.929×
Mixer2 runs

RunsGenerated 1.187× · real models 1.008×

MachineTime
  1. prod-amd64Seed 30556772331.159×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 30556772331.032×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal1.006×
  • Batch 1, sparse1.127×
  • Batch 1, wide1.330×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.029×
  • compact-mixer on digits0.898×
  • compact-mixer on fashion1.017×
  • compact-mixer on mnist1.014×
  • uci-har-mixer on UCI-HAR1.094×
Recurrent2 runs

RunsGenerated 1.078× · real models 1.122×

MachineTime
  1. prod-amd64Seed 30556772331.103×Done1 attempt0.365× of the baseline kernels' time
  2. prod-arm64Seed 30556772331.096×Done1 attempt0.361× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal1.084×
  • Batch 1, sparse1.069×
  • Batch 1, wide1.085×

Real models

  • appliance-gru on appliance-energy1.269×
  • mitbih-gru on mitbih1.371×
  • mnist-rows-lstm on mnist-rows1.036×
  • speech-commands-gru on speech-commands1.038×
  • speech-commands-lstm-gru on speech-commands1.056×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.042×
  • tiny-shakespeare-lstm on tiny-shakespeare1.042×
  • uci-har-gru on uci-har1.262×
  • uci-har-lstm on uci-har1.039×