Conjectures.io
Tensor kernels

Entry #41

Completed
Hotkey
5FpXYh…nDBBxg
Sent
11 Nov 2026, 09:05 UTC
Source
6c0743342ffd

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 #43Done2 of 2 runs1.006×Generated 1.037×, real models 0.976×Above the barGain lower bound -5.9%Guard passed
MLPvs #44Done2 of 2 runs0.962×Generated 0.992×, real models 0.934×Not significantGain lower bound -8.4%Guard passed
ResNetvs #40Done2 of 2 runs0.925×Generated 0.932×, real models 0.919×QualifiesGain lower bound 1.2%Guard passed
MobileNetvs #42Done2 of 2 runs1.077×Generated 1.210×, real models 0.959×Above the barGain lower bound -27.3%Guard passed
Normalized MLPvs #6Done2 of 2 runs0.964×Generated 0.984×, real models 0.944×Not significantGain lower bound -11.2%Guard passed
Mixervs #24Done2 of 2 runs1.031×Generated 0.967×, real models 1.098×Above the barGain lower bound -24.4%Guard passed
Recurrentvs #41Done2 of 2 runs0.952×Generated 0.984×, real models 0.920×QualifiesGain lower bound 1.3%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 1.037× · real models 0.976×

MachineTime
  1. prod-amd64Seed 41596942661.041×Done1 attempt0.010× of the baseline kernels' time
  2. prod-arm64Seed 41596942660.973×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.961×
  • Batch 1, sparse1.047×
  • Batch 1, wide1.035×

Mixer

  • Batch 1, normal1.043×
  • Batch 1, sparse1.005×
  • Batch 1, wide1.036×

ResNet

  • Batch 1, normal1.032×
  • Batch 1, sparse1.008×
  • Batch 1, wide1.031×

MobileNet

  • Batch 1, normal1.079×
  • Batch 1, sparse1.046×
  • Batch 1, wide1.040×

Recurrent

  • Batch 1, normal1.037×
  • Batch 1, sparse0.992×
  • Batch 1, wide1.003×

Normalized MLP

  • Batch 1, normal1.061×
  • Batch 1, sparse1.022×
  • Batch 1, wide1.066×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.940×
  • compact-mixer on digits0.812×
  • compact-mixer on fashion1.005×
  • compact-mixer on mnist0.995×
  • compact-mlp on digits0.779×
  • compact-mlp on fashion1.148×
  • compact-mlp on mnist1.051×
  • compact-mobilenet on digits1.034×
  • compact-mobilenet on fashion0.967×
  • compact-mobilenet on mnist0.983×
  • compact-normalized_mlp on digits0.839×
  • compact-normalized_mlp on fashion1.081×
  • compact-normalized_mlp on mnist1.164×
  • compact-resnet on digits1.082×
  • compact-resnet on fashion1.009×
  • compact-resnet on mnist0.982×
  • appliance-gru on appliance-energy0.906×
  • mitbih-gru on mitbih0.930×
  • mnist-rows-lstm on mnist-rows0.947×
  • speech-commands-gru on speech-commands0.930×
  • speech-commands-lstm-gru on speech-commands0.931×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.937×
  • tiny-shakespeare-lstm on tiny-shakespeare0.947×
  • uci-har-gru on uci-har0.925×
  • uci-har-lstm on uci-har0.937×
  • uci-har-mixer on UCI-HAR0.954×
  • uci-har-mlp on UCI-HAR0.993×
  • uci-har-mobilenet on UCI-HAR0.944×
  • uci-har-normalized_mlp on UCI-HAR1.021×
  • uci-har-resnet on UCI-HAR0.962×
MLP2 runs

RunsGenerated 0.992× · real models 0.934×

MachineTime
  1. prod-amd64Seed 41596942660.963×Done1 attempt0.017× of the baseline kernels' time
  2. prod-arm64Seed 41596942660.962×Done1 attempt0.016× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.013×
  • Batch 1, sparse0.966×
  • Batch 1, wide1.065×

Real models

  • compact-mlp on digits0.770×
  • compact-mlp on fashion1.011×
  • compact-mlp on mnist0.981×
  • uci-har-mlp on UCI-HAR0.997×
ResNet2 runs

RunsGenerated 0.932× · real models 0.919×

MachineTime
  1. prod-amd64Seed 41596942660.950×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 41596942660.902×Done1 attempt0.009× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal0.942×
  • Batch 1, sparse0.975×
  • Batch 1, wide0.978×

Real models

  • compact-resnet on digits0.773×
  • compact-resnet on fashion1.005×
  • compact-resnet on mnist0.996×
  • uci-har-resnet on UCI-HAR0.922×
MobileNet2 runs

RunsGenerated 1.210× · real models 0.959×

MachineTime
  1. prod-amd64Seed 41596942660.984×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 41596942661.179×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.204×
  • Batch 1, sparse1.025×
  • Batch 1, wide1.043×

Real models

  • compact-mobilenet on digits0.954×
  • compact-mobilenet on fashion0.951×
  • compact-mobilenet on mnist0.974×
  • uci-har-mobilenet on UCI-HAR0.959×
Normalized MLP2 runs

RunsGenerated 0.984× · real models 0.944×

MachineTime
  1. prod-amd64Seed 41596942660.950×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 41596942660.978×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.147×
  • Batch 1, sparse0.968×
  • Batch 1, wide0.941×

Real models

  • compact-normalized_mlp on digits0.927×
  • compact-normalized_mlp on fashion1.010×
  • compact-normalized_mlp on mnist0.867×
  • uci-har-normalized_mlp on UCI-HAR0.979×
Mixer2 runs

RunsGenerated 0.967× · real models 1.098×

MachineTime
  1. prod-amd64Seed 41596942660.947×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 41596942661.121×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.983×
  • Batch 1, sparse0.942×
  • Batch 1, wide1.001×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.993×
  • compact-mixer on digits1.865×
  • compact-mixer on fashion0.917×
  • compact-mixer on mnist0.967×
  • uci-har-mixer on UCI-HAR0.971×
Recurrent2 runs

RunsGenerated 0.984× · real models 0.920×

MachineTime
  1. prod-amd64Seed 41596942660.952×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 41596942660.951×Done1 attempt0.014× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.965×
  • Batch 1, sparse1.000×
  • Batch 1, wide0.991×

Real models

  • appliance-gru on appliance-energy0.939×
  • mitbih-gru on mitbih0.937×
  • mnist-rows-lstm on mnist-rows0.808×
  • speech-commands-gru on speech-commands0.927×
  • speech-commands-lstm-gru on speech-commands0.935×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.947×
  • tiny-shakespeare-lstm on tiny-shakespeare0.930×
  • uci-har-gru on uci-har0.926×
  • uci-har-lstm on uci-har0.937×