Conjectures.io
Tensor kernels

Entry #34

Completed
Hotkey
5F3yDg…RTpK4C
Sent
11 Nov 2026, 06:31 UTC
Source
c1cc1450be58

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 runs0.974×Generated 0.951×, real models 0.998×Above the barGain lower bound -3.7%Guard passed
MLPvs #1Done2 of 2 runs0.925×Generated 0.895×, real models 0.957×Not significantGain lower bound -18.2%Guard passed
ResNetvs #40Done2 of 2 runs0.989×Generated 1.013×, real models 0.966×Above the barGain lower bound -9.5%Guard passed
MobileNetvs #37Done2 of 2 runs0.980×Generated 0.887×, real models 1.083×Above the barGain lower bound -5.0%Guard passed
Normalized MLPvs #6Done2 of 2 runs0.930×Generated 0.936×, real models 0.924×Not significantGain lower bound -3.7%Guard passed
Mixervs #24Done2 of 2 runs1.075×Generated 1.019×, real models 1.134×Above the barGain lower bound -26.2%Guard passed
Recurrentvs #38Done2 of 2 runs0.798×Generated 0.787×, real models 0.809×QualifiesGain lower bound 18.0%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 0.951× · real models 0.998×

MachineTime
  1. prod-amd64Seed 38151098280.971×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 38151098280.978×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.059×
  • Batch 1, sparse0.910×
  • Batch 1, wide1.001×

Mixer

  • Batch 1, normal1.124×
  • Batch 1, sparse1.136×
  • Batch 1, wide1.210×

ResNet

  • Batch 1, normal0.933×
  • Batch 1, sparse0.905×
  • Batch 1, wide0.923×

MobileNet

  • Batch 1, normal0.905×
  • Batch 1, sparse0.919×
  • Batch 1, wide0.923×

Recurrent

  • Batch 1, normal0.786×
  • Batch 1, sparse0.775×
  • Batch 1, wide0.799×

Normalized MLP

  • Batch 1, normal1.098×
  • Batch 1, sparse0.959×
  • Batch 1, wide0.993×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.151×
  • compact-mixer on digits1.335×
  • compact-mixer on fashion1.042×
  • compact-mixer on mnist1.046×
  • compact-mlp on digits0.583×
  • compact-mlp on fashion0.983×
  • compact-mlp on mnist1.883×
  • compact-mobilenet on digits1.412×
  • compact-mobilenet on fashion1.181×
  • compact-mobilenet on mnist1.145×
  • compact-normalized_mlp on digits0.904×
  • compact-normalized_mlp on fashion1.173×
  • compact-normalized_mlp on mnist1.036×
  • compact-resnet on digits1.233×
  • compact-resnet on fashion0.935×
  • compact-resnet on mnist0.949×
  • appliance-gru on appliance-energy0.744×
  • mitbih-gru on mitbih0.779×
  • mnist-rows-lstm on mnist-rows0.941×
  • speech-commands-gru on speech-commands0.692×
  • speech-commands-lstm-gru on speech-commands0.788×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.886×
  • tiny-shakespeare-lstm on tiny-shakespeare0.880×
  • uci-har-gru on uci-har0.738×
  • uci-har-lstm on uci-har0.878×
  • uci-har-mixer on UCI-HAR0.966×
  • uci-har-mlp on UCI-HAR0.987×
  • uci-har-mobilenet on UCI-HAR0.773×
  • uci-har-normalized_mlp on UCI-HAR1.088×
  • uci-har-resnet on UCI-HAR0.727×
MLP2 runs

RunsGenerated 0.895× · real models 0.957×

MachineTime
  1. prod-amd64Seed 38151098280.828×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 38151098281.035×Done1 attempt0.022× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.971×
  • Batch 1, sparse0.914×
  • Batch 1, wide0.956×

Real models

  • compact-mlp on digits1.445×
  • compact-mlp on fashion0.884×
  • compact-mlp on mnist0.780×
  • uci-har-mlp on UCI-HAR0.841×
ResNet2 runs

RunsGenerated 1.013× · real models 0.966×

MachineTime
  1. prod-amd64Seed 38151098280.998×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 38151098280.980×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal1.019×
  • Batch 1, sparse0.997×
  • Batch 1, wide1.007×

Real models

  • compact-resnet on digits0.842×
  • compact-resnet on fashion1.015×
  • compact-resnet on mnist1.005×
  • uci-har-resnet on UCI-HAR1.012×
MobileNet2 runs

RunsGenerated 0.887× · real models 1.083×

MachineTime
  1. prod-amd64Seed 38151098280.958×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 38151098281.003×Done1 attempt0.013× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal0.894×
  • Batch 1, sparse0.898×
  • Batch 1, wide0.933×

Real models

  • compact-mobilenet on digits1.285×
  • compact-mobilenet on fashion1.189×
  • compact-mobilenet on mnist1.124×
  • uci-har-mobilenet on UCI-HAR0.800×
Normalized MLP2 runs

RunsGenerated 0.936× · real models 0.924×

MachineTime
  1. prod-amd64Seed 38151098280.975×Done1 attempt0.007× of the baseline kernels' time
  2. prod-arm64Seed 38151098280.888×Done1 attempt0.013× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal0.964×
  • Batch 1, sparse0.933×
  • Batch 1, wide0.924×

Real models

  • compact-normalized_mlp on digits1.223×
  • compact-normalized_mlp on fashion0.721×
  • compact-normalized_mlp on mnist0.955×
  • uci-har-normalized_mlp on UCI-HAR0.866×
Mixer2 runs

RunsGenerated 1.019× · real models 1.134×

MachineTime
  1. prod-amd64Seed 38151098281.043×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 38151098281.107×Done1 attempt0.006× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal1.033×
  • Batch 1, sparse0.965×
  • Batch 1, wide1.014×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.061×
  • compact-mixer on digits1.880×
  • compact-mixer on fashion0.900×
  • compact-mixer on mnist1.019×
  • uci-har-mixer on UCI-HAR1.025×
Recurrent2 runs

RunsGenerated 0.787× · real models 0.809×

MachineTime
  1. prod-amd64Seed 38151098280.784×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 38151098280.813×Done1 attempt0.014× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.782×
  • Batch 1, sparse0.789×
  • Batch 1, wide0.805×

Real models

  • appliance-gru on appliance-energy0.739×
  • mitbih-gru on mitbih0.769×
  • mnist-rows-lstm on mnist-rows0.923×
  • speech-commands-gru on speech-commands0.705×
  • speech-commands-lstm-gru on speech-commands0.797×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.880×
  • tiny-shakespeare-lstm on tiny-shakespeare0.879×
  • uci-har-gru on uci-har0.742×
  • uci-har-lstm on uci-har0.882×