Conjectures.io
Tensor kernels

Entry #72

Completed
Hotkey
5EWLL7…3QFePh
Sent
12 Nov 2026, 00:10 UTC
Source
ed8acdc5400e

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 runs19.155×Generated 20.640×, real models 17.776×Above the barGain lower bound -2098.9%Guard passed
MLPvs #44Done2 of 2 runs1.004×Generated 0.935×, real models 1.079×Above the barGain lower bound -35.1%Guard passed
ResNetvs #51Done2 of 2 runs105.660×Generated 116.401×, real models 95.910×Above the barGain lower bound -12307.8%Guard passed
MobileNetvs #53Done2 of 2 runs87.706×Generated 91.923×, real models 83.682×Above the barGain lower bound -10251.1%Guard passed
Normalized MLPvs #48Done2 of 2 runs53.185×Generated 76.997×, real models 36.738×Above the barGain lower bound -6395.4%Guard passed
Mixervs #47Done2 of 2 runs172.122×Generated 181.588×, real models 163.150×Above the barGain lower bound -18816.8%Guard passed
Recurrentvs #52Done2 of 2 runs1.060×Generated 1.051×, real models 1.069×Above the barGain lower bound -22.0%Guard passed

The source stays private while it holds a board, and for entries that never held one.

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 20.640× · real models 17.776×

MachineTime
  1. prod-amd64Seed 95888945117.659×Done1 attempt0.215× of the baseline kernels' time
  2. prod-arm64Seed 95888945120.778×Done1 attempt0.230× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.156×
  • Batch 1, sparse0.931×
  • Batch 1, wide0.940×

Mixer

  • Batch 1, normal139.027×
  • Batch 1, sparse141.932×
  • Batch 1, wide164.627×

ResNet

  • Batch 1, normal101.938×
  • Batch 1, sparse110.988×
  • Batch 1, wide98.318×

MobileNet

  • Batch 1, normal81.940×
  • Batch 1, sparse82.662×
  • Batch 1, wide82.571×

Recurrent

  • Batch 1, normal0.950×
  • Batch 1, sparse0.945×
  • Batch 1, wide0.932×

Normalized MLP

  • Batch 1, normal69.288×
  • Batch 1, sparse67.480×
  • Batch 1, wide75.700×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction203.047×
  • compact-mixer on digits178.218×
  • compact-mixer on fashion141.310×
  • compact-mixer on mnist133.560×
  • compact-mlp on digits1.131×
  • compact-mlp on fashion1.183×
  • compact-mlp on mnist1.673×
  • compact-mobilenet on digits57.273×
  • compact-mobilenet on fashion77.684×
  • compact-mobilenet on mnist75.913×
  • compact-normalized_mlp on digits67.560×
  • compact-normalized_mlp on fashion19.566×
  • compact-normalized_mlp on mnist23.062×
  • compact-resnet on digits61.299×
  • compact-resnet on fashion114.474×
  • compact-resnet on mnist111.693×
  • appliance-gru on appliance-energy0.963×
  • mitbih-gru on mitbih0.985×
  • mnist-rows-lstm on mnist-rows0.981×
  • speech-commands-gru on speech-commands0.955×
  • speech-commands-lstm-gru on speech-commands0.970×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.961×
  • tiny-shakespeare-lstm on tiny-shakespeare0.974×
  • uci-har-gru on uci-har0.992×
  • uci-har-lstm on uci-har0.960×
  • uci-har-mixer on UCI-HAR149.271×
  • uci-har-mlp on UCI-HAR0.941×
  • uci-har-mobilenet on UCI-HAR69.149×
  • uci-har-normalized_mlp on UCI-HAR21.033×
  • uci-har-resnet on UCI-HAR70.303×
MLP2 runs

RunsGenerated 0.935× · real models 1.079×

MachineTime
  1. prod-amd64Seed 9588894510.827×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 9588894511.219×Done1 attempt0.020× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.959×
  • Batch 1, sparse0.918×
  • Batch 1, wide0.988×

Real models

  • compact-mlp on digits0.605×
  • compact-mlp on fashion2.251×
  • compact-mlp on mnist1.169×
  • uci-har-mlp on UCI-HAR0.851×
ResNet2 runs

RunsGenerated 116.401× · real models 95.910×

MachineTime
  1. prod-amd64Seed 958889451110.900×Done1 attempt0.988× of the baseline kernels' time
  2. prod-arm64Seed 958889451100.667×Done1 attempt0.997× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal109.095×
  • Batch 1, sparse126.721×
  • Batch 1, wide112.013×

Real models

  • compact-resnet on digits75.674×
  • compact-resnet on fashion119.713×
  • compact-resnet on mnist116.020×
  • uci-har-resnet on UCI-HAR80.507×
MobileNet2 runs

RunsGenerated 91.923× · real models 83.682×

MachineTime
  1. prod-amd64Seed 95888945190.911×Done1 attempt1.180× of the baseline kernels' time
  2. prod-arm64Seed 95888945184.613×Done1 attempt1.160× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal91.247×
  • Batch 1, sparse92.346×
  • Batch 1, wide89.295×

Real models

  • compact-mobilenet on digits67.149×
  • compact-mobilenet on fashion95.155×
  • compact-mobilenet on mnist93.021×
  • uci-har-mobilenet on UCI-HAR82.503×
Normalized MLP2 runs

RunsGenerated 76.997× · real models 36.738×

MachineTime
  1. prod-amd64Seed 95888945154.354×Done1 attempt0.589× of the baseline kernels' time
  2. prod-arm64Seed 95888945152.041×Done1 attempt0.596× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal72.719×
  • Batch 1, sparse82.826×
  • Batch 1, wide84.052×

Real models

  • compact-normalized_mlp on digits84.553×
  • compact-normalized_mlp on fashion24.256×
  • compact-normalized_mlp on mnist41.303×
  • uci-har-normalized_mlp on UCI-HAR21.504×
Mixer2 runs

RunsGenerated 181.588× · real models 163.150×

MachineTime
  1. prod-amd64Seed 958889451174.963×Done1 attempt0.846× of the baseline kernels' time
  2. prod-arm64Seed 958889451169.328×Done1 attempt0.857× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal173.031×
  • Batch 1, sparse174.925×
  • Batch 1, wide204.461×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction229.717×
  • compact-mixer on digits149.486×
  • compact-mixer on fashion159.370×
  • compact-mixer on mnist148.579×
  • uci-har-mixer on UCI-HAR142.163×
Recurrent2 runs

RunsGenerated 1.051× · real models 1.069×

MachineTime
  1. prod-amd64Seed 9588894511.149×Done1 attempt0.015× of the baseline kernels' time
  2. prod-arm64Seed 9588894510.978×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal1.054×
  • Batch 1, sparse1.050×
  • Batch 1, wide1.062×

Real models

  • appliance-gru on appliance-energy1.018×
  • mitbih-gru on mitbih1.087×
  • mnist-rows-lstm on mnist-rows1.058×
  • speech-commands-gru on speech-commands1.063×
  • speech-commands-lstm-gru on speech-commands1.094×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.071×
  • tiny-shakespeare-lstm on tiny-shakespeare1.088×
  • uci-har-gru on uci-har1.058×
  • uci-har-lstm on uci-har1.085×