Conjectures.io
Tensor kernels

Entry #94

Completed
Hotkey
5Gq4Bz…HQQBFb
Sent
12 Nov 2026, 07:54 UTC
Source
c5c976d2e433

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 runs17.264×Generated 17.487×, real models 17.044×Above the barGain lower bound -1751.2%Guard passed
MLPvs #44Done2 of 2 runs1.118×Generated 0.964×, real models 1.296×Above the barGain lower bound -45.1%Guard passed
ResNetvs #51Done2 of 2 runs3.717×Generated 3.768×, real models 3.667×Above the barGain lower bound -358.7%Guard passed
MobileNetvs #53Done2 of 2 runs0.956×Generated 0.955×, real models 0.957×Not significantGain lower bound -15.9%Guard passed
Normalized MLPvs #48Done2 of 2 runs44.457×Generated 64.708×, real models 30.544×Above the barGain lower bound -5606.4%Guard passed
Mixervs #47Done2 of 2 runs174.604×Generated 180.960×, real models 168.471×Above the barGain lower bound -18844.1%Guard passed
Recurrentvs #55Done2 of 2 runs850.500×Generated 761.890×, real models 949.416×Above the barGain lower bound -88961.2%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 17.487× · real models 17.044×

MachineTime
  1. prod-amd64Seed 20554566418.028×Done1 attempt0.165× of the baseline kernels' time
  2. prod-arm64Seed 20554566416.533×Done1 attempt0.174× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.946×
  • Batch 1, sparse0.993×
  • Batch 1, wide1.033×

Mixer

  • Batch 1, normal166.798×
  • Batch 1, sparse167.909×
  • Batch 1, wide195.605×

ResNet

  • Batch 1, normal3.875×
  • Batch 1, sparse4.434×
  • Batch 1, wide3.839×

MobileNet

  • Batch 1, normal0.977×
  • Batch 1, sparse1.044×
  • Batch 1, wide1.000×

Recurrent

  • Batch 1, normal706.668×
  • Batch 1, sparse665.175×
  • Batch 1, wide656.220×

Normalized MLP

  • Batch 1, normal58.793×
  • Batch 1, sparse59.876×
  • Batch 1, wide62.058×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction198.268×
  • compact-mixer on digits221.105×
  • compact-mixer on fashion159.233×
  • compact-mixer on mnist153.961×
  • compact-mlp on digits0.927×
  • compact-mlp on fashion2.836×
  • compact-mlp on mnist1.653×
  • compact-mobilenet on digits0.838×
  • compact-mobilenet on fashion0.804×
  • compact-mobilenet on mnist0.811×
  • compact-normalized_mlp on digits79.587×
  • compact-normalized_mlp on fashion34.397×
  • compact-normalized_mlp on mnist27.620×
  • compact-resnet on digits2.634×
  • compact-resnet on fashion3.563×
  • compact-resnet on mnist3.696×
  • appliance-gru on appliance-energy719.719×
  • mitbih-gru on mitbih899.629×
  • mnist-rows-lstm on mnist-rows730.605×
  • speech-commands-gru on speech-commands710.523×
  • speech-commands-lstm-gru on speech-commands865.534×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1118.564×
  • tiny-shakespeare-lstm on tiny-shakespeare831.186×
  • uci-har-gru on uci-har826.498×
  • uci-har-lstm on uci-har1151.466×
  • uci-har-mixer on UCI-HAR161.665×
  • uci-har-mlp on UCI-HAR1.069×
  • uci-har-mobilenet on UCI-HAR0.891×
  • uci-har-normalized_mlp on UCI-HAR19.945×
  • uci-har-resnet on UCI-HAR5.691×
MLP2 runs

RunsGenerated 0.964× · real models 1.296×

MachineTime
  1. prod-amd64Seed 2055456640.915×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 2055456641.367×Done1 attempt0.014× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.992×
  • Batch 1, sparse0.969×
  • Batch 1, wide0.959×

Real models

  • compact-mlp on digits1.126×
  • compact-mlp on fashion2.150×
  • compact-mlp on mnist1.172×
  • uci-har-mlp on UCI-HAR0.996×
ResNet2 runs

RunsGenerated 3.768× · real models 3.667×

MachineTime
  1. prod-amd64Seed 2055456643.832×Done1 attempt0.031× of the baseline kernels' time
  2. prod-arm64Seed 2055456643.605×Done1 attempt0.033× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal3.370×
  • Batch 1, sparse4.278×
  • Batch 1, wide4.002×

Real models

  • compact-resnet on digits2.681×
  • compact-resnet on fashion3.618×
  • compact-resnet on mnist3.583×
  • uci-har-resnet on UCI-HAR5.199×
MobileNet2 runs

RunsGenerated 0.955× · real models 0.957×

MachineTime
  1. prod-amd64Seed 2055456640.950×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 2055456640.961×Done1 attempt0.008× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal0.947×
  • Batch 1, sparse0.981×
  • Batch 1, wide1.005×

Real models

  • compact-mobilenet on digits0.939×
  • compact-mobilenet on fashion0.961×
  • compact-mobilenet on mnist0.965×
  • uci-har-mobilenet on UCI-HAR0.961×
Normalized MLP2 runs

RunsGenerated 64.708× · real models 30.544×

MachineTime
  1. prod-amd64Seed 20554566446.473×Done1 attempt0.567× of the baseline kernels' time
  2. prod-arm64Seed 20554566442.528×Done1 attempt0.571× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal62.240×
  • Batch 1, sparse64.478×
  • Batch 1, wide63.968×

Real models

  • compact-normalized_mlp on digits97.723×
  • compact-normalized_mlp on fashion23.688×
  • compact-normalized_mlp on mnist20.623×
  • uci-har-normalized_mlp on UCI-HAR18.230×
Mixer2 runs

RunsGenerated 180.960× · real models 168.471×

MachineTime
  1. prod-amd64Seed 205545664181.672×Done1 attempt0.854× of the baseline kernels' time
  2. prod-arm64Seed 205545664167.811×Done1 attempt0.848× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal172.853×
  • Batch 1, sparse174.302×
  • Batch 1, wide208.426×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction242.951×
  • compact-mixer on digits161.223×
  • compact-mixer on fashion155.593×
  • compact-mixer on mnist153.111×
  • uci-har-mixer on UCI-HAR145.440×
Recurrent2 runs

RunsGenerated 761.890× · real models 949.416×

MachineTime
  1. prod-amd64Seed 205545664905.437×Done1 attempt7.870× of the baseline kernels' time
  2. prod-arm64Seed 205545664798.897×Done1 attempt7.972× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal804.428×
  • Batch 1, sparse732.425×
  • Batch 1, wide739.278×

Real models

  • appliance-gru on appliance-energy761.424×
  • mitbih-gru on mitbih982.273×
  • mnist-rows-lstm on mnist-rows768.918×
  • speech-commands-gru on speech-commands768.909×
  • speech-commands-lstm-gru on speech-commands973.127×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1310.932×
  • tiny-shakespeare-lstm on tiny-shakespeare903.035×
  • uci-har-gru on uci-har916.979×
  • uci-har-lstm on uci-har1341.777×