Conjectures.io
Tensor kernels

Entry #102

Completed
Hotkey
5DCowR…2HyW8U
Sent
12 Nov 2026, 10:38 UTC
Source
48d7b3bd6422

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 runs81.815×Generated 93.245×, real models 71.785×Above the barGain lower bound -8831.5%Guard passed
MLPvs #44Done2 of 2 runs0.823×Generated 0.973×, real models 0.697×Not significantGain lower bound -1.5%Guard passed
ResNetvs #51Done2 of 2 runs5548.696×Generated 6551.271×, real models 4699.551×Above the barGain lower bound -649885.2%Guard passed
MobileNetvs #53Done2 of 2 runs2524.300×Generated 2784.063×, real models 2288.773×Above the barGain lower bound -281039.7%Guard passed
Normalized MLPvs #48Done2 of 2 runs47.144×Generated 61.203×, real models 36.315×Above the barGain lower bound -5642.0%Guard passed
Mixervs #47Done2 of 2 runs609.334×Generated 638.806×, real models 581.221×Above the barGain lower bound -69391.9%Guard passed
Recurrentvs #56Done2 of 2 runs0.941×Generated 0.937×, real models 0.945×QualifiesGain lower bound 1.1%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 93.245× · real models 71.785×

MachineTime
  1. prod-amd64Seed 310132328884.999×Done1 attempt0.958× of the baseline kernels' time
  2. prod-arm64Seed 310132328878.750×Done1 attempt0.935× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.155×
  • Batch 1, sparse1.165×
  • Batch 1, wide1.172×

Mixer

  • Batch 1, normal591.367×
  • Batch 1, sparse571.565×
  • Batch 1, wide643.606×

ResNet

  • Batch 1, normal6147.887×
  • Batch 1, sparse6929.832×
  • Batch 1, wide6128.285×

MobileNet

  • Batch 1, normal2728.245×
  • Batch 1, sparse2638.776×
  • Batch 1, wide2684.125×

Recurrent

  • Batch 1, normal0.888×
  • Batch 1, sparse0.886×
  • Batch 1, wide0.870×

Normalized MLP

  • Batch 1, normal56.631×
  • Batch 1, sparse59.048×
  • Batch 1, wide58.497×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction853.579×
  • compact-mixer on digits449.981×
  • compact-mixer on fashion632.512×
  • compact-mixer on mnist644.390×
  • compact-mlp on digits0.878×
  • compact-mlp on fashion0.983×
  • compact-mlp on mnist0.541×
  • compact-mobilenet on digits1643.682×
  • compact-mobilenet on fashion2327.778×
  • compact-mobilenet on mnist2203.466×
  • compact-normalized_mlp on digits77.773×
  • compact-normalized_mlp on fashion24.403×
  • compact-normalized_mlp on mnist18.697×
  • compact-resnet on digits4581.222×
  • compact-resnet on fashion6954.951×
  • compact-resnet on mnist7138.002×
  • appliance-gru on appliance-energy0.887×
  • mitbih-gru on mitbih0.822×
  • mnist-rows-lstm on mnist-rows0.744×
  • speech-commands-gru on speech-commands0.856×
  • speech-commands-lstm-gru on speech-commands0.893×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.924×
  • tiny-shakespeare-lstm on tiny-shakespeare0.713×
  • uci-har-gru on uci-har0.915×
  • uci-har-lstm on uci-har0.921×
  • uci-har-mixer on UCI-HAR614.658×
  • uci-har-mlp on UCI-HAR1.095×
  • uci-har-mobilenet on UCI-HAR2353.455×
  • uci-har-normalized_mlp on UCI-HAR20.474×
  • uci-har-resnet on UCI-HAR2610.243×
MLP2 runs

RunsGenerated 0.973× · real models 0.697×

MachineTime
  1. prod-amd64Seed 31013232880.787×Done1 attempt0.015× of the baseline kernels' time
  2. prod-arm64Seed 31013232880.862×Done1 attempt0.017× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.002×
  • Batch 1, sparse1.014×
  • Batch 1, wide0.951×

Real models

  • compact-mlp on digits0.604×
  • compact-mlp on fashion0.653×
  • compact-mlp on mnist0.589×
  • uci-har-mlp on UCI-HAR1.014×
ResNet2 runs

RunsGenerated 6551.271× · real models 4699.551×

MachineTime
  1. prod-amd64Seed 31013232885702.655×Done1 attempt51.727× of the baseline kernels' time
  2. prod-arm64Seed 31013232885398.895×Done1 attempt52.753× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal6246.620×
  • Batch 1, sparse6922.829×
  • Batch 1, wide6163.682×

Real models

  • compact-resnet on digits3950.388×
  • compact-resnet on fashion6947.804×
  • compact-resnet on mnist7057.255×
  • uci-har-resnet on UCI-HAR2518.269×
MobileNet2 runs

RunsGenerated 2784.063× · real models 2288.773×

MachineTime
  1. prod-amd64Seed 31013232882601.194×Done1 attempt32.909× of the baseline kernels' time
  2. prod-arm64Seed 31013232882449.679×Done1 attempt33.576× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal2810.338×
  • Batch 1, sparse2690.827×
  • Batch 1, wide2712.746×

Real models

  • compact-mobilenet on digits1682.639×
  • compact-mobilenet on fashion2642.328×
  • compact-mobilenet on mnist2614.390×
  • uci-har-mobilenet on UCI-HAR2360.821×
Normalized MLP2 runs

RunsGenerated 61.203× · real models 36.315×

MachineTime
  1. prod-amd64Seed 310132328847.451×Done1 attempt0.544× of the baseline kernels' time
  2. prod-arm64Seed 310132328846.839×Done1 attempt0.545× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal57.353×
  • Batch 1, sparse61.014×
  • Batch 1, wide62.986×

Real models

  • compact-normalized_mlp on digits83.415×
  • compact-normalized_mlp on fashion20.270×
  • compact-normalized_mlp on mnist43.128×
  • uci-har-normalized_mlp on UCI-HAR23.851×
Mixer2 runs

RunsGenerated 638.806× · real models 581.221×

MachineTime
  1. prod-amd64Seed 3101323288613.975×Done1 attempt3.086× of the baseline kernels' time
  2. prod-arm64Seed 3101323288604.728×Done1 attempt3.104× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal693.963×
  • Batch 1, sparse650.192×
  • Batch 1, wide714.480×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction897.382×
  • compact-mixer on digits320.840×
  • compact-mixer on fashion626.357×
  • compact-mixer on mnist656.391×
  • uci-har-mixer on UCI-HAR560.346×
Recurrent2 runs

RunsGenerated 0.937× · real models 0.945×

MachineTime
  1. prod-amd64Seed 31013232880.905×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 31013232880.978×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.926×
  • Batch 1, sparse0.918×
  • Batch 1, wide0.949×

Real models

  • appliance-gru on appliance-energy0.852×
  • mitbih-gru on mitbih0.857×
  • mnist-rows-lstm on mnist-rows1.278×
  • speech-commands-gru on speech-commands0.805×
  • speech-commands-lstm-gru on speech-commands0.947×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.003×
  • tiny-shakespeare-lstm on tiny-shakespeare0.981×
  • uci-har-gru on uci-har0.838×
  • uci-har-lstm on uci-har1.020×