Conjectures.io
Tensor kernels

Entry #108

Completed
Hotkey
5E7RoH…v4pqyy
Sent
12 Nov 2026, 14:06 UTC
Source
2dd5615a067c

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 runs0.986×Generated 0.982×, real models 0.989×Above the barGain lower bound -2.4%Guard passed
MLPvs #44Done2 of 2 runs1.278×Generated 1.075×, real models 1.519×Above the barGain lower bound -49.2%Guard passed
ResNetvs #51Done2 of 2 runs0.961×Generated 1.013×, real models 0.911×QualifiesGain lower bound 0.4%Guard passed
MobileNetvs #53Done2 of 2 runs1.151×Generated 1.249×, real models 1.061×Above the barGain lower bound -45.0%Guard passed
Normalized MLPvs #48Done2 of 2 runs1.059×Generated 1.011×, real models 1.109×Above the barGain lower bound -19.6%Guard passed
Mixervs #58Done2 of 2 runs1.046×Generated 1.011×, real models 1.082×Above the barGain lower bound -12.9%Guard passed
Recurrentvs #57Done2 of 2 runs1.044×Generated 1.008×, real models 1.080×Above the barGain lower bound -15.5%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 0.982× · real models 0.989×

MachineTime
  1. prod-amd64Seed 26723518420.973×Done1 attempt0.010× of the baseline kernels' time
  2. prod-arm64Seed 26723518420.999×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.870×
  • Batch 1, sparse1.072×
  • Batch 1, wide1.122×

Mixer

  • Batch 1, normal1.015×
  • Batch 1, sparse0.938×
  • Batch 1, wide0.903×

ResNet

  • Batch 1, normal0.963×
  • Batch 1, sparse1.052×
  • Batch 1, wide0.979×

MobileNet

  • Batch 1, normal1.006×
  • Batch 1, sparse1.040×
  • Batch 1, wide1.006×

Recurrent

  • Batch 1, normal0.869×
  • Batch 1, sparse0.839×
  • Batch 1, wide0.869×

Normalized MLP

  • Batch 1, normal1.050×
  • Batch 1, sparse1.075×
  • Batch 1, wide1.049×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.995×
  • compact-mixer on digits1.220×
  • compact-mixer on fashion1.027×
  • compact-mixer on mnist1.038×
  • compact-mlp on digits0.950×
  • compact-mlp on fashion1.125×
  • compact-mlp on mnist1.419×
  • compact-mobilenet on digits0.799×
  • compact-mobilenet on fashion0.900×
  • compact-mobilenet on mnist0.913×
  • compact-normalized_mlp on digits0.864×
  • compact-normalized_mlp on fashion1.025×
  • compact-normalized_mlp on mnist1.171×
  • compact-resnet on digits0.854×
  • compact-resnet on fashion0.965×
  • compact-resnet on mnist0.950×
  • appliance-gru on appliance-energy0.896×
  • mitbih-gru on mitbih0.853×
  • mnist-rows-lstm on mnist-rows0.971×
  • speech-commands-gru on speech-commands0.863×
  • speech-commands-lstm-gru on speech-commands0.887×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.871×
  • tiny-shakespeare-lstm on tiny-shakespeare0.913×
  • uci-har-gru on uci-har0.879×
  • uci-har-lstm on uci-har0.860×
  • uci-har-mixer on UCI-HAR1.047×
  • uci-har-mlp on UCI-HAR0.902×
  • uci-har-mobilenet on UCI-HAR1.233×
  • uci-har-normalized_mlp on UCI-HAR0.982×
  • uci-har-resnet on UCI-HAR1.107×
MLP2 runs

RunsGenerated 1.075× · real models 1.519×

MachineTime
  1. prod-amd64Seed 26723518421.081×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 26723518421.510×Done1 attempt0.016× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.015×
  • Batch 1, sparse1.027×
  • Batch 1, wide1.032×

Real models

  • compact-mlp on digits1.067×
  • compact-mlp on fashion2.195×
  • compact-mlp on mnist2.238×
  • uci-har-mlp on UCI-HAR1.014×
ResNet2 runs

RunsGenerated 1.013× · real models 0.911×

MachineTime
  1. prod-amd64Seed 26723518421.018×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 26723518420.907×Done1 attempt0.009× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal0.987×
  • Batch 1, sparse1.009×
  • Batch 1, wide0.958×

Real models

  • compact-resnet on digits0.779×
  • compact-resnet on fashion0.964×
  • compact-resnet on mnist0.952×
  • uci-har-resnet on UCI-HAR0.965×
MobileNet2 runs

RunsGenerated 1.249× · real models 1.061×

MachineTime
  1. prod-amd64Seed 26723518421.057×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 26723518421.254×Done1 attempt0.013× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.415×
  • Batch 1, sparse1.074×
  • Batch 1, wide1.034×

Real models

  • compact-mobilenet on digits0.880×
  • compact-mobilenet on fashion1.061×
  • compact-mobilenet on mnist1.100×
  • uci-har-mobilenet on UCI-HAR1.231×
Normalized MLP2 runs

RunsGenerated 1.011× · real models 1.109×

MachineTime
  1. prod-amd64Seed 26723518421.031×Done1 attempt0.010× of the baseline kernels' time
  2. prod-arm64Seed 26723518421.087×Done1 attempt0.009× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.020×
  • Batch 1, sparse1.009×
  • Batch 1, wide1.133×

Real models

  • compact-normalized_mlp on digits1.068×
  • compact-normalized_mlp on fashion1.138×
  • compact-normalized_mlp on mnist1.031×
  • uci-har-normalized_mlp on UCI-HAR1.208×
Mixer2 runs

RunsGenerated 1.011× · real models 1.082×

MachineTime
  1. prod-amd64Seed 26723518421.098×Done1 attempt0.004× of the baseline kernels' time
  2. prod-arm64Seed 26723518420.996×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal1.017×
  • Batch 1, sparse1.009×
  • Batch 1, wide1.094×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.082×
  • compact-mixer on digits1.317×
  • compact-mixer on fashion1.019×
  • compact-mixer on mnist1.012×
  • uci-har-mixer on UCI-HAR1.009×
Recurrent2 runs

RunsGenerated 1.008× · real models 1.080×

MachineTime
  1. prod-amd64Seed 26723518421.054×Done1 attempt0.010× of the baseline kernels' time
  2. prod-arm64Seed 26723518421.033×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal1.026×
  • Batch 1, sparse0.998×
  • Batch 1, wide1.106×

Real models

  • appliance-gru on appliance-energy1.090×
  • mitbih-gru on mitbih1.064×
  • mnist-rows-lstm on mnist-rows1.305×
  • speech-commands-gru on speech-commands1.046×
  • speech-commands-lstm-gru on speech-commands0.983×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.969×
  • tiny-shakespeare-lstm on tiny-shakespeare1.318×
  • uci-har-gru on uci-har1.062×
  • uci-har-lstm on uci-har0.947×