Conjectures.io
Tensor kernels

Entry #30

Completed
Hotkey
5HTEvs…KozDZa
Sent
11 Nov 2026, 03:56 UTC
Source
4575d1782d1e

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 #36Done2 of 2 runs1.036×Generated 1.030×, real models 1.042×Above the barGain lower bound -9.6%Guard passed
MLPvs #1Done2 of 2 runs1.011×Generated 0.920×, real models 1.111×Above the barGain lower bound -29.9%Guard passed
ResNetvs #32Done2 of 2 runs0.997×Generated 0.999×, real models 0.995×Above the barGain lower bound -12.1%Guard passed
MobileNetvs #33Done2 of 2 runs0.985×Generated 1.007×, real models 0.964×Above the barGain lower bound -1.5%Guard passed
Normalized MLPvs #6Done2 of 2 runs1.008×Generated 1.084×, real models 0.936×Above the barGain lower bound -10.9%Guard passed
Mixervs #24Done2 of 2 runs0.975×Generated 1.008×, real models 0.943×Above the barGain lower bound -11.4%Guard passed
Recurrentvs #35Done2 of 2 runs1.000×Generated 1.004×, real models 0.996×Above the barGain lower bound -1.6%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 1.030× · real models 1.042×

MachineTime
  1. prod-amd64Seed 19810532231.080×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 19810532230.993×Done1 attempt0.014× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.210×
  • Batch 1, sparse1.078×
  • Batch 1, wide1.011×

Mixer

  • Batch 1, normal0.966×
  • Batch 1, sparse0.989×
  • Batch 1, wide1.030×

ResNet

  • Batch 1, normal0.996×
  • Batch 1, sparse0.996×
  • Batch 1, wide0.992×

MobileNet

  • Batch 1, normal1.006×
  • Batch 1, sparse0.976×
  • Batch 1, wide0.987×

Recurrent

  • Batch 1, normal0.998×
  • Batch 1, sparse1.005×
  • Batch 1, wide1.001×

Normalized MLP

  • Batch 1, normal1.239×
  • Batch 1, sparse1.015×
  • Batch 1, wide1.125×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.018×
  • compact-mixer on digits1.027×
  • compact-mixer on fashion0.971×
  • compact-mixer on mnist1.013×
  • compact-mlp on digits1.018×
  • compact-mlp on fashion1.156×
  • compact-mlp on mnist1.347×
  • compact-mobilenet on digits0.964×
  • compact-mobilenet on fashion0.998×
  • compact-mobilenet on mnist0.994×
  • compact-normalized_mlp on digits1.133×
  • compact-normalized_mlp on fashion1.221×
  • compact-normalized_mlp on mnist1.489×
  • compact-resnet on digits1.004×
  • compact-resnet on fashion0.994×
  • compact-resnet on mnist0.998×
  • appliance-gru on appliance-energy1.011×
  • mitbih-gru on mitbih0.994×
  • mnist-rows-lstm on mnist-rows1.002×
  • speech-commands-gru on speech-commands0.997×
  • speech-commands-lstm-gru on speech-commands0.999×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.002×
  • tiny-shakespeare-lstm on tiny-shakespeare1.004×
  • uci-har-gru on uci-har0.997×
  • uci-har-lstm on uci-har1.001×
  • uci-har-mixer on UCI-HAR1.012×
  • uci-har-mlp on UCI-HAR0.967×
  • uci-har-mobilenet on UCI-HAR1.002×
  • uci-har-normalized_mlp on UCI-HAR0.876×
  • uci-har-resnet on UCI-HAR0.970×
MLP2 runs

RunsGenerated 0.920× · real models 1.111×

MachineTime
  1. prod-amd64Seed 19810532230.866×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 19810532231.181×Done1 attempt0.025× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.932×
  • Batch 1, sparse0.946×
  • Batch 1, wide0.912×

Real models

  • compact-mlp on digits1.054×
  • compact-mlp on fashion2.180×
  • compact-mlp on mnist0.676×
  • uci-har-mlp on UCI-HAR0.983×
ResNet2 runs

RunsGenerated 0.999× · real models 0.995×

MachineTime
  1. prod-amd64Seed 19810532231.024×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 19810532230.971×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal0.981×
  • Batch 1, sparse0.999×
  • Batch 1, wide1.008×

Real models

  • compact-resnet on digits1.034×
  • compact-resnet on fashion0.960×
  • compact-resnet on mnist0.942×
  • uci-har-resnet on UCI-HAR1.047×
MobileNet2 runs

RunsGenerated 1.007× · real models 0.964×

MachineTime
  1. prod-amd64Seed 19810532231.003×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 19810532230.967×Done1 attempt0.014× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal0.989×
  • Batch 1, sparse1.000×
  • Batch 1, wide1.016×

Real models

  • compact-mobilenet on digits1.026×
  • compact-mobilenet on fashion0.951×
  • compact-mobilenet on mnist0.938×
  • uci-har-mobilenet on UCI-HAR0.942×
Normalized MLP2 runs

RunsGenerated 1.084× · real models 0.936×

MachineTime
  1. prod-amd64Seed 19810532230.993×Done1 attempt0.010× of the baseline kernels' time
  2. prod-arm64Seed 19810532231.022×Done1 attempt0.013× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal0.986×
  • Batch 1, sparse1.122×
  • Batch 1, wide0.955×

Real models

  • compact-normalized_mlp on digits1.030×
  • compact-normalized_mlp on fashion0.798×
  • compact-normalized_mlp on mnist0.911×
  • uci-har-normalized_mlp on UCI-HAR1.025×
Mixer2 runs

RunsGenerated 1.008× · real models 0.943×

MachineTime
  1. prod-amd64Seed 19810532231.003×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 19810532230.948×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.965×
  • Batch 1, sparse1.005×
  • Batch 1, wide0.975×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.072×
  • compact-mixer on digits0.673×
  • compact-mixer on fashion1.003×
  • compact-mixer on mnist0.979×
  • uci-har-mixer on UCI-HAR1.054×
Recurrent2 runs

RunsGenerated 1.004× · real models 0.996×

MachineTime
  1. prod-amd64Seed 19810532231.002×Done1 attempt0.024× of the baseline kernels' time
  2. prod-arm64Seed 19810532230.998×Done1 attempt0.024× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal1.003×
  • Batch 1, sparse0.988×
  • Batch 1, wide1.009×

Real models

  • appliance-gru on appliance-energy1.004×
  • mitbih-gru on mitbih0.987×
  • mnist-rows-lstm on mnist-rows0.986×
  • speech-commands-gru on speech-commands0.985×
  • speech-commands-lstm-gru on speech-commands0.993×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.011×
  • tiny-shakespeare-lstm on tiny-shakespeare0.996×
  • uci-har-gru on uci-har0.997×
  • uci-har-lstm on uci-har1.001×