Conjectures.io
Tensor kernels

Entry #58

Completed
Hotkey
5Fe1gv…D2daGy
Sent
11 Nov 2026, 16:14 UTC
Source
9a5199dab083

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 #43Done2 of 2 runs0.954×Generated 0.929×, real models 0.980×QualifiesGain lower bound 0.1%Guard passed
MLPvs #44Done2 of 2 runs0.876×Generated 0.936×, real models 0.821×Not significantGain lower bound -9.0%Guard passed
ResNetvs #45Done2 of 2 runs0.958×Generated 0.944×, real models 0.973×Not significantGain lower bound -4.5%Guard passed
MobileNetvs #42Done2 of 2 runs0.995×Generated 1.049×, real models 0.944×Above the barGain lower bound -30.5%Guard passed
Normalized MLPvs #48Done2 of 2 runs1.010×Generated 1.014×, real models 1.006×Above the barGain lower bound -11.1%Guard passed
Mixervs #47Done2 of 2 runs1.015×Generated 1.090×, real models 0.946×Above the barGain lower bound -20.9%Guard passed
Recurrentvs #49Done2 of 2 runs0.992×Generated 1.035×, real models 0.950×Above the barGain lower bound -22.3%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 0.929× · real models 0.980×

MachineTime
  1. prod-amd64Seed 9693346920.958×Done1 attempt0.010× of the baseline kernels' time
  2. prod-arm64Seed 9693346920.951×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.904×
  • Batch 1, sparse0.935×
  • Batch 1, wide0.964×

Mixer

  • Batch 1, normal0.899×
  • Batch 1, sparse0.863×
  • Batch 1, wide0.888×

ResNet

  • Batch 1, normal0.946×
  • Batch 1, sparse0.939×
  • Batch 1, wide0.930×

MobileNet

  • Batch 1, normal1.001×
  • Batch 1, sparse1.010×
  • Batch 1, wide1.009×

Recurrent

  • Batch 1, normal0.831×
  • Batch 1, sparse0.888×
  • Batch 1, wide0.863×

Normalized MLP

  • Batch 1, normal1.047×
  • Batch 1, sparse1.040×
  • Batch 1, wide0.907×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.863×
  • compact-mixer on digits0.562×
  • compact-mixer on fashion0.924×
  • compact-mixer on mnist0.931×
  • compact-mlp on digits1.031×
  • compact-mlp on fashion1.777×
  • compact-mlp on mnist1.736×
  • compact-mobilenet on digits0.909×
  • compact-mobilenet on fashion0.922×
  • compact-mobilenet on mnist0.958×
  • compact-normalized_mlp on digits1.275×
  • compact-normalized_mlp on fashion1.018×
  • compact-normalized_mlp on mnist0.963×
  • compact-resnet on digits0.982×
  • compact-resnet on fashion0.924×
  • compact-resnet on mnist0.929×
  • appliance-gru on appliance-energy0.870×
  • mitbih-gru on mitbih0.816×
  • mnist-rows-lstm on mnist-rows1.044×
  • speech-commands-gru on speech-commands0.831×
  • speech-commands-lstm-gru on speech-commands0.866×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.868×
  • tiny-shakespeare-lstm on tiny-shakespeare0.944×
  • uci-har-gru on uci-har0.810×
  • uci-har-lstm on uci-har0.868×
  • uci-har-mixer on UCI-HAR0.898×
  • uci-har-mlp on UCI-HAR0.941×
  • uci-har-mobilenet on UCI-HAR0.963×
  • uci-har-normalized_mlp on UCI-HAR0.984×
  • uci-har-resnet on UCI-HAR0.955×
MLP2 runs

RunsGenerated 0.936× · real models 0.821×

MachineTime
  1. prod-amd64Seed 9693346920.997×Done1 attempt0.014× of the baseline kernels' time
  2. prod-arm64Seed 9693346920.770×Done1 attempt0.014× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.965×
  • Batch 1, sparse0.996×
  • Batch 1, wide0.917×

Real models

  • compact-mlp on digits0.573×
  • compact-mlp on fashion0.773×
  • compact-mlp on mnist1.021×
  • uci-har-mlp on UCI-HAR1.003×
ResNet2 runs

RunsGenerated 0.944× · real models 0.973×

MachineTime
  1. prod-amd64Seed 9693346921.009×Done1 attempt0.007× of the baseline kernels' time
  2. prod-arm64Seed 9693346920.910×Done1 attempt0.008× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal0.939×
  • Batch 1, sparse0.952×
  • Batch 1, wide0.931×

Real models

  • compact-resnet on digits0.962×
  • compact-resnet on fashion0.983×
  • compact-resnet on mnist0.967×
  • uci-har-resnet on UCI-HAR0.979×
MobileNet2 runs

RunsGenerated 1.049× · real models 0.944×

MachineTime
  1. prod-amd64Seed 9693346920.992×Done1 attempt0.012× of the baseline kernels' time
  2. prod-arm64Seed 9693346920.999×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal0.984×
  • Batch 1, sparse0.995×
  • Batch 1, wide1.009×

Real models

  • compact-mobilenet on digits0.907×
  • compact-mobilenet on fashion0.922×
  • compact-mobilenet on mnist0.958×
  • uci-har-mobilenet on UCI-HAR0.993×
Normalized MLP2 runs

RunsGenerated 1.014× · real models 1.006×

MachineTime
  1. prod-amd64Seed 9693346920.979×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 9693346921.042×Done1 attempt0.010× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.035×
  • Batch 1, sparse1.010×
  • Batch 1, wide0.953×

Real models

  • compact-normalized_mlp on digits0.925×
  • compact-normalized_mlp on fashion1.005×
  • compact-normalized_mlp on mnist1.096×
  • uci-har-normalized_mlp on UCI-HAR1.003×
Mixer2 runs

RunsGenerated 1.090× · real models 0.946×

MachineTime
  1. prod-amd64Seed 9693346921.004×Done1 attempt0.005× of the baseline kernels' time
  2. prod-arm64Seed 9693346921.027×Done1 attempt0.005× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal1.065×
  • Batch 1, sparse1.062×
  • Batch 1, wide1.149×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction1.062×
  • compact-mixer on digits0.814×
  • compact-mixer on fashion0.980×
  • compact-mixer on mnist1.055×
  • uci-har-mixer on UCI-HAR0.846×
Recurrent2 runs

RunsGenerated 1.035× · real models 0.950×

MachineTime
  1. prod-amd64Seed 9693346920.956×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 9693346921.029×Done1 attempt0.014× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal1.031×
  • Batch 1, sparse0.990×
  • Batch 1, wide1.000×

Real models

  • appliance-gru on appliance-energy0.934×
  • mitbih-gru on mitbih0.916×
  • mnist-rows-lstm on mnist-rows0.965×
  • speech-commands-gru on speech-commands0.861×
  • speech-commands-lstm-gru on speech-commands0.934×
  • spoken-arabic-digit-lstm on spoken-arabic-digit1.010×
  • tiny-shakespeare-lstm on tiny-shakespeare1.021×
  • uci-har-gru on uci-har0.906×
  • uci-har-lstm on uci-har1.012×