Categories Companies Research

Loading the SOTA2 catalog…

Explore

Categories Companies Research

Contribute

Join SOTA2 Contact

SOTA2

© 2026 SOTA2

Build in Bellevue, USA

WizWand is now SOTA2 Research

Papers Benchmarks Datasets Tasks

Pythia dataset and benchmarks – Page 2 · SOTA2 Research

Research Datasets

Pythia

Benchmarks

Task NameDataset NameSOTA MethodSort

Task NameDataset NameSOTA ResultTrendResults

Activation Reconstruction

Pythia 1.4b

0.22MSE

4

Pythia 410m

77.5Average F1 Score

4

Activation Reconstruction

Pythia 410m

0.03MSE

4

Sparse Autoencoding

Pythia 70M (layer-3 residuals)

33.9Achieved L0

4

Jailbreak Attack

Pythia-12B

70Attack Success Rate

4

Automated Interpretability

Pythia-160M

0.857AutoInterp Score

3

SAE Pareto Dominance Analysis

Pythia-1B activations

100Dominance Fraction

3

Influential training example identification

Pythia 6.9B

83.1Top-1 Accuracy

2

Page 2 of 2

Popular tasks

Sparse Autoencoding Activation Reconstruction Token Extrapolation Scaling Law Modeling Model Extrapolation

Follow for updates

Join SOTA2 Contact