← all datasets

Pythia-14M

Emerging
1papers using it
2026first seen

Pythia-14M is a dataset used to evaluate the susceptibility of language models by measuring their response to perturbations in the data distribution, resulting in the identification of interpretable clusters related to various patterns and structures.

Papers using Pythia-14M (1)

Pythia-14M dataset β€” papers, benchmarks & downloads Β· AI for Science