Pythia

text · generative · anchor · artifact

Suite of language models trained on one corpus in a fixed data order across sizes, built to make training dynamics comparable between scales. Cited in the corpus from 410M to 2.8B.

Note
anchor found by search rather than in a citing paper, and checked against this entry's own description before it was recorded: "Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling"
Variants
Pythia 410M, Pythia 1B, Pythia 1.4B, Pythia 2.8B, Pythia-6.9B, Pythia-12B, Pythia-160m, Pythia-70m, Pythia-70m-deduped

Findings

Shared mechanisms