# EleutherAI/pythia

The hub for EleutherAI's work on interpretability and learning dynamics

Repository: https://github.com/EleutherAI/pythia
Canonical: https://ross.abutalabs.com/products/pythia
Language: Jupyter Notebook
License: Apache-2.0
License Family: permissive
Last push: 2025-11-15T17:39:52+00:00

## Health v2 (maintenance only)
Score: 56/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 52, release rhythm 35, longevity 100
- inputs: {"age_days": 1712, "days_push": 291, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2911, forks 222 (observed 2026-08-28T04:07:29.277095+00:00)

## What it is
EleutherAI's Pythia project: a suite of autoregressive transformer language models released with 154 training checkpoints each, plus code and reproducibility instructions for interpretability research. The repository serves as a hub for the model suite and associated papers on memorization and pre-training stability.

## Use cases
- study how knowledge develops in LLMs during training
- research learning dynamics across model scales
- run causal interventions on the training data order
- reproduce interpretability analyses of transformers
- study memorization behavior in large language models
- compare stability across many pre-training runs

## When to choose
- you need checkpoints throughout training to analyze learning dynamics
- you want fully reproducible open models for interpretability research
- you need models trained on identical data in identical order for causal studies

## When to avoid
- you need a production-ready LLM for inference or deployment
- you want the most capable model rather than a research suite
- you need instruction-tuned or chat models

## Facets
- artifact type: dataset
- maturity: stable
- function: machine-learning, deep-learning, llm-training, data-science
- domain: large-language-models, machine-learning, artificial-intelligence, tutorials
- platform: python, cross-platform
- tags: interpretability, model-suite, checkpoints, scaling-laws, learning-dynamics, transformers, research, open-science, gpu

## Member repositories
- EleutherAI/pythia (main) score 56

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:29.277095+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T07:34:15.021959+00:00, confidence not recorded.
  - readme: https://github.com/EleutherAI/pythia (fetched 2026-08-28T04:07:29.277095+00:00, sha 164a248fac4c)
- Data as of 2026-08-30T08:39:29.467469+00:00.