# openai/transformer-debugger

Repository: https://github.com/openai/transformer-debugger
Canonical: https://ross.abutalabs.com/products/transformer-debugger
Language: Python
License: MIT
License Family: permissive
Last push: 2026-04-15T17:10:05+00:00

## Health v2 (maintenance only)
Score: 60/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 77, release rhythm 35, longevity 64
- inputs: {"age_days": 905, "days_push": 140, "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 4124, forks 244 (observed 2026-08-28T04:08:36.014567+00:00)

## What it is
Transformer Debugger (TDB) is an OpenAI tool for investigating specific behaviors of small language models, combining automated interpretability with sparse autoencoders. It provides a React neuron viewer and a backend activation server that lets users explore model components (neurons, attention heads, autoencoder latents) and intervene in the forward pass without writing code.

## Use cases
- investigate why a model outputs token A instead of token B for a prompt
- understand why an attention head attends to a specific token
- explore neurons and autoencoder latents in GPT-2 without writing code
- trace circuits connecting model components
- browse top-activating dataset examples for neurons and attention heads
- intervene in the forward pass to test hypotheses about model behavior

## When to choose
- you are doing mechanistic interpretability research on small language models like GPT-2
- you want to explore sparse autoencoder latents and neuron activations interactively
- you need to identify which components contribute to a specific model behavior before writing code

## When to avoid
- you need to debug or interpret large frontier-scale models
- you want a general-purpose LLM debugging or monitoring tool rather than interpretability research
- you need a production inference or serving solution

## Facets
- artifact type: application
- maturity: active
- function: machine-learning, llm-inference, data-visualization, developer-tools
- domain: deep-learning, large-language-models, machine-learning, developer-tools
- platform: python, cross-platform
- tags: interpretability, sparse-autoencoders, mechanistic-interpretability, neuron-viewer, gpt-2, superalignment, activation-analysis, nodejs, web-server

## Member repositories
- openai/transformer-debugger (main) score 60

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:36.014567+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 2026-08-29T18:23:04.860178+00:00, confidence not recorded.
  - readme: https://github.com/openai/transformer-debugger (fetched 2026-08-28T04:08:36.014567+00:00, sha a1109fba6a49)
- Data as of 2026-08-30T08:39:29.467469+00:00.