# PAIR-code/lit

The Learning Interpretability Tool: Interactively analyze ML models to understand their behavior in an extensible and framework agnostic interface.

Repository: https://github.com/PAIR-code/lit
Canonical: https://ross.abutalabs.com/products/pair-code-lit
Homepage: https://pair-code.github.io/lit
Language: TypeScript
License: Apache-2.0
License Family: permissive
Topics: machine-learning, natural-language-processing, visualization
Last push: 2026-07-29T23:56:00+00:00

## Health v2 (maintenance only)
Score: 73/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 95, release rhythm 28, longevity 100
- inputs: {"age_days": 2227, "days_push": 35, "days_rel": 621, "gap_med": 59, "n_releases_24m": 2}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3660, forks 369 (observed 2026-08-28T04:08:13.284906+00:00)

## What it is
The Learning Interpretability Tool (LIT) is a visual, interactive web-based tool for understanding and debugging ML model behavior across text, image, and tabular data. It is framework-agnostic (TensorFlow, PyTorch, HuggingFace, etc.), extensible via plug-ins, and runs as a standalone server or inside notebooks like Colab and Jupyter.

## Use cases
- debug why my NLP model makes certain predictions
- visualize salience maps for model explanations
- compare two models side by side on the same examples
- find examples my model performs poorly on
- generate counterfactual examples to test model robustness
- inspect LLM text generation behavior interactively
- visualize embedding spaces and slice model metrics

## When to choose
- you need interactive, visual exploration of model predictions and errors
- you want a framework-agnostic tool that works with TensorFlow, PyTorch, or HuggingFace models
- you work in notebooks (Colab, Jupyter, Vertex AI) and want inline model analysis
- you need extensible interpretability with custom metrics, generators, and visualizations

## When to avoid
- you need fully automated, headless explainability pipelines without a UI
- you only need simple feature-importance plots from a training library
- your team requires a lightweight CLI-only workflow with no browser interface
- you need production model monitoring rather than research debugging

## Facets
- artifact type: application
- maturity: active
- function: data-visualization, machine-learning, nlp, developer-tools
- domain: machine-learning, data-visualization, artificial-intelligence
- platform: python, cross-platform, browser
- tags: interpretability, explainable-ai, model-debugging, salience-maps, counterfactual-analysis, llm-debugging, notebooks, natural-language-processing, web-server

## Member repositories
- PAIR-code/lit (main) score 73

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:13.284906+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 2026-08-29T18:31:35.361650+00:00, confidence not recorded.
  - readme: https://github.com/PAIR-code/lit (fetched 2026-08-28T04:08:13.284906+00:00, sha 6895714497cd)
  - homepage: https://pair-code.github.io/lit (fetched 2026-08-29T09:25:53.319656+00:00, sha 9edef8a58a44)
- Data as of 2026-08-30T08:39:29.467469+00:00.