Learning Interpretability Tool
The Learning Interpretability Tool (🔥LIT) is a visual, interactive ML model-understanding tool that supports text, image, and tabular data.

overview of LIT

The Learning Interpretability Tool (🔥LIT) is for researchers and practitioners looking to understand NLP model behavior through a visual, interactive, and extensible tool.

Use LIT to ask and answer questions like:

  • What kind of examples does my model perform poorly on?
  • Why did my model make this prediction? Can it attribute it to adversarial behavior, or undesirable priors from the training set?
  • Does my model behave consistently if I change things like textual style, verb tense, or pronoun gender?

LIT contains many built-in capabilities but is also customizable, with the ability to add custom interpretability techniques, metrics calculations, counterfactual generators, visualizations, and more.

In addition to language, LIT also includes preliminary support for models operating on tabular and image data. For a similar tool built to explore general-purpose machine learning models, check out the What-If Tool.

LIT can be run as a standalone server, or inside of python notebook environments such as Colab, Jupyter, and Google Cloud Vertex AI Notebooks.

Flexible and powerful model probing

Built-in capabilities

Salience maps

Attention visualization

Metrics calculations

Counterfactual generation

Model and datapoint comparison

Embedding visualization


And more...

Supported task types



Text generation / seq2seq

Masked language models

Span labeling

Multi-headed models

Image and tabular data

And more...

Framework agnostic

TensorFlow 1.x

TensorFlow 2.x


Notebook compatibility

Custom inference code

Remote Procedure Calls

And more...

What's the latest


Version 1.1

Input salience for text-to-text LLMs, with wrappers for HuggingFace Transformers and KerasNLP models.


LIT is open-source and easily extensible to new models, tasks, and more.

Demo Paper at EMNLP ‘20

Read about what went into LIT in our demo paper, presented at EMNLP ‘20.