Description
Ecco is a Python library that creates interactive visualizations to help you look inside language models. It lets you see what an NLP model is thinking. You can give GPT-2 some input text and watch it generate output token by token, see which input words contributed most to each output word through input saliency, examine the neuron activations the library captures from the model, and look for patterns in those activations using non-negative matrix factorization. The site has example visualizations and two Colab notebooks you can run in your browser, one for neuron activations and one for input saliency. The project is an alpha release of a research project focused on GPT-based transformer language models, and the author invites feedback. The code is on GitHub under jalammar.…
Highlights
- Visualize token-by-token generation
- Input saliency for each output word
- Neuron activation exploration
- Colab notebooks to try in the browser
Ratings & Reviews
No ratings yet
Be the first to rate Ecco.
People also visit
Ecco FAQ
What is Ecco?
A Python library that creates interactive visualizations of what language models are doing.
Which models does it support?
The site says it focuses on GPT-based transformer language models.
Can I try it without installing?
Yes. The site links two Colab notebooks you can run in the browser.
Is it production ready?
No. The site calls it an alpha release of a research project.
What are the best Ecco alternatives?
Popular alternatives to Ecco include Perplexity and Lucid. See all Ecco alternatives









