Ecco: Inspect Transformer Models in Jupyter

Ecco

Ecco is a Python library that creates interactive Jupyter notebook visualizations for exploring and explaining pre-trained NLP models, especially Transformer-based language models. It provides views of token predictions, feature attributions, layer-by-layer token evolution and neuron activations.

Independent OpenViST project profile, based on linked sources.

Understand the project

Practical overview

When it fits

Use Ecco in Jupyter notebooks when you want to explore or explain an existing pre-trained Transformer model’s behavior rather than train or fine-tune a model.

Who it is for

NLP researchers exploring how pre-trained models process text.
Model developers investigating predictions, input attribution or internal activations.

Documented uses

Inspect tokenization and candidate output-token probabilities for a model input.
Explore which input words contribute to a model classification using feature attribution.
Trace token rankings across model layers with logit-lens visualizations.
Capture and visualize Transformer feed-forward neuron activations and compare activation spaces.

Before you choose

Requirements & limits

The repository provides installation instructions for pip and conda and identifies the project as BSD-3-Clause licensed.

Ecco runs in Jupyter and is built on PyTorch and Hugging Face Transformers.
The library focuses on exploring pre-trained models; it does not handle training or fine-tuning.
The documentation identifies the project as an alpha release of a research project.
Local models are supported when based on Hugging Face PyTorch models.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install Ecco for notebook-based model exploration

Install Ecco using either pip or conda to prepare for exploring pre-trained language models in Jupyter notebooks.

Before you start

A Python environment with pip or conda available.
Ecco is intended to run inside Jupyter notebooks and is built on PyTorch and Transformers.

Steps

Choose either pip or conda as the installation method.
For pip, run the command below.
Documented command
```
pip install ecco
```
For conda, run conda install -c conda-forge ecco.

Expected outcome

Ecco is installed and ready for notebook-based exploration of existing pre-trained NLP models.

Guide limitations

Ecco is for exploring and understanding existing pre-trained models, not training or fine-tuning.
The project describes itself as an alpha release of a research project.

Documentation 1 ↗

Evidence and transparency

Sources & review

Official repository ↗

Source record details

Source availability: Active
Sources checked: 15 October 2026
Original website title: jalammar/ecco · Official homepage title · retained in its source language

How we review sources Suggest a correction