BertViz is a tool for visualizing attention in the Transformer model, supporting all models from the transformers library (BERT, GPT-2, XLNet, RoBERTa, XLM, CTRL, etc.). It extends the Tensor2Tensor visualization tool by Llion Jones and the transformers library from HuggingFace.

Blog post:

• Deconstructing BERT, Part 2: Visualizing the Inner Workings of Attention (Part 1 is not a prerequisite)

Paper:

• A Multiscale Visualization of Attention in the Transformer Model

Related blog posts:

• OpenAI GPT-2: Understanding Language Generation through Visualization
• Deconstructing BERT: Distilling 6 Patterns from 100 Million Parameters

The attention-head view visualizes the attention patterns produced by one or more attention heads in a given transformer layer.

The attention view supports all models from the Transformers library, including:
BERT: [Notebook] [Colab]
GPT-2: [Notebook] [Colab]
XLNet: [Notebook]
RoBERTa: [Notebook]
XLM: [Notebook]
Albert: [Notebook]
DistilBert: [Notebook]
(and others)

The model view provides a birds-eye view of attention across all of the model’s layers and heads.

The model view supports all models from the Transformers library, including:
BERT: [Notebook] [Colab]
GPT2: [Notebook] [Colab]
XLNet: [Notebook]
RoBERTa: [Notebook]
XLM: [Notebook]
Albert: [Notebook]
DistilBert: [Notebook]
(and others)

The neuron view visualizes the individual neurons in the query and key vectors and shows how they are used to compute attention.

The neuron view supports the following three models:
BERT: [Notebook] [Colab (view only)]
GPT-2 [Notebook] [Colab (view only)]
RoBERTa [Notebook]

• Transformers (version required depends on models used)
• PyTorch >=1.0.0
• Jupyter
• tqdm
• boto3
• IPython
• requests
• regex
• sentencepiece

(See requirements.txt)

git clone https://github.com/jessevig/bertviz.git
cd bertviz
jupyter notebook

Click on any of the sample notebooks. Note that the sample notebooks do not cover all Huggingface models, but the code should be similar for those not included.

Click on any of the Colab links above, and scroll to the bottom of the page. It should be pre-loaded with the visualization, so you don't need to actually run anything.

If you write your own code for executing BertViz in Colab, note that some of the steps are different from those in the Jupyter notebooks (see Colab examples above).

The visualizations works best with shorter sentences and may fail if the input text is very long. The tool is designed such that only one visualization should be included per notebook.

Jesse Vig

When referencing BertViz, please cite this paper.

@article{vig2019transformervis,
  author    = {Jesse Vig},
  title     = {A Multiscale Visualization of Attention in the Transformer Model},
  journal   = {arXiv preprint arXiv:1906.05714},
  year      = {2019},
  url       = {https://arxiv.org/abs/1906.05714}
}

This project is licensed under the Apache 2.0 License - see the LICENSE file for details

This project incorporates code from the following repos:

• https://github.com/tensorflow/tensor2tensor
• https://github.com/huggingface/pytorch-pretrained-BERT