visuallm 0.3.0

Visualization tool for various generation tasks on Language Models.

Conditional Language Model Generation Visualization

• when evaluating language models it is often pain to see what is generated and why
• this little package is a vue.js frontend together with flask backend and it is designed to easily show some interesting visualizations on conditional generation models
• it handles frontend-backend communication as well as frontend rendering
• hence the developper can focus only on ML aspects of his work!

VERSION: 0.3.0 changelog

• added visuallm.components.ChatComponent, so you can now chat with the models!
• added support for OpenAI API, so providing the token, you can chat with the OpenAI models!
• refactoring and better code quality
• more documentation strings

Installation

• install from pypi:
  • pip install visuallm

Usage

The documentation is a WIP as of now, however here you can see several snippets of what the library can do.

If you want to use the app with the personachat dataset, you can play with prepared example by running:

flask --app examples_py.persona_chat_example.app run

Generation Playground

Select which parameters you want to use for generation, plug in a HuggingFace model, or an OpenAI token and have fun with experimenting with various generation hyperparameters!

Chat Playground

Select which parameters you want to use for generation, plug in a HuggingFace model, or an OpenAI token and have fun with chatting with the model!

Visualize Next Token Predictions

By using visuallm.components.NextTokenPredictionComponent.NextTokenPredictionComponent you can just plug the HuggingFace model in and go through the generation process step by step.