k2t

Introduction

Keytotext is a model designed to convert keywords into coherent sentences. It leverages the capabilities of the T5 model and is available through the Hugging Face platform. The model is trained using datasets like WebNLG and Dart, and its performance is evaluated using natural language generation (NLG) metrics.

Architecture

Keytotext is built on the T5 (Text-to-Text Transfer Transformer) architecture. The model comes in different sizes, including k2t, k2t-tiny, and k2t-base, each tailored for various computational and application needs.

Training

The model has been trained on datasets such as WebNLG and Dart, known for their applicability in text generation tasks. Training notebooks are available for those interested in understanding the training process or customizing it further. These resources can be accessed in the Training Notebooks directory.

Guide: Running Locally

1. Installation:

   • Install the keytotext package via pip:

     pip install keytotext
     

   • To run the Streamlit UI, install streamlit-tags:

     pip install streamlit-tags
     

2. Usage:

   • Example notebooks can be explored via Google Colab or locally.
   • The Streamlit app provides an interactive UI to test the model.

3. Hardware Recommendations:

   • For optimal performance, especially with larger models, utilizing cloud GPUs like those offered by AWS, GCP, or Azure is recommended.

License

Keytotext is released under the MIT License, allowing for broad use and modification by the community.