README.md

metadata

title: Adaptive T5 Summarization
emoji: 📝
colorFrom: blue
colorTo: indigo
sdk: gradio
app_file: app.py

Adaptive T5 Summarization

This project builds a meta-model on top of T5 to adapt model selection based on text complexity.

Installation

To use this project, clone the repository and install the required dependencies:

git clone https://huggingface.co/spaces/tlmk22/OptimAbstract
cd OptimAbstract
pip install -r requirements.txt
python -m spacy download en_core_web_lg

Repository Structure

.
├── demo.ipynb                   #  Jupyter notebook demonstrating the usage.
├── model.py                    # Contains the T5 models and the meta-model implementation.
├── requirements.txt                     # Lists the required Python packages.
└── README.md

Usage

What the meta model does:

• Extract complexity-based features from input texts.
• Apply multiple T5-based summarization models.
• Use BertScore to determine the best-performing model for each text.
• Train a classifier to predict the best model based on extracted features.
• At inference, the classifier selects the appropriate model dynamically.

TODO

The results are not satisfactory. Improvements should be made:

• Reduce the tolerance (the small model is nearly always used).
• Include feature computation time in the meta-model cost analysis.
• Analyze feature relevance and classifier performance more deeply.
• Modify the MetaModel structure since it is too large to commit (4GB).

This is an ongoing project, and contributions or feedback are welcome!