Skip to content

This repository contains an implementation of a text summarization model using the T5 (Text-To-Text Transfer Transformer) architecture. The goal of this project is to build a system that can generate concise summaries from lengthy news articles, specifically using the CNN/DailyMail dataset.

Notifications You must be signed in to change notification settings

AliakbarMehdizadeh/Transformers-Summarizer-T5-CNN

Repository files navigation

Text Summarizer Using T5 Fine Tuned on CNN/DailyMail Dataset

This repository contains an implementation of a text summarization model based on the T5 (Text-To-Text Transfer Transformer) architecture. The primary goal of this project is to develop an efficient system capable of generating concise, human-readable summaries from lengthy news articles, making it easier to digest key information. This project specifically utilizes the CNN/DailyMail dataset, which is widely used for training models on the task of summarizing long-form text into brief highlights.

Key Features:

  1. Model Architecture: This project leverages the T5 model, a transformer-based architecture designed for various text generation tasks, including summarization. The model is trained to convert a long text input (news article) into a concise output (summary).

  2. Dataset: The model is trained on the CNN/DailyMail dataset, which consists of news articles and their respective highlights, making it an ideal dataset for summarization tasks.

  3. API Integration: An API is created for real-world use, allowing users to send documents or text inputs and receive summarized versions. This API is built using Flask (or FastAPI/Streamlit depending on your preference) for easy integration into applications.

    Sample Output

Usage

  1. Clone the repository:
  2. Create and activate a virtual environment
  3. pip install -r requirements.txt
  4. python main.py for training and saving the fine tuned model
  5. Start the FastAPI server: uvicorn app:app --reload
  6. Start the Streamlit app in a new terminal: streamlit run streamlit_app.py
  7. Open your browser and go to http://localhost:8501 to access the Streamlit app.

About

This repository contains an implementation of a text summarization model using the T5 (Text-To-Text Transfer Transformer) architecture. The goal of this project is to build a system that can generate concise summaries from lengthy news articles, specifically using the CNN/DailyMail dataset.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published