/AI_Jokes_With_HuggingFace

Primary LanguagePython

AI Joke Generator

This repository contains a Streamlit application that generates humorous jokes based on news headlines using a Hugging Face model. The application allows users to input a news article and receive a joke in response. Additionally, users can provide feedback, and the system saves the prompts and responses locally.

Installation

Install the required packages:

```sh
pip install -r requirements.txt
```

Usage

  1. Run the Streamlit application:

    streamlit run streamlit.py

  2. Interact with the application:

    • Enter the content of a news article in the provided text area.
    • Click the "Run" button to generate a joke.
    • Review the joke and provide feedback in the feedback text area.
    • Click the "Submit" button to save the prompt, response, and feedback locally.

Fine-Tuning the Model

This repository also includes scripts for fine-tuning the model using PEFT (Parameter-Efficient Fine-Tuning).

  1. Load and preprocess the dataset:

    The dataset is loaded from a local JSON file (responses.json) that contains prompts and their corresponding responses.

  2. Initialize the model and tokenizer:

    The google/flan-t5-base model is used as the base model.

  3. Apply PEFT and train the model:

    The model is fine-tuned using the PEFT configuration.

  4. Save the fine-tuned model:

    The fine-tuned model is saved locally for further use.

Folder Structure

  • streamlit.py: Contains the Streamlit application code.
  • app.py: Contains the core logic for generating jokes using the Hugging Face model.
  • responses.json: Stores the user prompts and responses.
  • requirements.txt: Lists the required Python packages.
  • README.md: This file.

License

This project is licensed under the MIT License.

Acknowledgements

  • Hugging Face for providing the pre-trained models and libraries.
  • Streamlit for the interactive web application framework.

Author