This repository contains implementations of prompt generation using the Qlora method and the PEFT (Parameter-Efficient Fine-Tuning) approach. The project leverages the TinyLlama model and utilizes the BitsAndBytesConfig for loading the model in 8-bit configuration, ensuring efficient computation. It includes notebooks for both Kaggle and Google Colab, showcasing model performance and prompt generation capabilities.
- Efficient prompt generation utilizing the Qlora method
- Implemented in both Kaggle and Google Colab notebooks
- Demonstrates model performance metrics using TensorBoard
- Easy-to-use interface for generating prompts based on user-defined titles
- Model: The project uses the TinyLlama model: TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T from Hugging Face.
- Dataset: The dataset used for training and prompt generation is fka/awesome-chatgpt-prompts from Hugging Face.
To run the notebooks, ensure you have the following libraries installed. A requirements.txt file is included for easy installation:
pip install -r requirements.txtThe trained model is saved in zip format for easy access and deployment. Ensure to unzip the model before use. The model is configured with BitsAndBytesConfig to load it in 8-bit format for optimized performance.
The notebooks in this repository demonstrate how to generate prompts based on user-defined titles. Both Kaggle and Google Colab notebooks are included for ease of access.
- Kaggle Notebook: Open Kaggle Notebook
- Google Colab Notebook: Open Google Colab Notebook
Performance metrics of the model can be found within the Performance Directory, with visualizations provided through TensorBoard.
You can view the results generated with new data in the Results Directory.
Contributions are welcome! If you'd like to contribute to this project, please fork the repository and submit a pull request.
This project is licensed under the Apache 2.0 License. See the LICENSE file for details.