Jessedbrown/fp16_image_lora_loader

FP16 Image and LoRA Loader

This package provides separate classes for loading images in FP16 format and applying LoRA weights in FP16 to PyTorch models. It is designed for optimized performance when working with large images and deep learning models, especially on systems with limited memory.

Features

• FP16 Image Loader: Load images as FP16 tensors compatible with PyTorch, reducing memory usage and improving processing speed.
• LoRA Loader: Load LoRA (Low-Rank Adaptation) weights in FP16 format and apply them to PyTorch models, useful for fine-tuning large models efficiently.

Installation

To use this package, you need Python 3.7 or higher and pip installed on your system.

1. Clone or Download the Repository

   Clone this repository or download the code files to your local machine.

   git clone https://github.com/JesseBrown1980/fp16_image_lora_loader.git
   cd fp16_image_lora_loader
   

Build the Package

Open a terminal in the fp16_image_lora_loader directory and run the following command to build the package:

bash

python setup.py sdist bdist_wheel Install the Package Locally

After building, you can install the package by running:

bash

pip install . This will install the package and its dependencies (torch, opencv-python, numpy, and safetensors).

Usage Once installed, you can import and use the FP16ImageLoader and LoRALoader classes in your Python code.

Loading an FP16 Image Use the FP16ImageLoader to load an image in FP16 format as a PyTorch tensor:

python

from fp16_image_lora_loader import FP16ImageLoader

Initialize the image loader

image_loader = FP16ImageLoader()

Load an image as an FP16 tensor

tensor_image = image_loader.load_fp16_image("path/to/your/image.png")

tensor_image is now a PyTorch tensor in FP16 format

Make sure that the image path (path/to/your/image.png) is valid.

Loading LoRA Weights Use the LoRALoader to load LoRA weights in FP16 format into a PyTorch model:

python

from fp16_image_lora_loader import LoRALoader

Initialize the LoRA loader

lora_loader = LoRALoader()

Load the LoRA weights into the model

model = ... # replace with your PyTorch model instance lora_loader.load_lora_weights(model, "path/to/lora_weights.safetensors")

The model now has LoRA weights applied in FP16 format

Ensure that the path to the LoRA weights file (path/to/lora_weights.safetensors) is valid.

Requirements Python 3.7 or higher torch (PyTorch) opencv-python numpy safetensors These dependencies are listed in the setup.py file and will be installed automatically.

Project Structure fp16_image_loader.py: Contains the FP16ImageLoader class for loading images in FP16 format. lora_loader.py: Contains the LoRALoader class for loading LoRA weights in FP16 format. init.py: Imports both classes to make them accessible at the package level. License This project is licensed under the MIT License. See the LICENSE file for details.

Contributing If you would like to contribute to this project, please:

Fork the repository on GitHub. Clone your fork and create a new branch for your feature or bug fix. Make your changes and commit them with descriptive messages. Push your changes to your fork on GitHub. Submit a pull request to the original repository. For major changes, please open an issue to discuss what you would like to change.

Contact For questions or further information, please contact:

Author: Jesse Daniel Brown Email: plasmatoid@gmail.com GitHub: JesseBrown1980 LinkedIn: https://www.linkedin.com/in/jessedanielbrown1980/ ResearchGate: https://www.researchgate.net/profile/Jesse-Brown-21