Introduction

Needle (Necessary Elements of Deep Learning) is a lightweight deep learning framework designed by CMU for educational purposes. It provides essential components of deep learning frameworks, including modular programming abstraction, dynamic computational graphs, automatic differentiation, and tensor acceleration backends (CUDA). The link below features our final project for the course 10714 Deep Learning System where we implemented a Deformable ViT in Needle.

For more details about the project and to run the code, refer to the Github Link. Additionally, you can watch the presentation on YouTube.