AlloyMoE: GPU Kernels for Mixture of Experts
In Progress
This project is looking at developing performant and memory efficient GPU kernels for MoE modules.
GPU Programming
BERN-NN-IBF: Enhancing Neural Network Bound Propagation Through Implicit Bernstein Form and Optimized Tensor Operations
IEEE TCAD 2024
This paper explores methods to perform efficient bounds-propagation on neural networks.