Inference Kernel Optimization

Low-level code acceleration techniques targeting the forward pass of deep learning models to increase token throughput per unit