Optimizing Transformer-Based Diffusion Models for Video Generation with NVIDIA TensorRT