large model R&D, distributed training inference acceleration, Diffusion Models, deep learning algorithms, video generation, training inference optimization, CPU GPU acceleration