Cache-DiT is built on top of the Hugging Face Diffusers library and supports nearly all Diffusion Transformers from Diffusers including Flux. It provides hybrid cache acceleration including DBCache, TaylorSeer, and SCM techniques, along with comprehensive parallelism optimizations such as Context Parallelism, Tensor Parallelism, and hybrid 2D/3D parallelism. The engine also supports quantization and compilation for further performance gains, with dedicated parallelism support for Text Encoders, VAE, and ControlNet components.