ViDiT-Q: Efficient and Accurate Quantization of Diffusion Transformers for Image and Video Generation | Synapse