OrbitQuant enables data-agnostic quantization for image and video diffusion transformers

OrbitQuant is a data-agnostic weight-activation quantizer designed to reduce inference costs for diffusion transformers (DiTs) by bypassing the need for per-checkpoint calibration data. It quantizes in a normalized, rotated basis using randomized permuted block-Hadamard rotation, allowing a single Lloyd-Max codebook to serve all timesteps, prompts, and layers.

• The method extends to weight rows offline, absorbing the rotation into weights so only a forward rotation on activations remains at runtime.
• It transfers from image to video generation with no per-modality tuning required.
• OrbitQuant sets the state of the art for post-training quantization (PTQ) across FLUX.1, Z-Image-Turbo, Wan 2.1, and CogVideoX.
• It pushes PTQ of image diffusion transformers to W2A4 while maintaining usable generation quality.

This approach provides a unified solution for low-bit quantization across different modalities without the overhead of re-fitting calibration data for every new checkpoint or input type.