OrbitQuant: Data-Agnostic Quantization for Image and Video Diffusion Transformers
Abstract
OrbitQuant enables efficient post-training quantization for diffusion transformers by using a normalized rotated basis that eliminates the need for recalibration across different timesteps and modalities.
Diffusion transformers (DiTs) achieve state-of-the-art image and video generation, but their multi-step sampling and growing parameter count make inference expensive. Post-training quantization (PTQ) is the natural remedy, yet DiT activations shift across timesteps, prompts, and guidance branches, forcing prior methods to re-fit calibration data for every new checkpoint or modality. We present OrbitQuant, a data-agnostic weight-activation quantizer that bypasses range estimation by quantizing in a normalized, rotated basis. In this basis, a randomized permuted block-Hadamard (RPBH) rotation concentrates each coordinate around one fixed, known marginal regardless of the input, so a single Lloyd-Max codebook serves all timesteps, prompts, and layers of a given input dimension. We extend the same quantizer to weight rows offline, absorbing the rotation into the weights so that it cancels inside each linear layer and only a forward rotation on the activations remains at runtime. The same recipe transfers from image to video with no per-modality tuning. Across FLUX.1, Z-Image-Turbo, Wan 2.1, and CogVideoX, it sets the state of the art for PTQ at several low-bit settings. It also pushes PTQ of image diffusion transformers to W2A4 with usable generation quality.
Community
OrbitQuant is a data-agnostic weight–activation quantizer for image and video diffusion transformers that uses no calibration data. DiT activations drift across timesteps, prompts, and guidance branches, which forces prior PTQ methods to re-fit calibration for every new checkpoint or modality. OrbitQuant instead quantizes in a normalized, rotated basis: a randomized permuted block-Hadamard (RPBH) rotation concentrates each coordinate around one fixed, known marginal regardless of input, so a single Lloyd–Max codebook serves all timesteps, prompts, and layers. The rotation is folded into the weights offline and cancels inside each linear layer, leaving only a cheap forward rotation on activations at runtime and the same recipe transfers from image to video with no per-modality tuning. Across FLUX.1, Z-Image-Turbo, Wan 2.1, and CogVideoX it sets the state of the art at several low-bit settings. It can produce usable images at 2-bit weights (W2A4) where prior approaches collapse.
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- DiRotQ: Rotation-Aware Quantization for 4-bit Diffusion Transformers (2026)
- Timestep-Aware SVDQuant-GPTQ for W4A4 Quantization of Wan2.2-I2V (2026)
- OSCAR: Offline Spectral Covariance-Aware Rotation for 2-bit KV Cache Quantization (2026)
- Tail-Aware HiFloat4: W4A4 Post-Training Quantization for Wan2.2 (2026)
- PermuQuant: Lowering Per-Group Quantization Error by Reordering Channels for Diffusion Models (2026)
- OCTOPUS: Optimized KV Cache for Transformers via Octahedral Parametrization Under optimal Squared error quantization (2026)
- DynamicPTQ: Mitigating Activation Quantization Collapse via Residual-Stream Dynamics (2026)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: @librarian-bot recommend
Get this paper in your agent:
hf papers read 2607.02461 Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper