Researchers have developed OrbitQuant, a novel method for post-training quantization of diffusion transformers (DiTs). This technique allows for efficient inference by quantizing in a normalized, rotated basis, eliminating the need for recalibration across different timesteps, prompts, or modalities. OrbitQuant achieves state-of-the-art performance in low-bit settings for image and video generation models like FLUX.1, Z-Image-Turbo, Wan 2.1, and CogVideoX, even enabling usable generation quality at 2-bit weights. AI
IMPACT Reduces computational cost for diffusion model inference, potentially enabling wider deployment on resource-constrained devices.
RANK_REASON Research paper detailing a new method for model quantization. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
