← 研踪 PaperTrailarXiv:2609.33496 · v2
Chameleon: Dynamic Format Adapter for Efficient Diffusion
首次提交 2026-09-27 20:04
版本更新 2026-11-16 21:20
雷达观测 2026-11-17 17:11
摘要
Post-training quantization (PTQ) is the standard way to run modern diffusion models on memory-constrained accelerators, yet every existing diffusion PTQ scheme fixes the in advance and only tunes the scale, zero point, or per-layer bit-width. At a fixed bit-width the best format depends on the distribution being encoded, and that distribution differs across weight channels, across layers, and along the diffusion timestep, where activation distributions slide from heavy-tailed and noise-dominated to tightly clustered and structured. We propose Chameleon, a PTQ framework that holds the bit-width fixed and treats the format itself as a discrete variable, chosen per weight channel and per activation layer, optionally refined per timestep bucket. Activation formats come from {INT8, FP8 E4M3, FP8 E5M2, MXFP8, MXINT8}, selected ahead of time from two cheap statistics (empirical kurtosis and the closed-form diffusion SNR) and stored in a lookup table; weight formats come from {INT8, MXINT8} at 8 bits or {INT4, NF4, FP4 E2M1, MXINT4, MXFP4} at 4 bits, selected offline by reconstruction error. An architectural fork adapts the same selection layer to multi-step UNets, single-step distilled models, and Diffusion Transformers. Most of the gain comes from the per-layer choice; the timestep refinement helps mainly at 6-bit activations. Across SDXL, SDXL-Turbo, and PixArt- on COCO-2014, Chameleon achieves the best FID in all six backbone bit-width settings, with CLIP within 0.24 of the FP16 reference and the best of all quantized methods at . On PixArt-, a real mixed-format kernel runs 1.09 faster than FP16 end-to-end with 3 smaller weights.
作者备注:Under Review
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v2 · 2026-11-16 21:20 · 最新观测版本
Chameleon: Dynamic Format Adapter for Efficient Diffusion
Post-training quantization (PTQ) is the standard way to run modern diffusion models on memory-constrained accelerators, yet every existing diffusion PTQ scheme fixes the in advance and only tunes the scale, zero point, or per-layer bit-width. At a fixed bit-width the best format depends on the distribution being encoded, and that distribution differs across weight channels, across layers, and along the diffusion timestep, where activation distributions slide from heavy-tailed and noise-dominated to tightly clustered and structured. We propose Chameleon, a PTQ framework that holds the bit-width fixed and treats the format itself as a discrete variable, chosen per weight channel and per activation layer, optionally refined per timestep bucket. Activation formats come from {INT8, FP8 E4M3, FP8 E5M2, MXFP8, MXINT8}, selected ahead of time from two cheap statistics (empirical kurtosis and the closed-form diffusion SNR) and stored in a lookup table; weight formats come from {INT8, MXINT8} at 8 bits or {INT4, NF4, FP4 E2M1, MXINT4, MXFP4} at 4 bits, selected offline by reconstruction error. An architectural fork adapts the same selection layer to multi-step UNets, single-step distilled models, and Diffusion Transformers. Most of the gain comes from the per-layer choice; the timestep refinement helps mainly at 6-bit activations. Across SDXL, SDXL-Turbo, and PixArt- on COCO-2014, Chameleon achieves the best FID in all six backbone bit-width settings, with CLIP within 0.24 of the FP16 reference and the best of all quantized methods at . On PixArt-, a real mixed-format kernel runs 1.09 faster than FP16 end-to-end with 3 smaller weights.
查看此版本解读 →查看此版本原文 ↗