From c386136200e6c1fd43df40bcfe17d12f49ab79e3 Mon Sep 17 00:00:00 2001 From: Mengni Wang Date: Thu, 30 Oct 2025 05:29:16 -0400 Subject: [PATCH] enhance flux doc --- auto_round/compressors/diffusion/README.md | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) diff --git a/auto_round/compressors/diffusion/README.md b/auto_round/compressors/diffusion/README.md index ca9adb93f..907bbaacb 100644 --- a/auto_round/compressors/diffusion/README.md +++ b/auto_round/compressors/diffusion/README.md @@ -4,9 +4,15 @@ This feature is experimental and may be subject to changes, including potential ## Quantization +Quantization for diffusion models is limited: + +1. Only transformer module of diffusion models will be quantized.. +2. Loading quantized model is not supported yet, so please use `fake` format for quantization. +3. Calibration dataset only supports `coco2014` and user customized `.tsv` file. + + ### API Usage (CPU/GPU) Recommended -By default, AutoRoundDiffusion only quantizes the transformer module of diffusion models and uses `COCO2014 captions` for calibration. ```python import torch