eDMD:用於少步擴散蒸餾的端點分佈匹配
eDMD: Endpoint Distribution Matching for Few-Step Diffusion Distillation
eDMD 提出端點分佈匹配蒸餾方法,只以學生模型的最終輸出訓練共享假模型,並加強教師監督,以改善少步擴散模型的蒸餾。實驗顯示,在 SD3-Medium、FLUX.2-klein(4B 和 9B)及 Qwen-Image-2.1 上,eDMD 在 4 步生成中持續勝過現有的分佈匹配方法。
Published on Nov 12
Authors:
,
,
,
,
,
Abstract
Endpoint Distribution Matching Distillation trains a shared fake model only on final student outputs and adds precise teacher supervision to better align few-step diffusion students with teacher distributions for 4-step generation.
Distribution matching distillation (DMD) has emerged as a leading paradigm for distilling diffusion models into few-step students, where an auxiliary fake score model tracks the student's distribution and the discrepancy between the teacher and fake predictions drives the student toward the teacher. In multi-step distillation, although only the student's final output is delivered, existing methods train the fake model on intermediate states or across all denoising steps. Consequently, the student's update signal becomes entangled with intermediate distributions, which can distort the matching direction and remain nonzero even after the student's final outputs already match the teacher distribution. To resolve this mismatch, we propose Endpoint Distribution Matching Distillation (eDMD), which trains a single shared fake model exclusively on the student's final outputs and uses it to correct every denoising step. Our theoretical analysis proves that this endpoint reference guarantees that the distribution matching term vanishes if and only if the student's final output distribution matches the teacher's, regardless of intermediate distributions. Beyond the fake reference, we further strengthen teacher supervision through a precise teacher (PT) that replaces single-step guided predictions with short multi-step trajectories on the final noise interval, paired with a calibrated query density for intermediate steps. Experiments on SD3-Medium, FLUX.2-klein (4B and 9B), and Qwen-Image-2.1 demonstrate that eDMD consistently outperforms existing distribution matching methods for 4-step generation.
View arXiv page View PDF Add to collection
Get this paper in your agent:
hf papers read 2611.17194
Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash
Models citing this paper 1
Text-to-Image • Updated 3 days ago • 3zttzzt/eDMD
Datasets citing this paper 0
No dataset linking this paper
Cite arxiv.org/abs/2611.17194 in a dataset README.md to link it from this page.
Spaces citing this paper 0
No Space linking this paper
Cite arxiv.org/abs/2611.17194 in a Space README.md to link it from this page.
Collections including this paper 0
No Collection including this paper
Add this paper to a collection to link it from this page.
來源:HuggingFace Daily Papers(社區熱門論文) · huggingface.co