Research map / Image generation / Fast diffusion sampling
MPQ-DMv2: Flexible Residual Mixed Precision Quantization for Low-Bit Diffusion Models With Temporal Distillation
Wei Feng, Chuanguang Yang, Haotong Qin, Yuqi Li, Xiangqi Li, Zhulin An, Libo Huang, Boyu Diao, Fuzhen Zhuang, Michele Magno, Yongjun Xu, Yingli Tian and 1 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add FZRMQ residual quantizer (ablation) | MPQ-DM · LDM-4 ImageNet W2A4 | FID: 36.59 -> 33.58 | Improved |
| add flexible optimization strategies (FOS) to FZRMQ (ablation) | FZRMQ · LDM-4 ImageNet W2A4 | FID: 34.95 -> 33.58 | Improved |
| add MTRD temporal relation distillation (ablation) | MPQ-DM + FZRMQ · LDM-4 ImageNet W2A4 | FID: 33.58 -> 33.02 | Improved |
| add OOLRI SVD-based LoRA initialization (ablation) | MPQ-DM + FZRMQ + MTRD · LDM-4 ImageNet W2A4 | FID: 33.02 -> 32.55 | Improved |
| use KL divergence instead of MSE in MTRD (ablation) | MTRD · LDM-4 ImageNet W2A4 | FID: 33.26 -> 33.02 | Improved |
| vary distillation weight alpha from 0.1 to 2.0 (ablation) | MTRD · LDM-4 ImageNet W2A4 | FID: 33.58 -> 33.02 | Improved |
| allocate extra 10% channels to 2-bit (MPQ-DMv2+) | MPQ-DMv2 · LDM-4 ImageNet W2A4 | FID: 32.55 -> 26.98 | Improved |
| replace MPQ-DM with MPQ-DMv2 | MPQ-DM · LSUN-Churches LDM-8 W3A4 | FID: 14.08 -> 10.58 | Improved |
| replace MPQ-DM with MPQ-DMv2 | MPQ-DM · LSUN-Bedrooms LDM-4 W2A4 | FID: 21.69 -> 21.65 | No clear effect |
| replace MPQ-DM with MPQ-DMv2 | MPQ-DM · DiT-XL/2 ImageNet 512 W3A8 | FID: 60.96 -> 53.63 | Improved |
| replace MPQ-DM with MPQ-DMv2 | MPQ-DM · Stable Diffusion COCO W3A4 | CLIP Score: 26.96 -> 28.28 | Improved |
| add MTRD, FZRMQ, OOLRI together vs MPQ-DM calibration cost | MPQ-DM · LDM-4 ImageNet W2A4 | calibration time (hours): 3.12 -> 3.13 | No clear effect |
Closest papers by meaning
- MPQ-DM: Mixed Precision Quantization for Extremely Low Bit Diffusion Models (AAAI 2025 · similarity 0.94)
- EfficientDM: Efficient Quantization-Aware Fine-Tuning of Low-Bit Diffusion Models (ICLR 2024 · similarity 0.92)
- Qua2SeDiMo: Quantifiable Quantization Sensitivity of Diffusion Models (AAAI 2025 · similarity 0.90)
- TCAQ-DM: Timestep-Channel Adaptive Quantization for Diffusion Models (AAAI 2025 · similarity 0.90)
- Pioneering 4-Bit FP Quantization for Diffusion Models: Mixup-Sign Quantization and Timestep-Aware Fine-Tuning (CVPR 2025 · similarity 0.90)
- SVDQuant: Absorbing Outliers by Low-Rank Component for 4-Bit Diffusion Models (ICLR 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
