Research map / Image generation / Fast diffusion sampling
Distilling Parallel Gradients for Fast ODE Solvers of Diffusion Models
Beier Zhu, Ruoyu Wang, Tong Zhao, Hanwang Zhang, Chi Zhang
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 learnable output scale o_n (ablation) | EPD-Solver · CIFAR-10 (5 Para. NFE) | FID: 5.84 -> 4.33 | Improved |
| add learnable timestep shift delta_n^k (ablation) | EPD-Solver · CIFAR-10 (5 Para. NFE) | FID: 5.47 -> 4.33 | Improved |
| add both o_n and delta_n^k (ablation) | EPD-Solver · CIFAR-10 (3 Para. NFE) | FID: 16.01 -> 10.40 | Improved |
| use time-uniform schedule instead of LogSNR (ablation) | EPD-Solver · CIFAR-10 (5 Para. NFE) | FID: 8.88 -> 4.33 | Improved |
| use time-uniform schedule instead of EDM schedule (ablation) | EPD-Solver · CIFAR-10 (7 Para. NFE) | FID: 4.50 -> 2.82 | Improved |
| use DPM-Solver-2 teacher instead of Heun (ablation) | EPD-Solver · CIFAR-10 (5 Para. NFE) | FID: 6.65 -> 4.33 | Improved |
| use DPM-Solver-2 teacher instead of iPNDM (ablation) | EPD-Solver · CIFAR-10 (3 Para. NFE) | FID: 13.69 -> 10.40 | Improved |
| increase parallel directions K from 1 to 2 (ablation) | EPD-Solver · FFHQ 64x64 (3 Para. NFE) | FID: 26.0 -> 22.7 | Improved |
| increase K from 2 to 3 (ablation) | EPD-Solver · ImageNet 64x64 (9 Para. NFE) | FID: 4.20 -> 4.18 | No clear effect |
| increase K from 1 to 3 (ablation) | EPD-Solver · LSUN Bedroom latency (9 Para. NFE) | latency (ms): 131.1 -> 147.8 | Hurt |
| increase K from 1 to 2 (ablation) | EPD-Solver · CIFAR-10 latency (9 Para. NFE) | latency (ms): 80.5 -> 79.8 | No clear effect |
| replace single learned midpoint (AMED-Solver) with ensemble of parallel gradients | EPD-Solver · LSUN Bedroom 256x256 (3 NFE) | FID: 58.21 -> 13.21 | Improved |
Closest papers by meaning
- Distilling ODE Solvers of Diffusion Models into Smaller Steps (CVPR 2024 · similarity 0.91)
- PFDiff: Training-Free Acceleration of Diffusion Models Combining Past and Future Scores (ICLR 2025 · similarity 0.90)
- TADA: Improved Diffusion Sampling with Training-free Augmented DynAmics (NeurIPS 2025 · similarity 0.90)
- Fast ODE-based Sampling for Diffusion Models in Around 5 Steps (CVPR 2024 · similarity 0.90)
- Adaptive Stochastic Coefficients for Accelerating Diffusion Sampling (NeurIPS 2025 · similarity 0.89)
- Simple and Fast Distillation of Diffusion Models (NeurIPS 2024 · similarity 0.89)
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.
