Research map / Image generation / Fast diffusion sampling
Robust Parallel Diffusion Sampling via Dynamic Jacobian Bandwidth
Zile Huang, Ser-Nam Lim
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 adaptive Jacobian bandwidth control to sequential sampling (ablation) | ROPA · HunyuanVideo VBench (4 cores) | speedup: 1.5 alone vs 2.1 full | Improved |
| add score-aligned low-rank preconditioning (ablation) | ROPA · HunyuanVideo VBench (4 cores) | speedup: 1.4 alone vs 2.1 full | Improved |
| combine ROPA with SpargeAttention (ablation) | ROPA · HunyuanVideo VBench (4 cores) | VBench quality: 73.6 -> 72.8 (speedup 2.1 -> 2.4) | Mixed |
| combine ROPA with ToCa token caching (ablation) | ROPA · HunyuanVideo VBench (4 cores) | speedup: 2.1 -> 2.3 | Improved |
| use ROPA instead of ParaDIGMS (Picard iteration) | HunyuanVideo (8 cores) · HunyuanVideo VBench | latent RMSE: 0.189 -> 0.055 | Improved |
| use ROPA instead of ParaSolver at 8 cores | HunyuanVideo · HunyuanVideo VBench | speedup: 2.3 -> 2.9 | Improved |
| use ROPA instead of ParaSolver at 4 cores on SD-3.5-Large | SD-3.5-Large · COCO2017 1000 prompts | speedup: 1.1 -> 1.8 | Improved |
| use ROPA instead of sequential sampling | HunyuanVideo · VBench | VBench quality: 73.8% -> 73.6% | No clear effect |
| adaptive bandwidth widens late in denoising (stiff region) | ROPA · FLUX | fraction of steps with w_i >= 5: 0% early -> 85% late stiff | Improved |
| use ROPA instead of ParaSolver at 8 cores on SD-3.5-Large | SD-3.5-Large · COCO2017 1000 prompts | latent RMSE: 0.324 -> 0.224 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CHORDS — CHORDS: Diffusion Sampling Accelerator with Multi-core Hierarchical ODE Solvers (ICCV 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Boosting Diffusion Models with Moving Average Sampling in Frequency Domain (CVPR 2024 · similarity 0.90)
- TADA: Improved Diffusion Sampling with Training-free Augmented DynAmics (NeurIPS 2025 · similarity 0.90)
- A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models (ICML 2024 · similarity 0.89)
- Accelerating Diffusion Models with Parallel Sampling: Inference at Sub-Linear Time Complexity (NeurIPS 2024 · similarity 0.89)
- Accelerating Parallel Sampling of Diffusion Models (ICML 2024 · similarity 0.89)
- AsyncDiff: Parallelizing Diffusion Models by Asynchronous Denoising (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.
