Research map / Image generation / Fast diffusion sampling
Planned Diffusion
Daniel Mingyi Israel, Tian Jin, Ellie Y Cheng, Guy Van den Broeck, Aditya Grover, Suvinay Subramanian, Michael Carbin
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 |
|---|---|---|---|
| use planned diffusion instead of Fast-dLLM on the diffusion model | Dream-7B-Base · AlpacaEval | LC win rate: 40.2 -> 44.6 with 22.4x speedup | Improved |
| use planned diffusion instead of autoregressive decoding | Dream-7B-Base · AlpacaEval | speedup / LC win rate: 1.27x speedup, win rate 50.0 -> 49.2 | Mixed |
| use block-sparse attention between chunks (PDSA) | Planned diffusion · AlpacaEval | speedup / LC win rate: speedup 1.81x vs AR, win rate 44.6 | Mixed |
| increase fine-tuning epochs from 2 to 16 (ablation) | Planned diffusion · AlpacaEval | LC win rate: 44.9 -> 49.2 | Improved |
| increase fine-tuning epochs from 2 to 16 (ablation) | Autoregressive · AlpacaEval | LC win rate: flat at 50.0 | No clear effect |
| increase fine-tuning epochs from 2 to 16 (ablation) | Diffusion · AlpacaEval | LC win rate: 41.1 -> 52.6 | Improved |
| increase fine-tuning epochs from 2 to 16 (ablation) | PDSA · AlpacaEval | LC win rate: 40.2 -> 43.7 | Improved |
| include topic attribute in the plan (ablation) | Planned diffusion (4 epochs) · AlpacaEval | LC win rate: 23.33 -> 46.65 | Improved |
| include <sync/> tokens in the plan (ablation) | Planned diffusion (4 epochs) · AlpacaEval | LC win rate and latency: without sync: win rate 46.65 -> 42.96, latency 3.23 s -> 5.81 s | Mixed |
| scale predicted chunk length away from 1.0 (0.5 to 2.5) (ablation) | Planned diffusion · AlpacaEval | LC win rate | Hurt |
| lower step ratio and confidence threshold (ablation) | Planned diffusion · AlpacaEval | quality-latency trade-off | Mixed |
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.
- Fast-dLLM — Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decoding (ICLR 2026) · 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
- Breaking AR’s Sampling Bottleneck: Provable Acceleration via Diffusion Language Models (NeurIPS 2025 · similarity 0.91)
- Energy-Based Diffusion Language Models for Text Generation (ICLR 2025 · similarity 0.91)
- Beyond Autoregression: Fast LLMs via Self-Distillation Through Time (ICLR 2025 · similarity 0.90)
- Scaling Diffusion Language Models via Adaptation from Autoregressive Models (ICLR 2025 · similarity 0.90)
- Denoising Time Matters: Diverse Generation in Diffusion Language Models (NeurIPS 2026 · similarity 0.90)
- The Diffusion Duality (ICML 2025 · 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.
