Session 7 // Feb 2: Planned Diffusion
Diffusion LLM
Hello folks,
Diffusion language models are capable of parallelizing text generation but can struggle with coherence in low time-step regimes. Planned Diffusion unlocks a new axis of parallelism by moving from token-level parallelism to semantic parallelism. Planned diffusion first generates a structured plan, then diffuses semantically independent spans of text in parallel according to the plan.
This Monday, Daniel Israel (UCLA) and Tian Jin (MIT) will discuss their exciting Planned Diffusion paper as the joint first authors.
Title: Planned Diffusion
Meeting Link: click here
Time: Feb 2 (Monday) 1pm ET / 10am PT / 7pm CET / 11:30pm IST
Paper: https://arxiv.org/abs/2510.18087
Prior knowledge:
Fundamentals of discrete diffusion (video by Sasha Rush)
Abstract: A central challenge in large language model inference is the trade-off between generation speed and output quality. Autoregressive models produce high-quality text but generate tokens sequentially. Diffusion models can generate tokens in parallel but often need many iterations to match the same quality. We propose planned diffusion, a hybrid method that combines the strengths of both paradigms. Planned diffusion works in two stages: first, the model creates a short autoregressive plan that breaks the output into smaller, independent spans. Second, the model generates these spans simultaneously using diffusion. This approach expands the speed-quality Pareto frontier and provides a practical path to faster, high-quality text generation. On AlpacaEval, a suite of 805 instruction-following prompts, planned diffusion achieves Pareto-optimal trade-off between quality and latency, achieving 1.27x to 1.81x speedup over autoregressive generation with only 0.87% to 5.4% drop in win rate, respectively. Our sensitivity analysis shows that the planning mechanism of planned diffusion is minimal and reliable, and simple runtime knobs exist to provide flexible control of the quality-latency trade-off.
Yours truly,
Subham, Justin, Zhihan
Website, Twitter, Discord, YouTube