Intel / video

Making Diffusion LLMs Faster: A Survey of Speedup Techniques

Source: youtube.com
Author: Julia TurcTop Viber
Date: Published Feb 9, 2026

Context

Generating several text positions together does not eliminate repeated computation. This diffusion-speed survey explains that, when positions can attend to both earlier and later text, changing one part of a draft can change internal representations elsewhere. Previously calculated values may therefore need updating. The video is a paid collaboration with Inception.

Block diffusion offers a compromise: generate blocks in order, while predicting multiple positions within each block. Completed blocks cannot depend on future ones, so their internal values can be stored and reused. This extends the Transformer explanation: controlling which positions can influence each other determines which computations remain reusable.

Terms in this piece · Glossary

attention — The mechanism that lets a model weigh which earlier words matter for the word it's currently processing — the core operation of a transformer.
transformer — The neural network architecture behind modern AI models, built on attention — letting every word directly consider every other word in parallel.
distillation — Training a small, cheap model to imitate a big one's outputs, keeping much of the capability at a fraction of the cost.
KV cache — The memory a model keeps about text it has already read, so generating each new token doesn't require reprocessing the whole conversation.

Key quotes

“It's a very new technology and approach and I think we're still far from even a local optimum.”
Stefano Ermon

“If you need a lot of diffusion steps, then there is no benefit compared to to one autoregressive model.”
Stefano Ermon

“KV caching as implemented for auto-regressive models is simply not possible for the fusion models.”
Julia Turc

“So, the virus basically spreads to the entire context window.”
Julia Turc

“I see a future where diffusion models will become the leading paradigm not just for image and video generation but also for for discrete object like text and code.”
Stefano Ermon

Chapters

00:00Intro
02:00Auto-regressive vs diffusion LLMs
03:06Reducing refinement steps
05:54Self-Distillation Through Time
07:15Curriculum learning
08:17Speeding up sampling
09:40Confidence scores
11:35Guided diffusion (FlashDLM)
13:24Approximate KV caching (dLLM-Cache, dKV-Cache)
19:03Block diffusion
21:19Where to find diffusion models

Read the source www.youtube.com