Trendoday

MMFlash: Faster Speculative Decoding for Vision-Language Models

Speculative decoding accelerates generation by letting a lightweight drafter propose tokens for parallel verification by a target model. Diffusion drafters such as DFlash propose an entire token block

  • Upvotes21
–Rank now
#5Peak rank
2Days on list
345Score · 7d
Nov 16, 2026First seen
Rank · 7 days

About this paper

MMFlash improves multimodal speculative decoding by adaptively upweighting visually dependent tokens and using sparse attention for long video, yielding up to 3.58× lossless speedup.

Speculative decoding accelerates generation by letting a lightweight drafter propose tokens for parallel verification by a target model. Diffusion drafters such as DFlash propose an entire token block in a single forward pass, but mainly focus on text-only tasks, leaving multimodal drafting underexplored. To study this setting, we first introduce a metric of token-level visual dependence based on the KL divergence between next-token distributions with and without visual input. We find that acceptance rates decrease from 46% in the lowest visual-dependence quartile to 22% in the highest, with this disparity persisting after multimodal training. Motivated by this observation, we propose…

speculative decodinglightweight draftertarget modelparallel verificationdiffusion draftersDFlashtoken blocksingle forward passtext-only tasksmultimodal drafting
Upvotes
21
Authors
5
Published
Nov 13, 2026

Summary and abstract excerpt via Hugging Face Papers ↗ · arXiv.

Trending history

MMFlash: Faster Speculative Decoding for Vision-Language Models was first seen on Hugging Face · Daily papers on Nov 16, 2026. Since then it has appeared in 15 checks on 2 different days. Its best position was #5 of 30, reached on Nov 16, 15:28 UTC. It dropped off the list after Nov 17, 11:28 UTC.

Rank history

Position on Hugging Face · Daily papers at every check (lower is better).

#1#11#20#30 07:2814:2821:2804:2811:28 Nov 16, 07:28 UTC · #12Nov 16, 09:28 UTC · #10Nov 16, 11:28 UTC · #8Nov 16, 13:28 UTC · #7Nov 16, 15:28 UTC · #5Nov 16, 17:28 UTC · #5Nov 16, 19:28 UTC · #6Nov 16, 21:28 UTC · #6Nov 16, 23:28 UTC · #6Nov 17, 01:28 UTC · #5Nov 17, 03:28 UTC · #5Nov 17, 05:28 UTC · #5Nov 17, 07:28 UTC · #6Nov 17, 09:28 UTC · #5Nov 17, 11:28 UTC · #7

Last 14 days

Best rank per day (UTC).

Nov 9today

Metric history

Platform metric as reported by Hugging Face.

0122436 07:2814:2821:2804:2811:28 Nov 16, 07:28 UTC · 21Nov 16, 09:28 UTC · 25Nov 16, 11:28 UTC · 30Nov 16, 13:28 UTC · 33Nov 16, 15:28 UTC · 35Nov 16, 17:28 UTC · 35Nov 16, 19:28 UTC · 35Nov 16, 21:28 UTC · 35Nov 16, 23:28 UTC · 35Nov 17, 01:28 UTC · 35Nov 17, 03:28 UTC · 36Nov 17, 05:28 UTC · 36Nov 17, 07:28 UTC · 35Nov 17, 09:28 UTC · 35Nov 17, 11:28 UTC · 36

Questions

Is MMFlash: Faster Speculative Decoding for Vision-Language Models trending on HF Daily Papers right now?

Not at the moment. It was last on the list on Nov 17, 11:28 UTC, after 2 days of trending.

What is the highest rank MMFlash: Faster Speculative Decoding for Vision-Language Models reached?

#5 on Hugging Face · Daily papers, on Nov 16, 15:28 UTC. Trendoday checks this list every 120 minutes.

How long has MMFlash: Faster Speculative Decoding for Vision-Language Models been trending?

It was first seen on Nov 16, 2026 and has been on the list on 2 days (15 checks) since.