Newshot
Papers Still developing Event #01M7J58K978DNJNYFNCC5HW4EN

MMFlash: Faster Speculative Decoding for Vision-Language Models

Event overview(model summary, not the source's own text)

Three arXiv papers posted in November 2026 address efficiency in speculative decoding. MMFlash (arXiv:2611.18482v1) presents a method for faster speculative decoding in vision-language models, where a lightweight drafter proposes tokens for parallel verification by a target model. VisHead-Spec (arXiv:2611.21839v1) proposes head-routed visual attention for multimodal speculative decoding, noting that speedup diminishes under certain conditions. SpecSearch (arXiv:2611.22296v1) targets hardware-aware efficient speculative decoding, arguing that optimal configuration depends jointly on workload and hardware. All three are new-announcement preprint abstracts and report no deployment or benchmark results in the supplied text.

This summary was distilled by a model from the 3 reports below; it is not the source text. Treat the reports themselves as the facts.

How this story unfolded

Left: what it concluded — the key points the model distilled from the reports below, each one listing the reports it rests on. Right: what happened — ordered by time, each with a short summary so you can follow the thread without clicking through. The numbered links work both ways: tap a source under a point to jump to that report, tap the cited-by chip on a report to jump back to the point.

Key points7 items Model summary · ordered by when the evidence was published

  1. Point 01 11/16, 13:00

    MMFlash is posted as arXiv:2611.18482v1 under the new-announcement category.

  2. Point 02 11/16, 13:00

    MMFlash speeds generation by having a lightweight drafter propose tokens for parallel verification by a target model.

  3. Point 03 11/18, 13:00

    VisHead-Spec is posted as arXiv:2611.21839v1 and targets multimodal speculative decoding for VLMs.

  4. Point 04 11/18, 13:00

    VisHead-Spec proposes head-routed visual attention and states that speculative decoding preserves the target model's outputs.

  5. Point 05 11/18, 13:00

    VisHead-Spec notes that the speedup from speculative decoding diminishes.

  6. Point 06 11/18, 13:00

    SpecSearch is posted as arXiv:2611.22296v1 and focuses on hardware-aware efficient speculative decoding.

  7. Point 07 11/18, 13:00

    SpecSearch argues that the optimal speculative decoding configuration depends jointly on workload and hardware.

ReportsBy publish time · newer further down

  1. November 16, 2026 · Mon
  2. Start

    MMFlash: Faster Speculative Decoding for Vision-Language Models

    Original summary arXiv:2611.18482v1 Announce Type: new Abstract: Speculative decoding accelerates generation by letting a lightweight drafter propose tokens for parallel verification by a target mo

    arXiv cs.CV Original ↗
    Cited by Point 01 Point 02 Cited
  3. November 18, 2026 · Wed
  4. +2d 0h

    VisHead-Spec: Head-Routed Visual Attention for Multimodal Speculative Decoding

    Original summary arXiv:2611.21839v1 Announce Type: new Abstract: Speculative decoding accelerates vision-language models(VLMs) while preserving the target model's outputs, but its speedup diminishe

    arXiv cs.LG Original ↗
  5. +2d 0h

    SpecSearch: Towards Hardware-Aware Efficient Speculative Decoding

    Original summary arXiv:2611.22296v1 Announce Type: new Abstract: Speculative decoding (SD) accelerates large language model (LLM) inference, but its optimal configuration depends jointly on workloa

    arXiv cs.AI Original ↗
    Cited by Point 06 Point 07 Cited
  6. Still developing · later reports will keep extending this axis

Related reading

VLN-Flash: Parallel Action Decoding with Adaptive Execution for Vision-and-Language Navigation
arXiv:2611.19287v1 Announce Type: new Abstract: When a vision-and-language navigation policy generates an action sequenc
arXiv cs.RO 11/18/2026 Full text here
72pts
Stochastic Self-Distillation for Flow-Based Vision-Language-Action Models
arXiv:2611.20460v1 Announce Type: new Abstract: Flow-based vision-language-action (VLA) models achieve strong performanc
arXiv cs.RO 11/18/2026 Full text here
77pts
Scaling Laws for Vision-Language Fusion
arXiv:2611.22351v1 Announce Type: new Abstract: Vision-language models (VLMs) have emerged as central foundation models
arXiv cs.LG 11/18/2026 Full text here
82pts
PlanDiv: Test-Time Scaling with Latent Plans for Vision-Language-Action Models
arXiv:2611.20010v1 Announce Type: new Abstract: Test-time scaling improves vision-language-action (VLA) policies by draw
arXiv cs.RO 11/18/2026 Full text here
77pts
VoxelSBFL: Voxel-Level Spectrum-Based Failure Prediction for Vision-Language-Action Models
arXiv:2611.22453v1 Announce Type: cross Abstract: Vision-Language-Action (VLA) models integrate visual perception, langu
arXiv cs.SE 11/18/2026 Full text here
72pts