AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Look Less, Think Faster: Joint Token-Compute Adaptation for Multimodal LLMs

Pengcheng Wang, Zhiquan Wang, Jayoung Lee, Zhuoyan Xu, Ran Xu, Saurabh Bagchi, Yin Li, Somali Chaterji

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Training-free token reduction for MLLMs · 1 citations (OpenAlex)

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 changedWhereMeasuredEffect
add vision-side token controller jointly with LLM compute controllerAdaLLaVA (compute-only) · VQAv2 at 50% FLOPsaccuracy (%): 67.9 -> 74.4Improved
add token controller to compute-only adaptationAdaLLaVA · TextVQA at 50% FLOPsaccuracy (%): 29.8 -> 54.4Improved
joint end-to-end training vs sequential AdaLLaVA-PruMergeSmartVL · VQAv2 at 50% FLOPsaccuracy (%): 74.5 (sequential) vs 74.4No clear effect
joint control vs sequential composition on GQASmartVL · GQA at 50% FLOPsaccuracy (%): 60.1 (AdaLLaVA-PruMerge) vs 59.8Hurt
SmartVL vs AdaLLaVA on MMBench at 50% FLOPsSmartVL · MMBenchaccuracy (%): 63.3 (AdaLLaVA) vs 62.0Hurt
add head-group (width) gating on top of layer gating (ablation)SmartVL-L · TextVQA full budgetaccuracy (%): 58.1 -> 57.5Hurt
jointly train token and compute controllers rather than separately (ablation)SmartVL · POPE/VQAv2/TextVQAoptimal allocationImproved
scale joint adaptation to 13B backboneSmartVL · VQAv2 val 13Baccuracy (%): 72.27 (8.27T) -> 75.45 (7.25T)Improved
learned budget-conditioned token controller vs FastV / PruMerge+SmartVL token controller · VQAv2accuracy vs FLOPsImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy