Research map / Vision-language and multimodal / Multimodal LLMs
Turbo: Informativity-Driven Acceleration Plug-In for Vision-Language Large Models
Chen Ju, Haicheng Wang, Haozhe Cheng, Xu Chen, Zhonghua Zhai, Weilin Huang, Jinsong Lan, Shuai Xiao, Bo Zheng
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add mutual redundancy alone to token merging (ablation) | Turbo (BLIP captioning) · COCO captioning | CIDEr: 112.8 -> 120.9 | Improved |
| add semantic value alone to token merging (ablation) | Turbo (BLIP captioning) · COCO captioning | CIDEr: 112.8 -> 123.8 | Improved |
| combine mutual redundancy and semantic value instead of either alone (ablation) | Turbo (BLIP captioning) · COCO captioning | CIDEr: 123.8 -> 130.0 | Improved |
| use multiplicative instead of additive fusion of the two components (ablation) | Turbo · COCO captioning | CIDEr: 130.0 -> 129.9 | No clear effect |
| use dynamic per-layer fusion (S4) instead of simple weighted difference (ablation) | Turbo · COCO captioning | CIDEr: +0.3 CIDEr | No clear effect |
| add minimum token-length threshold at large drop ratio (ablation) | Turbo · COCO captioning, ratio 65 | CIDEr: 95.5 -> 115.3 | Improved |
| vary balancing coefficient alpha from 1 to 20 | Turbo · BLIP captioning | performance | No clear effect |
| use Turbo instead of UPop pruning for retrieval acceleration | BLIP · Flickr30k retrieval | image-to-text R@1: 92.5 -> 96.8 | Improved |
| use Turbo instead of ToMe for token merging | ViT-B/16 AugReg · ImageNet-1k | top-1 accuracy: 82.6 -> 83.1 | Improved |
| apply GPTQ 4-bit quantization to BLIP2 | BLIP2 · COCO captioning | throughput: 29.5 -> 30.1 (1.02x) | No clear effect |
| stack Turbo on top of UPop pruning | BLIP · COCO captioning | throughput speedup: 1.98x -> 2.19x | Improved |
Later papers that compared with this method
4 later accepted papers report Turbo-8B as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- TableMix: Enhancing Multimodal Table Reasoning in MLLMs from a Data-Centric Perspective (CVPR 2026 · better on 6 of 6 shared results)
- S³-MSD: Large Vision-Language Model for Explainable and Generalizable Multi-modal Sarcasm Detection (AAAI 2026 · better on 1 of 1 shared result)
- FOLDER: Accelerating Multi-Modal Large Language Models with Enhanced Performance (ICCV 2025 · better on 6 of 6 shared results)
- Libra-Merging: Importance-redundancy and Pruning-merging Trade-off for Acceleration Plug-in in Large Vision-Language Model (CVPR 2025 · better on 5 of 5 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Gated Relational Alignment via Confidence-based Distillation for Efficient VLMs (ICML 2026 · similarity 0.87)
- VLsI: Verbalized Layers-to-Interactions from Large to Small Vision Language Models (CVPR 2025 · similarity 0.87)
- A Stitch in Time Saves Nine: Small VLM is a Precise Guidance for Accelerating Large VLMs (CVPR 2025 · similarity 0.87)
- Lifting the Veil on Visual Information Flow in MLLMs: Unlocking Pathways to Faster Inference (CVPR 2025 · similarity 0.87)
- What matters when building vision-language models? (NeurIPS 2024 · similarity 0.87)
- Collaborative Multi-Mode Pruning for Vision-Language Models (CVPR 2026 · similarity 0.87)
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.
