Research map / Vision-language and multimodal / Multimodal LLMs
Growing a Twig to Accelerate Large Vision-Language Models
Zhenwei Shao, Mingyang Wang, Zhou Yu, Wenwen Pan, Yan Yang, Tao Wei, Hongyuan Zhang, Ning Mao, Wei Chen, Jun Yu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| guide token pruning with last twig layer attention instead of backbone layer K (ablation) | TwigVLM · 6 VLM benchmarks | RelAcc: 82.3 -> 96.0 | Improved |
| use twig layer attention instead of backbone layer K+T at same depth (ablation) | TwigVLM · 6 VLM benchmarks | RelAcc: 86.2 -> 96.0 | Improved |
| add SSD to token pruning (TTP & SSD vs FastV) (ablation) | TwigVLM · MM-Vet | RelSpd: 104.3 -> 153.6 | Improved |
| add self-speculative decoding without token pruning (ablation) | TwigVLM · MM-Vet | RelSpd: 104.3 -> 146.7 | Improved |
| add TTP on top of SSD (ablation) | TwigVLM · MM-Vet | RelSpd: 146.7 -> 153.6 | Improved |
| initialize twig from layers K..K+T instead of random (ablation) | TwigVLM · 6 VLM benchmarks | RelAcc: 87.2 -> 96.0 | Improved |
| initialize twig from layers K..K+T instead of last T layers (ablation) | TwigVLM · 6 VLM benchmarks | RelAcc: 90.4 -> 96.0 | Improved |
| increase twig layers from 1 to 3 (ablation) | TwigVLM · 6 VLM benchmarks | RelAcc: 93.9 -> 96.0 | Improved |
| increase twig layers from 3 to 4 (ablation) | TwigVLM · MM-Vet RelSpd | RelSpd: 153.6 -> 145.4 | Hurt |
| prune at layer K=2 instead of K=0 (ablation) | TwigVLM · 6 VLM benchmarks | RelAcc: 93.6 -> 96.0 | Improved |
| prune at layer K=2 instead of K=3 (ablation) | TwigVLM · 6 VLM benchmarks | RelAcc: 85.5 -> 96.0 | Improved |
| add FinalWipe with Kf=24 (ablation) | TwigVLM · 6 VLM benchmarks | RelAcc: 93.1 -> 96.0 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VisionZip — VisionZip: Longer is Better but Not Necessary in Vision Language Models (CVPR 2025) · better on 8 of 8 shared results
- SparseVLM — SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference (ICML 2025) · better on 2 of 2 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
- A Stitch in Time Saves Nine: Small VLM is a Precise Guidance for Accelerating Large VLMs (CVPR 2025 · similarity 0.93)
- ATP-LLaVA: Adaptive Token Pruning for Large Vision Language Models (CVPR 2025 · similarity 0.91)
- TopV: Compatible Token Pruning with Inference Time Optimization for Fast and Low-Memory Multimodal Vision Language Model (CVPR 2025 · similarity 0.91)
- SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference (ICML 2025 · similarity 0.91)
- SwiftVLM: Efficient Vision-Language Model Inference via Cross-Layer Token Bypass (NeurIPS 2026 · similarity 0.91)
- Balanced Token Pruning: Accelerating Vision Language Models Beyond Local Optimization (NeurIPS 2025 · similarity 0.90)
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.
