Research map / Vision-language and multimodal / Multimodal LLMs
Parallel Vision Token Scheduling for Fast and Accurate Multimodal LMMs Inference
Wengyi Zhan, Mingbao Lin, Zhihang Lin, Rongrong Ji
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 |
|---|---|---|---|
| increase migration depth n from 3 to 15 (ablation) | ParVTS · LLaVA-1.5-7B, 64 tokens | average retained (%): 92.91% -> 94.38% | Improved |
| increase migration depth n from 3 to 15 (ablation) | ParVTS · LLaVA-1.5-7B, 192 tokens | average retained (%): 97.95% -> 95.86% | Hurt |
| use parallel scheduling instead of non-subject-first sequential scheduling (ablation) | ParVTS · LLaVA-1.5-7B, 64 tokens | average retained (%): 89.80% -> 92.91% | Improved |
| use parallel scheduling instead of subject-first sequential scheduling (ablation) | ParVTS · LLaVA-1.5-7B, 64 tokens | average retained (%): 85.97% -> 92.91% | Improved |
| use parallel scheduling instead of subject-first sequential scheduling (ablation) | ParVTS · LLaVA-1.5-7B, 192 tokens | average retained (%): 77.77% -> 97.95% | Improved |
| use parallel scheduling instead of non-subject-first sequential scheduling (ablation) | ParVTS · LLaVA-1.5-7B, 192 tokens | average retained (%): 97.28% -> 97.95% | Improved |
| add non-subject path with equal fusion weights alpha=beta=0.5 instead of subject-only beta=1.0 (ablation) | ParVTS · LLaVA-1.5-7B, 64 tokens | average retained (%): 91.55% -> 92.91% | Improved |
| use equal fusion weights instead of alpha=1.0 beta=0.0 (ablation) | ParVTS · LLaVA-1.5-7B, 64 tokens | average retained (%): 92.52% -> 92.91% | Improved |
| use ParVTS instead of SAINT at 64 tokens | SAINT · LLaVA-1.5-7B, 64 tokens | average retained (%): 85.69% -> 92.45% | Improved |
| use ParVTS instead of FastV at -88.89% tokens | FastV · Qwen2.5-VL-3B | average retained (%): 67.46% -> 77.21% | Improved |
| use ParVTS instead of FastV at -88.89% tokens | FastV · InternVL2.5-2B | average retained (%): 91.43% -> 95.88% | Improved |
| use ParVTS instead of FastV at -49.80% tokens | FastV · Video-LLaVA-7B | accuracy: 58.1 -> 57.8 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most 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
- EfficientLLaVA: Generalizable Auto-Pruning for Large Vision-language Models (CVPR 2025 · similarity 0.91)
- Boosting Multimodal Large Language Models with Visual Tokens Withdrawal for Rapid Inference (AAAI 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)
- Accelerating Multimodal Large Language Models via Dynamic Visual-Token Exit and the Empirical Findings (NeurIPS 2025 · similarity 0.90)
- Dynamic-LLaVA: Efficient Multimodal Large Language Models via Dynamic Vision-language Context Sparsification (ICLR 2025 · similarity 0.90)
- VisionPulse: Dynamic Visual Sparsity for Efficient Multimodal Reasoning (ICML 2026 · 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.
