Research map / Vision-language and multimodal / Multimodal LLMs
Accelerating Multimodal Large Language Models with Prior-Corrected Token Reduction
Zengjie Chen, Yuxiang Cai, Jingcai Guo, Taotao Cai, Jianwei Yin, Zhi Chen
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 |
|---|---|---|---|
| subtract the prior from posterior (P - Q instead of P) (ablation) | token scoring · Scoring ablation, LLaVA-1.5-7B, 64 tokens | avg (%): 86.7 -> 87.6 | Improved |
| use P log(P/Q) instead of P - Q (ablation) | token scoring · Scoring ablation, LLaVA-1.5-7B, 64 tokens | avg (%): 87.6 -> 91.5 | Improved |
| use P log(P/Q) instead of posterior P (ablation) | token scoring · Scoring ablation, LLaVA-1.5-7B, 128 tokens | avg (%): 93.7 -> 95.8 | Improved |
| use P log(P/Q) instead of posterior P (ablation) | token scoring · Scoring ablation, LLaVA-1.5-7B, 32 tokens | avg (%): 74.5 -> 84.5 | Improved |
| rank by prior Q only instead of posterior P (ablation) | token scoring · Scoring ablation, LLaVA-1.5-7B, 64 tokens | avg (%): 86.7 -> 79.3 | Hurt |
| use entropy score instead of posterior (ablation) | token scoring · Scoring ablation, LLaVA-1.5-7B, 64 tokens | avg (%): 86.7 -> 72.3 | Hurt |
| use log ratio instead of P - Q (ablation) | token scoring · Scoring ablation, LLaVA-1.5-7B, 64 tokens | avg (%): 87.6 -> 81.7 | Hurt |
| add prior correction to FastV-style ranking (ablation) | FastV · LLaVA-1.5-7B, 64 tokens | normalized avg (%): 70.7 -> 94.5 | Improved |
| add prior correction to FastV-style ranking (ablation) | FastV · Qwen3-VL-8B, 11.1% tokens | normalized avg (%): 75.5 -> 87.8 | Improved |
| add prior correction to FastV pruning (ablation) | FastV · Video-LLaVA video QA (GPT avg acc) | accuracy: 26.1 -> 30.7 | Improved |
| add prior-guided token merging (ablation) | PriorTR pruning · Video-LLaVA video QA (GPT avg acc) | accuracy: 30.7 -> 38.7 | Improved |
| add prior-guided merging instead of SparseVLM merging (ablation) | SparseVLM · Video-LLaVA video QA (Gemini avg acc) | accuracy: 40.8 -> 41.6 | Improved |
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 4 of 4 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
- ST3: Accelerating Multimodal Large Language Model by Spatial-Temporal Visual Token Trimming (AAAI 2025 · similarity 0.93)
- Beyond Attention or Similarity: Maximizing Conditional Diversity for Token Pruning in MLLMs (NeurIPS 2025 · similarity 0.92)
- Fit and Prune: Fast and Training-free Visual Token Pruning for Multi-modal Large Language Models (AAAI 2025 · similarity 0.92)
- Task-Related Token Compression in Multimodal Large Language Models from an Explainability Perspective (ICLR 2026 · similarity 0.92)
- Boosting Multimodal Large Language Models with Visual Tokens Withdrawal for Rapid Inference (AAAI 2025 · similarity 0.91)
- Accelerating Multimodal Large Language Models via Dynamic Visual-Token Exit and the Empirical Findings (NeurIPS 2025 · similarity 0.91)
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.
