Research map / Vision-language and multimodal / Multimodal LLMs
METEOR: Multi-Encoder Collaborative Token Pruning for Efficient Vision Language Models
Yuchen Liu, Yaoming Wang, Bowen Shi, Xiaopeng Zhang, Wenrui Dai, Chenglin Li, Hongkai Xiong, Qi Tian
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 |
|---|---|---|---|
| use post-projection instead of pre-projection fusion (ablation) | multi-encoder MLLM · Ablation (558K PT + 1M SFT, EAGLE encoders) | average accuracy: 67.5 -> 68.3 | Improved |
| use collaborative pruning instead of separate per-encoder pruning (ablation) | METEOR · Ablation (558K PT + 1M SFT, EAGLE encoders) | average accuracy: 66.9 -> 68.3 | Improved |
| use collaborative pruning instead of random pruning (ablation) | METEOR · Ablation (558K PT + 1M SFT, EAGLE encoders) | average accuracy: 61.4 -> 68.3 | Improved |
| use collaborative pruning instead of MLP compression (ablation) | METEOR · Ablation (558K PT + 1M SFT, EAGLE encoders) | average accuracy: 64.1 -> 68.3 | Improved |
| use collaborative pruning instead of Resampler compression (ablation) | METEOR · Ablation (558K PT + 1M SFT, EAGLE encoders) | average accuracy: 60.3 -> 68.3 | Improved |
| use collaborative pruning instead of PixelShuffle (ablation) | METEOR · Ablation (558K PT + 1M SFT, EAGLE encoders) | average accuracy: 63.9 -> 68.3 | Improved |
| add top-k attention head filtering (ablation) | text-guided pruning, 297 tokens · Ablation (558K PT + 1M SFT, EAGLE encoders) training-free | OCR average: 59.1 -> 59.9 | Improved |
| use instance-adaptive ratio instead of fixed ratio at 242 tokens (ablation) | text-guided pruning · Ablation (558K PT + 1M SFT, EAGLE encoders) training-free | average accuracy: 66.9 -> 67.9 | Improved |
| use instance-adaptive ratio instead of fixed ratio at 242 tokens (ablation) | text-guided pruning · Ablation (558K PT + 1M SFT, EAGLE encoders) training-free | OCR average: 62.4 -> 65.0 | Improved |
| prune to 126 tokens instead of 242 tokens (ablation) | METEOR with adaptive ratio · Ablation (558K PT + 1M SFT, EAGLE encoders) training-free | average accuracy: 67.9 -> 67.2 | Hurt |
| replace EAGLE with METEOR pruning | EAGLE (Vicuna-7B) · Efficient MLLM comparison (EAGLE-based, Vicuna-7B) | TFLOPS: 26.21 -> 13.42 | Improved |
| replace EAGLE with METEOR pruning | EAGLE (Vicuna-7B) · Efficient MLLM comparison (EAGLE-based, Vicuna-7B) | average accuracy: 69.3 -> 69.0 | Hurt |
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 1 of 1 shared result
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
- ViTCoP: Accelerating Large Vision-Language Models via Visual and Textual Semantic Collaborative Pruning (AAAI 2026 · similarity 0.90)
- Balanced Token Pruning: Accelerating Vision Language Models Beyond Local Optimization (NeurIPS 2025 · similarity 0.90)
- ATP-LLaVA: Adaptive Token Pruning for Large Vision Language Models (CVPR 2025 · similarity 0.89)
- LearnPruner: Rethinking Attention-based Token Pruning in Vision Language Models (ICLR 2026 · similarity 0.89)
- TopV: Compatible Token Pruning with Inference Time Optimization for Fast and Low-Memory Multimodal Vision Language Model (CVPR 2025 · similarity 0.89)
- iLLaVA: An Image is Worth Fewer Than 1/3 Input Tokens in Large Multimodal Models (ICLR 2026 · similarity 0.88)
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.
