Research map / Vision-language and multimodal / Multimodal LLMs
D-Attn: Decomposed Attention for Large Vision-and-Language Model
Chia-Wen Kuo, Sijie Zhu, Fan Chen, Xiaohui Shen, Longyin Wen
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 |
|---|---|---|---|
| add V2V diagonalization (ablation) | D-Attn (Mistral 0.3 7B) · visual tokens processed in training | max |V|: 9k -> 74k | Improved |
| add V2V diagonalization (ablation) | D-Attn (Mistral 0.3 7B) · training speed | sec/it: 11.25 -> 2.24 | Improved |
| add V2V diagonalization alone (ablation) | D-Attn (Mistral 0.3 7B) · GQA | accuracy: 61.8 -> 63.4 | Improved |
| add V2V diagonalization alone (ablation) | D-Attn (Mistral 0.3 7B) · MMStar | accuracy: 36.8 -> 32.6 | Hurt |
| add V2V diagonalization alone (ablation) | D-Attn (Mistral 0.3 7B) · MME | perception score: 1533.1 -> 1507.6 | Mixed |
| add debiased positional encodings on top of diagonalization (ablation) | D-Attn (Mistral 0.3 7B) · MME | perception score: 1507.6 -> 1598.6 | Improved |
| add debiased positional encodings on top of diagonalization (ablation) | D-Attn (Mistral 0.3 7B) · MMStar | accuracy: 32.6 -> 38.3 | Improved |
| add debiased positional encodings on top of diagonalization (ablation) | D-Attn (Mistral 0.3 7B) · VQA-T | accuracy: 63.4 -> 68.3 | Improved |
| use alpha-weighting instead of Tanh gate merging (ablation) | D-Attn (Mistral 0.3 7B) · MME | perception score: 1337.6 -> 1598.6 | Improved |
| use alpha-weighting instead of Sigmoid gate merging (ablation) | D-Attn (Mistral 0.3 7B) · MMStar | accuracy: 35.8 -> 38.3 | Improved |
| use alpha-weighting instead of Cascade merging (ablation) | D-Attn (Mistral 0.3 7B) · MME | perception score: 1586.1 -> 1598.6 | Improved |
| replace self-attention with D-Attn | Gemma 2 9B LVLM · MMVP | accuracy: 29.3 -> 45.3 | Improved |
Closest papers by meaning
- To Sink or Not to Sink: Visual Information Pathways in Large Vision-Language Models (ICLR 2026 · similarity 0.91)
- Rethinking Visual Token Reduction in LVLMs Under Cross-Modal Misalignment (AAAI 2026 · similarity 0.91)
- See What You Are Told: Visual Attention Sink in Large Multimodal Models (ICLR 2025 · similarity 0.91)
- Boosting Multimodal Large Language Models with Visual Tokens Withdrawal for Rapid Inference (AAAI 2025 · similarity 0.90)
- An Empirical Study of Attention and Diversity for Adaptive Visual Token Pruning in Large Vision-Language Models (ICLR 2026 · similarity 0.90)
- Accelerating Multimodal Large Language Models via Dynamic Visual-Token Exit and the Empirical Findings (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.
