Research map / Vision-language and multimodal / Cross-modal learning
VEQ: Modality-Adaptive Quantization for MoE Vision-Language Models
Guangshuo Qin, Zhiteng Li, Zheng Chen, Weihang Zhang, Linghe Kong, Yulun Zhang
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 gradient scaling factor gamma to expert importance (ablation) | VEQ-ME · MMMU/InfoVQA/ScienceQA (Kimi-VL W3) | average accuracy (%): 62.02 -> 63.27 | Improved |
| add quantity normalization factor beta to expert importance (ablation) | VEQ-ME · MMMU/InfoVQA/ScienceQA (Kimi-VL W3) | average accuracy (%): 62.30 -> 63.27 | Improved |
| add router affinity p to Hessian weighting (ablation) | VEQ-MA · MMMU/InfoVQA/ScienceQA (Kimi-VL W3) | average accuracy (%): 63.63 -> 63.96 | Improved |
| add modality indicator alpha to Hessian weighting (ablation) | VEQ-MA · MMMU/InfoVQA/ScienceQA (Kimi-VL W3) | average accuracy (%): 62.46 -> 63.96 | Improved |
| replace GPTQ with VEQ-MA at W3 | GPTQ · Kimi-VL average of 9 benchmarks (W3) | average accuracy (%): 62.33 -> 65.41 | Improved |
| replace AWQ with VEQ-ME at W3 | AWQ · Kimi-VL average of 9 benchmarks (W3) | average accuracy (%): 63.37 -> 64.00 | Improved |
| replace GPTQ with VEQ-MA at W3 | GPTQ · Qwen3-VL average of 9 benchmarks (W3) | average accuracy (%): 64.16 -> 67.14 | Improved |
| replace GPTQ with VEQ-MA at W4 | GPTQ · Qwen3-VL average of 9 benchmarks (W4) | average accuracy (%): 73.20 -> 74.36 | Improved |
| replace GPTQ with VEQ-MA at W4 | GPTQ · Kimi-VL average of 9 benchmarks (W4) | average accuracy (%): 72.96 -> 73.57 | Improved |
| replace AWQ with VEQ-ME at W4 | AWQ · Kimi-VL average of 9 benchmarks (W4) | average accuracy (%): 72.63 -> 72.86 | Improved |
| replace AWQ with VEQ-ME at W3 | AWQ · Qwen3-VL average of 9 benchmarks (W3) | average accuracy (%): 62.22 -> 63.93 | Improved |
| proportionally scale gamma and beta together (ablation) | VEQ-ME · validation PPL (64 MMMU samples) | PPL | No clear effect |
Closest papers by meaning
- Fine-Grained Post-Training Quantization for Large Vision Language Models with Quantization-Aware Integrated Gradients (CVPR 2026 · similarity 0.93)
- MBQ: Modality-Balanced Quantization for Large Vision-Language Models (CVPR 2025 · similarity 0.92)
- Breaking Modality Heterogeneity in Low-Bit Quantization for Large Vision-Language Models (NeurIPS 2026 · similarity 0.92)
- Quant Experts: Token-aware Adaptive Error Reconstruction with Mixture of Experts for Large Vision-Language Models Quantization (CVPR 2026 · similarity 0.92)
- Q-VLM: Post-training Quantization for Large Vision-Language Models (NeurIPS 2024 · similarity 0.92)
- VLM-PTQ: Efficient Post-Training Quantization for Large Vision-Language Models (CVPR 2026 · similarity 0.92)
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.
