SeGO: Sensitivity-Aware Golden Optimization for Large-Scale VLM Quantization
A cross-modal structural sensitivity asymmetry in VLMs is revealed and SeGO is proposed, a unified structural sensitivity-aware sparse optimization framework that achieves the balance among model parameter amount, quantization accuracy and scaling factors’ search efficiency on InternVL2 and LLaVA series.
Tianqi Zhao, Xinrui Cheng, Yang Su et al.
· 0 citations