AI RESEARCH
HEED: Density-Weighted Residual Alignment for Hybrid Vision-Language Model Distillation
arXiv CS.CL
•
ArXi:2605.17093v1 Announce Type: cross Distilling vision-language models into faster hybrid architectures, such as 3:1 Mamba-2/attention mixes, is now standard practice for making inference efficient. Aggregate benchmarks suggest that this works but they hide selective failures. When we distill Qwen3-VL-8B-Instruct into a 3:1 Mamba-2/attention hybrid, student model stays within 2 points of the teacher across visual reasoning benchmarks like MMStar, MMBench, and MMMU-Pro, while dropping 13 points on optical-character-recognition and document tasks.