TY  - RPRT
TI  - Delta-LLaVA: Base-then-Specialize Alignment for Token-Efficient Vision-Language Models
AU  - Mohamad Zamini
AU  - Diksha Shukla
PY  - 2025
UR  - https://arxiv.org/abs/2512.18910
ID  - 2512.18910
ER  -