Towards Long-window Anchoring in Vision-Language Model Distillation

While large vision-language models (VLMs) demonstrate strong long-context understanding, their prevalent small branches fail on linguistics-photography alignment for a limited window size. We discover that knowledge distillation improves students' capability as a complement to Rotary Position Embeddings (RoPE) on window sizes (anchored from large models). Building on this insight, we propose LAid, which directly aims at the transfer of long-range attention mechanisms through two complementary components: (1) a progressive distance-weighted attention matching that dynamically emphasizes longer position differences during training, and (2) a learnable RoPE response gain modulation that selectively amplifies position sensitivity where needed. Extensive experiments across multiple model families demonstrate that LAid-distilled models achieve up to 3.2 times longer effective context windows compared to baseline small models, while maintaining or improving performance on standard VL benchmarks. Spectral analysis also suggests that LAid successfully preserves crucial low-frequency attention components that conventional methods fail to transfer. Our work not only provides practical techniques for building more efficient long-context VLMs but also offers theoretical insights into how positional understanding emerges and transfers during distillation.

Building Flexible Machine Learning Models for Scientific Computing at Scale

Foundation models have revolutionized knowledge acquisition across domains, and our study introduces OmniArch, a paradigm-shifting approach designed for building foundation models in multi-physics scientific computing. OmniArch's pre-training involves a versatile pipeline that processes multi-physics spatio-temporal data, casting forward problem learning into scalable auto-regressive tasks, while our novel Physics-Informed Reinforcement Learning (PIRL) technique during fine-tuning ensures alignment with physical laws. Pre-trained on the comprehensive PDEBench dataset, OmniArch not only sets new performance benchmarks for 1D, 2D and 3D PDEs but also demonstrates exceptional adaptability to new physics via few-shot and zero-shot learning approaches. The model's representations further extend to inverse problem-solving, highlighting the transformative potential of AI-enabled Scientific Computing(AI4SC) foundation models for engineering applications and physics discovery.