# Towards Long-window Anchoring in Vision-Language Model Distillation

arXiv: https://arxiv.org/abs/2512.21576 · HTML: https://mlsanity.com/paper/2512.21576/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in MLLM training and visual instruction tuning, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 23 of its 28 references that are in the map.
- First posted on arXiv 25 Dec 2025.

## Placed in

- Multimodal LLMs / vision-language models / [MLLM training and visual instruction tuning](https://mlsanity.com/mllm/mllm-training-and-visual-instruction-tuning/)

## Builds on (most cited first, 23 of 23)

- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://mlsanity.com/paper/2104.09864/) (2021)
- [Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution](https://mlsanity.com/paper/2409.12191/) (2024)
- [Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolation](https://mlsanity.com/paper/2108.12409/) (2021)
- [Extending Context Window of Large Language Models via Positional Interpolation](https://mlsanity.com/paper/2306.15595/) (2023)
- [Yi: Open Foundation Models by 01.AI](https://mlsanity.com/paper/2403.04652/) (2024)
- [Model Tells You What to Discard: Adaptive KV Cache Compression for LLMs](https://mlsanity.com/paper/2310.01801/) (2023)
- [LongRoPE: Extending LLM Context Window Beyond 2 Million Tokens](https://mlsanity.com/paper/2402.13753/) (2024)
- [LongVILA: Scaling Long-Context Visual Language Models for Long Videos](https://mlsanity.com/paper/2408.10188/) (2024)
- [Effective Long-Context Scaling of Foundation Models](https://mlsanity.com/paper/2309.16039/) (2023)
- [LLM Maybe LongLM: Self-Extend LLM Context Window Without Tuning](https://mlsanity.com/paper/2401.01325/) (2024)
- [Focused Transformer: Contrastive Training for Context Scaling](https://mlsanity.com/paper/2307.03170/) (2023)
- [Randomized Positional Encodings Boost Length Generalization of Transformers](https://mlsanity.com/paper/2305.16843/) (2023)
- [PoSE: Efficient Context Window Extension of LLMs via Positional Skip-wise Training](https://mlsanity.com/paper/2309.10400/) (2023)
- [Compressing Visual-linguistic Model via Knowledge Distillation](https://mlsanity.com/paper/2104.02096/) (2021)
- [Found in the Middle: How Language Models Use Long Contexts Better via Plug-and-Play Positional Encoding](https://mlsanity.com/paper/2403.04797/) (2024)
- [Visual Haystacks: A Vision-Centric Needle-In-A-Haystack Benchmark](https://mlsanity.com/paper/2407.13766/) (2024)
- [Fourier Position Embedding: Enhancing Attention's Periodic Extension for Length Generalization](https://mlsanity.com/paper/2412.17739/) (2024)
- [LLAVADI: What Matters For Multimodal Large Language Models Distillation](https://mlsanity.com/paper/2407.19409/) (2024)
- [Robustifying State-space Models for Long Sequences via Approximate Diagonalization](https://mlsanity.com/paper/2310.01698/) (2023)
- [LongReD: Mitigating Short-Text Degradation of Long-Context Large Language Models via Restoration Distillation](https://mlsanity.com/paper/2502.07365/) (2025)
- [On the token distance modeling ability of higher RoPE attention dimension](https://mlsanity.com/paper/2410.08703/) (2024)
- [VLM-KD: Knowledge Distillation from VLM for Long-Tail Visual Recognition](https://mlsanity.com/paper/2408.16930/) (2024)

## Built on by (most cited first, 2 of 2)

- [Can Vision Replace Text in Working Memory? Evidence from Spatial n-Back in Vision-Language Models](https://mlsanity.com/paper/2602.04355/) (2026)
- [MCD: Causal Distillation of Multimodal In-Context Learning in Large Vision-Language Models](https://mlsanity.com/paper/2609.39920/) (2026)

## Abstract (arXiv)

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.