# An Empirical Study on How Video-LLMs Answer Video Questions

arXiv: https://arxiv.org/abs/2508.15360 · HTML: https://mlsanity.com/paper/2508.15360/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Builds on 32 of its 43 references that are in the map.
- First posted on arXiv 21 Aug 2025.

## Builds on (most cited first, 30 of 32)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [LLaVA-OneVision: Easy Visual Task Transfer](https://mlsanity.com/paper/2408.03326/) (2024)
- [Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling](https://mlsanity.com/paper/2412.05271/) (2024)
- [Video-LLaMA: An Instruction-tuned Audio-Visual Language Model for Video Understanding](https://mlsanity.com/paper/2306.02858/) (2023)
- [Video-MME: The First-Ever Comprehensive Evaluation Benchmark of Multi-modal LLMs in Video Analysis](https://mlsanity.com/paper/2405.21075/) (2024)
- [MVBench: A Comprehensive Multi-modal Video Understanding Benchmark](https://mlsanity.com/paper/2311.17005/) (2023)
- [MiniCPM-V: A GPT-4V Level MLLM on Your Phone](https://mlsanity.com/paper/2408.01800/) (2024)
- [EVA-CLIP: Improved Training Techniques for CLIP at Scale](https://mlsanity.com/paper/2303.15389/) (2023)
- [LLaMA-Adapter: Efficient Fine-tuning of Language Models with Zero-init Attention](https://mlsanity.com/paper/2303.16199/) (2023)
- [VideoLLaMA 2: Advancing Spatial-Temporal Modeling and Audio Understanding in Video-LLMs](https://mlsanity.com/paper/2406.07476/) (2024)
- [EgoSchema: A Diagnostic Benchmark for Very Long-form Video Language Understanding](https://mlsanity.com/paper/2308.09126/) (2023)
- [LLaVA-Video: Video Instruction Tuning With Synthetic Data](https://mlsanity.com/paper/2410.02713/) (2024)
- [Dissecting Recall of Factual Associations in Auto-Regressive Language Models](https://mlsanity.com/paper/2304.14767/) (2023)
- [Long Context Transfer from Language to Vision](https://mlsanity.com/paper/2406.16852/) (2024)
- [Generic Attention-model Explainability for Interpreting Bi-Modal and Encoder-Decoder Transformers](https://mlsanity.com/paper/2103.15679/) (2021)
- [LongVILA: Scaling Long-Context Visual Language Models for Long Videos](https://mlsanity.com/paper/2408.10188/) (2024)
- [Zero-Shot Video Question Answering via Frozen Bidirectional Language Models](https://mlsanity.com/paper/2206.08155/) (2022)
- [InternVideo2.5: Empowering Video MLLMs with Long and Rich Context Modeling](https://mlsanity.com/paper/2501.12386/) (2025)
- [Boosting Multimodal Large Language Models with Visual Tokens Withdrawal for Rapid Inference](https://mlsanity.com/paper/2405.05803/) (2024)
- [Kangaroo: A Powerful Video-Language Model Supporting Long-context Video Input](https://mlsanity.com/paper/2408.15542/) (2024)
- [Causal Attention for Vision-Language Tasks](https://mlsanity.com/paper/2103.03493/) (2021)
- [Probing Image-Language Transformers for Verb Understanding](https://mlsanity.com/paper/2106.09141/) (2021)
- [Behind the Scene: Revealing the Secrets of Pre-trained Vision-and-Language Models](https://mlsanity.com/paper/2005.07310/) (2020)
- [Vision-and-Language or Vision-for-Language? On Cross-Modal Influence in Multimodal Transformers](https://mlsanity.com/paper/2109.04448/) (2021)
- [DrVideo: Document Retrieval Based Long Video Understanding](https://mlsanity.com/paper/2406.12846/) (2024)
- [VideoRoPE: What Makes for Good Video Rotary Position Embedding?](https://mlsanity.com/paper/2502.05173/) (2025)
- [Multimodal Neurons in Pretrained Text-Only Transformers](https://mlsanity.com/paper/2308.01544/) (2023)
- [DIME: Fine-grained Interpretations of Multimodal Models via Disentangled Local Explanations](https://mlsanity.com/paper/2203.02013/) (2022)
- [VideoQA in the Era of LLMs: An Empirical Study](https://mlsanity.com/paper/2408.04223/) (2024)

## Built on by (most cited first, 4 of 4)

- [CoEvolve: Training LLM Agents via Agent-Data Mutual Evolution](https://mlsanity.com/paper/2604.15840/) (2026)
- [WAVE: Learning Unified & Versatile Audio-Visual Embeddings with Multimodal LLM](https://mlsanity.com/paper/2509.21990/) (2025)
- [MLLMs Know When Before Speaking: Revealing and Recovering Temporal Grounding via Attention Cues](https://mlsanity.com/paper/2605.21954/) (2026)
- [JRDB-AVR: An Active Visual Reasoning Benchmark for Embodied Agents in Real-World Environments](https://mlsanity.com/paper/2609.35032/) (2026)

## Abstract (arXiv)

Taking advantage of large-scale data and pretrained language models, Video Large Language Models (Video-LLMs) have shown strong capabilities in answering video questions. However, most existing efforts focus on improving performance, with limited attention to understanding their internal mechanisms. This paper aims to bridge this gap through a systematic empirical study. To interpret existing VideoLLMs, we adopt attention knockouts as our primary analytical tool and design three variants: Video Temporal Knockout, Video Spatial Knockout, and Language-to-Video Knockout. Then, we apply these three knockouts on different numbers of layers (window of layers). By carefully controlling the window of layers and types of knockouts, we provide two settings: a global setting and a fine-grained setting. Our study reveals three key findings: (1) Global setting indicates Video information extraction primarily occurs in early layers, forming a clear two-stage process -- lower layers focus on perceptual encoding, while higher layers handle abstract reasoning; (2) In the fine-grained setting, certain intermediate layers exert an outsized impact on video question answering, acting as critical outliers, whereas most other layers contribute minimally; (3) In both settings, we observe that spatial-temporal modeling relies more on language-guided retrieval than on intra- and inter-frame self-attention among video tokens, despite the latter's high computational cost. Finally, we demonstrate that these insights can be leveraged to reduce attention computation in Video-LLMs. To our knowledge, this is the first work to systematically uncover how Video-LLMs internally process and understand video content, offering interpretability and efficiency perspectives for future research.