This is
LLaVA-NeXT-7B-Video-FT, a video-language model fine-tuned for our ICLR 2026 paper
Map the Flow: Revealing Hidden Pathways of Information in VideoLLMs.
We fine-tuned
llava-hf/llava-v1.6-vicuna-7b-hf on the video portion of
VideoChat2-IT (our cleaned annotations:
VideoChat2-IT-clean) for 3epochs to study how video instruction tuning shapes information flow in VideoLLMs.
This model is used to analyze temporal reasoning patterns via causal intervention tools such as Attention Knockout and Logit Lens.
We identify effective information pathways in VideoLLMs and show that these sparse pathways are sufficient for solving VideoQA tasks.
With only 42% of attention edges in LLaVA-NeXT-7B-Video-FT composing these effective pathways, the model retains its VideoQA performance.
1@inproceedings{kim2026map,
2 author = {Kim, Minji and Kim, Taekyung and Han, Bohyung},
3 title = {Map the Flow: Revealing Hidden Pathways of Information in VideoLLMs},
4 booktitle = {International Conference on Learning Representations (ICLR)},
5 year = {2026},
6}
7
8@article{kim2025map,
9 author = {Kim, Minji and Kim, Taekyung and Han, Bohyung},
10 title = {Map the Flow: Revealing Hidden Pathways of Information in VideoLLMs},
11 journal = {arXiv preprint arXiv:2510.13251},
12 year = {2025},
13}
