Metraipolis › archive › dc_9ae1007d7afb

Token-Budget Distillation: Transferring Full-Token Semantics to Compressed Video Vision-Language Models — arXiv:2608.28138

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.28138
License (as archived)
CC0-1.0
Archived
2026-10-17T11:33:10.278Z
SHA-256 of the archived text
dc895a7265ecd601ab4f12d0f11e47fba467d9c9e58befd3406304676867029d
Stable URL of this page
https://metraipolis.com/documento/dc_9ae1007d7afb

How to cite

Token-Budget Distillation: Transferring Full-Token Semantics to Compressed Video Vision-Language Models — arXiv:2608.28138 — arXiv (en), https://arxiv.org/abs/2608.28138 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_9ae1007d7afb (Metraipolis archive dc_9ae1007d7afb, sha256 dc895a7265ecd601ab4f12d0f11e47fba467d9c9e58befd3406304676867029d)
BibTeX
@misc{dc9ae1007d7afb,
  title = {Token-Budget Distillation: Transferring Full-Token Semantics to Compressed Video Vision-Language Models — arXiv:2608.28138},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.28138},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_9ae1007d7afb; sha256 dc895a7265ecd601ab4f12d0f11e47fba467d9c9e58befd3406304676867029d},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.28138. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Adapting video vision-language models (VLMs) is computationally expensive because video inputs produce a large number of visual tokens, making both fine-tuning and inference costly. Although visual token compression can reduce this overhead, direct adaptation on compressed inputs often causes semantic drift and noticeable performance degradation. We present Token-Budget Distillation (TBD), a parameter-efficient fine-tuning framework for adapting video VLMs under a fixed token budget. TBD freezes the pretrained backbone, updates only LoRA adapters, and integrates FlashVID-based visual token compression into the video pathway. To preserve full-token semantics under compression, TBD employs a dual-path teacher-student design, where a full-token teacher provides stable supervision and a compressed student is optimized with task loss, answer-region KL distillation, GT-anchored margin distillation, and reliability-aware KD control. This design enables the student to recover the semantic behavior of the full-token model while remaining efficient under aggressive token reduction. We evaluate TBD on three video VLM backbones, including LLaVA-Video, LLaVA-OneVision, and Qwen3-VL-8B-Instruct, across four video understanding benchmarks. TBD consistently outperforms compression-only baselines under both moderate and aggressive compression. On LLaVA-Video at retention ratio R = 10 percent, TBD preserves 97.0 percent of the Vanilla model's average accuracy; on LLaVA-OneVision at R = 10 percent, it achieves an average score of 58.4 and matches 100.0 percent relative accuracy.

Tema degli esploratori: statistica e scienza dei dati (retro-catalogo arXiv: annunci del 2026-08-31 in cs.CV).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV).
Pagina del lavoro: https://arxiv.org/abs/2608.28138 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-17T11:10:05.306Z dalla pagina https://arxiv.org/catchup/cs.CV/2026-08-31?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).