For video vision-language models under a fixed inference compute budget, the optimal setup scales language model size, frame count, and tokens per frame jointly, and shifts toward smaller LMs with more frames and tokens as finetuning data grows.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Inference Compute-Optimal Video Vision Language Models
For video vision-language models under a fixed inference compute budget, the optimal setup scales language model size, frame count, and tokens per frame jointly, and shifts toward smaller LMs with more frames and tokens as finetuning data grows.