A selector trained once on LLaVA-665K in CLIP space selects 15% of instructions to reach 98.3% of full-data performance and generalizes to an unseen dataset and different VLMs.
Preprint
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
verdicts
UNVERDICTED 2representative citing papers
LCG applies centroid clustering and confidence-guided semi-supervised selection to curate 6K-sample subsets that yield superior MT-bench performance compared with prior filtering methods when used for instruction tuning.
citing papers explorer
-
Once-For-All: A Train-Once and Select-Anytime Framework for Multimodal Instruction Tuning
A selector trained once on LLaVA-665K in CLIP space selects 15% of instructions to reach 98.3% of full-data performance and generalizes to an unseen dataset and different VLMs.
-
Low-Confidence Gold: Refining Low-Confidence Samples for Efficient Instruction Tuning
LCG applies centroid clustering and confidence-guided semi-supervised selection to curate 6K-sample subsets that yield superior MT-bench performance compared with prior filtering methods when used for instruction tuning.