LLaVA-665K

The 'LLaVA-665K' dataset is a multimodal instruction dataset used to evaluate the effectiveness of visual reasoning in training models by identifying and selecting samples that genuinely require visual input.

Papers using LLaVA-665K (3)

MAGIC: Multimodal Alignment & Grounding-aware Instruction Coreset for Vision-Language Models2026VisNec: Measuring and Leveraging Visual Necessity for Multimodal Instruction Tuning2026Visual Compositional Tuning2025