LLaVA

LLaVA is a benchmark used to evaluate multimodal large language models (MLLMs) by assessing their performance on vision-language tasks while addressing the challenges of continual learning and mitigating performance degradation in natural language understanding and generation.

Papers using LLaVA (3)

Cascaded Sparse Autoencoders Learn Multi-Level Visual Concepts in Multimodal LLMs2026Online Dynamic Batching with Formal Guarantees for LLM Training2026Improving Multimodal Large Language Models Using Continual Learning2024