LLaVA, or Large Language and Vision Assistant, is an open-source research project for multimodal models that combine visual understanding with language instruction following. It connects a vision encoder with a large language model and uses visual instruction tuning so the resulting assistant can reason and converse about images.

The repository provides model releases, training and evaluation code, datasets or data preparation guidance, and examples for running LLaVA variants. The project has been influential in demonstrating how relatively simple multimodal alignment and instruction-tuning recipes can produce strong general-purpose visual assistants.

LLaVA is primarily useful for researchers and developers working on vision-language models, multimodal evaluation, and image-grounded assistants. It can be used as a research baseline, a source of pretrained checkpoints, or a foundation for experimenting with new visual instruction datasets and architectures.