Models

← All models

LLaVA v1.5 7B

LLaVA Authors · 2023-10-05 · 7.1B parameters

LLaVA-1.5 7B is a visual instruction-tuned vision-language model developed by Haotian Liu and collaborators, released in October 2023, built on a Vicuna 7B language backbone combined with a CLIP visual encoder. It was a significant milestone among open-weight multimodal LLMs, demonstrating that carefully curated visual instruction tuning data could match larger models on standard vision-language benchmarks. The model is hosted at liuhaotian/llava-v1.5-7b on Hugging Face and supports image-grounded question answering and conversation.

Benchmark scores