Record 056 · Vision · 2023
50%23.56.I
LLaVA
UW / Wisconsin + collaborators
“The open vision-language model that made 'look at this screenshot' a local call.”
LLaVA connected a vision encoder to a Llama-class LLM. Moondream, Llama 3.2 Vision, and Qwen-VL sit on this idea.
Opened local multimodal chat.
- Open VLM architecture
- Visual instruction tuning
- Screenshot/document VQA
A pair of eyes on the local model.
