Are LLaVA variants better than original?

LLaVA is an open-source large multi-modal that uses a combination of the Vicuna LLM and CLIP vision encoder. In this video, we're going to compare the initial LLaVA model with more recently trained LLaVA models based on Meta's llama3 and Microsoft's phi3.
We'll see if they can extract code from a SQL query, tell us who Cristiano Ronaldo is, understand a graph/network diagram, and more!
#lmmssong #llms #llava #llama3 #phi3 #ollama
LLaVA - github.com/LLa...
LLaVA models on Ollama - ollama.com/sea...
Code - github.com/mne...
• LLaVA 1.6 is here...bu...
• LLaVA: A large multi-m...