Using LLAVA to categorize, label, and find photos, and memes.

[themanyone]

The complete writeup is here https://github.com/themanyone/FindAImage

Using batch mode of llama-lava-cli is essential because it could potentially save a lot of time leaving part of the model in memory between describing images. But batch mode is missing a couple features. One, it does not print the image name with the description. And two, there is no template system that works for it. There is a template system for other models, but even that is not aware of file names. And thus it has no way of inserting them.

It was necessary to create a pull request to add the necessary functionality.

But it works!

We were able to generate a web page with llava.