/vt/ - Virtual Youtubers

>>65657977
Which is what I am trying to fix now. >>65631014 Unfortunately the pixart-a script for Llava tagging uses the full unquantized llava version and I'm not smart enough to change the script to use 4-bit nor do i have 50gb vram.
The next best thing I can think of is using oobabooga api loaded with quantized llava and pass off the images through requests then saving the result as a caption.