Redownloading All The Old Models

So the new gguf format looks cool in that it supposedly automatically calculates the rope-freq for you. However, I do actually have data caps on my internet so I really want to try my hand at converting my existing ggmls.

It looks like you can do it a llama.cpp script:

python convert-llama-ggmlv3-to-gguf.py -i ~/models/llama-2-7b.ggmlv3.q5_1.bin -o ~/models/llama-2-7b.gguf.q5_1.bin

but it only works with ggmlv3 models; I think a lot of my older ones are not v3 so I may not get away with just converting them. I'll probably hang on to an old copy of gpt4all and oobabooga just so that I can keep running them a bit longer.

Anyhow- if you do find yourself trying to convert your ggmls, that script seems to be the magic tool to do it.