It abstracts away llama.cpp in a way that, frankly, leaves a lot of performance and quality on the table.
OP, do you have any telemetry you can show us comparing the performance difference between what you setup on this guide and an Ollama setup? Otherwise, at face value, I’m going to assume this is another thing on the internet i have to assume is uncorroborated bullshit. Apologies for sounding rude.
I don’t like some things about the devs. I won’t rant, but I especially don’t like the hint they’re cooking up something commercial.
This concerns me. Please provide links for us to read here about this. I would like any excuse to uninstall Ollama. Thank you!
I mean, if the data was written to a HDD, then any forensic tool can read the magnetic residual patterns on the metal platters instead of looking for file headers?