Programmatically working with a LLM is like sending off a request to an Indian computer center and then spending all the development time you would have done to do it yourself to try to validate and babysit it.
It's my first time doing it this way, as in running a llm server and then querying it.
It's insanity though. It just makes shit up sometimes.
I mean, it *mostly* works, but it feels way more hammered dogshit than it needs to be fr.
I don't understand why every company wants to integrate this shit. Going through all these words in a manner like this runs my GPU at 200W - and it looks like it's going to take a good 30-40 minutes to go through 11,000 words that I need to find equivalents to.
Did you know there's like 130 THOUSAND or so language models on HuggingFace??