Hacker News Reader

"It's a ~200 LOC change that only took me a few hours and it worked the first time."

jart is on another plane of existence—and 100% earned this flex

For knowing mmap exists (i.e. soft page faults are intensely useful explicitly for this kind of thing)? Or am I reading this wrong?

Due to nuances of how data is split across the various model files, the implementation involved some non-trivial steps in order for everything to be allocated correctly. The changes are here [1].

[1] https://github.com/ggerganov/llama.cpp/commit/5b8023d9354010...

I've been producing about 300 LOC/hr with the help of GPT-4.

Generated code practically always runs correctly the first time.

I'm not a particularly proficient and before this I would probably be happy with averaging 30 LOC/hr.

It's all about quality, not quantity—call me when you can do what jart does with a LLM. Actually, don't call me, I'll be in a bunker by then

Reply on news.ycombinator.com