Fun update to this: Daniel Lemire added another optimization to make this even faster. https://github.com/jadidbourbaki/llama.cpp/pull/12

I’ll benchmark his change and add it to the article, crediting him for this improvement.

Btw, if anyone has experience with the open source community in general and llama.cpp in specific, I would greatly appreciate some advice. I’m facing a bit of an interpersonal issue that I really hope is resolved without any ill will. Here is the context:

https://www.reddit.com/r/LocalLLaMA/comments/1wr5ylm/comment...

Any advice for what I can do? Due to this, I cannot create a PR or issue in the llama.cpp repository. However, I am worried about bothering the maintainers on other channels in case it aggravates them further. Thank you for your help!