Hacker News

Show HN: Gemma 3 inference in pure C++ with Metal acceleration

6 points by ybubnov ago | 2 comments

k1r111 [-]

Looks really cool, thank you. I can't find anything about performance. Is it faster? Or is it just a cool demo?

ybubnov |root |parent [-]

That’s in my short list of next things to do. In the recent releases my primary focus was on compact size of the executable and modern C++ API.