[einsteinx2]

You can also change it manually on macOS [0], though yes by default it’s automatic but has a much lower limit (I think 2/3 is total RAM or something).

[0]: https://github.com/ggml-org/llama.cpp/discussions/2182#discu...