Cross compiling

yzma uses no CGo, so cross compilation is the normal Go workflow. Set GOOS and GOARCH, and build.

GOOS=linux GOARCH=arm64 go build -o myapp-arm64 .

You need no C compiler and no cross toolchain.

Why this works

A CGo program must compile C code for the target, so it needs a cross toolchain for each target. yzma calls llama.cpp through purego and libffi. Nothing about llama.cpp is part of your binary, so the Go compiler alone is sufficient.

See Architecture.

The libraries are still per platform

Your Go binary is not enough. The machine that runs it also needs the llama.cpp shared libraries for its own platform.

You have two choices.

Let the application install them. The program calls download.Install the first time it runs. See Install from your program.

Ship them with the application. Put the library files beside the binary and give the path to llama.Load.

exe, err := os.Executable()
if err != nil {
	return err
}

llama.Load(filepath.Join(filepath.Dir(exe), "lib"))

Download the libraries for each target at build time:

yzma install --lib ./dist/linux-arm64/lib --os linux
yzma install --lib ./dist/darwin-arm64/lib --os darwin --processor metal
yzma install --lib ./dist/windows-amd64/lib --os windows

The targets

Raspberry Pi and Arduino UNO Q

These boards run a Debian based Linux on arm64. The Go build is the normal arm64 build. The libraries need the --os flag, because these systems use their own library versions.

GOOS=linux GOARCH=arm64 go build -o myapp .
yzma install --lib ./lib --processor cpu --os trixie

Use --os bookworm for an older Raspberry Pi OS.

A browser

A browser build uses TinyGo:

tinygo build -target wasm -o app.wasm ./examples/wasm/chat

The standard Go toolchain also works. The binary is larger.

GOOS=js GOARCH=wasm go build -o app.wasm ./examples/wasm/chat

See Build for a browser.

Test on the target

A cross compiled binary builds, but that does not prove that it runs. The llama.cpp libraries and the drivers on the target machine decide that. Run yzma system on the target to see the devices that llama.cpp found.