coremlcompiler

Overview ¶

Package coremlcompiler compiles CoreML model packages (.mlpackage) into compiled model bundles (.mlmodelc) without requiring xcrun or Apple's proprietary coremlcompiler binary.

The primary path converts mlprogram models: the protobuf-encoded MIL program is serialized to MIL text, weights are copied byte-for-byte, and a coremldata.bin header is generated. The result can be loaded directly by CoreML, the ANE runtime, or x/ane.

Three entry points cover different use cases:

Compile takes a .mlpackage directory or .mlmodel file on disk and produces a compiled .mlmodelc bundle. This is the simplest path and handles reading the protobuf, locating weights, and writing all output files:

err := coremlcompiler.Compile("model.mlpackage", "model.mlmodelc")

CompileMLProgram accepts raw model protobuf bytes and a weight directory. Use this when you already have the serialized model in memory or need to supply the weight directory separately:

err := coremlcompiler.CompileMLProgram(protoBytes, weightDir, "model.mlmodelc")

CompileMILText takes pre-built MIL text, a ModelDescription, and an optional weight root directory. Use this when you generate MIL text programmatically rather than decoding it from a protobuf:

err := coremlcompiler.CompileMILText(milText, 8, desc, weightRoot, "model.mlmodelc")

package main

import (
	"fmt"

	"github.com/tmc/apple/x/coremlcompiler"
)

func main() {
	// Compile a CoreML model package to a compiled bundle.
	err := coremlcompiler.Compile("model.mlpackage", "model.mlmodelc")
	if err != nil {
		fmt.Println(err)
	}

	// The compiled bundle can be loaded by CoreML or x/ane:
	//
	//   rt, _ := ane.Open()
	//   k, _ := rt.Compile(ane.CompileOptions{
	//       ModelType:   ane.ModelTypePackage,
	//       PackagePath: "model.mlmodelc",
	//   })
}

Output: