Implementation:Mlc ai Mlc llm Prepare libs

Knowledge Sources	MLC-LLM
Domains	Deep_Learning, Mobile_Deployment
Last Updated	2026-02-09 00:00 GMT

Overview

Concrete tools for setting up cross-compilation toolchains and native build environments for mobile platforms provided by MLC-LLM.

Description

MLC-LLM provides two platform-specific build scripts that automate the cross-compilation of the inference engine into native static libraries:

ios/prepare_libs.sh -- A Bash script that configures CMake for iOS cross-compilation, registers the appropriate Rust target, and builds the mlc_llm_static library with Metal GPU support. It supports building for both physical devices (arm64 with iphoneos sysroot) and simulators (arm64 or x86_64 with iphonesimulator sysroot). The script also creates a symbolic link from ios/MLCSwift/tvm_home to the TVM third-party dependency so that the Swift package can locate TVM headers.

android/mlc4j/prepare_libs.py -- A Python script that configures CMake for Android cross-compilation using the Android NDK toolchain. It registers the aarch64-linux-android Rust target, invokes CMake with OpenCL GPU support enabled, builds the tvm4j_runtime_packed target, and installs the resulting libraries. The script requires the ANDROID_NDK environment variable to be set and optionally accepts TVM_SOURCE_DIR for custom TVM builds.

Usage

Use these scripts when:

Initially setting up the MLC-LLM mobile build environment on a new development machine
Rebuilding the native inference libraries after updating the MLC-LLM source code
Switching between iOS device and simulator targets
The mlc_llm package command needs to invoke the platform-specific build step as part of the full packaging pipeline

Code Reference

Source Location

Repository: MLC-LLM
File (iOS): ios/prepare_libs.sh (Lines 3-79)
File (Android): android/mlc4j/prepare_libs.py (Lines 86-107)

Signature (iOS)

# ios/prepare_libs.sh
# OPTION:
#   -s, --simulator                      Build for Simulator
#   -a, --arch        x86_64 | arm64     Simulator arch
#   -h, --help                           Prints this help
#
# Environment Variables:
#   MLC_LLM_SOURCE_DIR  Path to MLC-LLM source (default: "..")

bash ios/prepare_libs.sh [-s|--simulator] [-a|--arch x86_64|arm64]

Signature (Android)

def main(mlc_llm_source_dir: Path):
    """Build MLC LLM and tvm4j native libraries for Android.

    Sets up Rust target, creates build directory, configures CMake
    with Android NDK toolchain, and builds tvm4j_runtime_packed.
    """

Command-line invocation:

python android/mlc4j/prepare_libs.py --mlc-llm-source-dir <path>

I/O Contract

Inputs (iOS)

Name	Type	Required	Description
`-s` / `--simulator`	flag	No	When set, builds for the iOS Simulator instead of a physical device
`-a` / `--arch`	string	No	Simulator architecture: `x86_64` or `arm64` (default: `arm64`)
`MLC_LLM_SOURCE_DIR`	env var	No	Path to MLC-LLM source root (default: `..`)

Inputs (Android)

Name	Type	Required	Description
`--mlc-llm-source-dir`	Path	No	Path to MLC-LLM source root (default: auto-detected or from `MLC_LLM_SOURCE_DIR` env var)
`ANDROID_NDK`	env var	Yes	Path to the Android NDK installation directory
`TVM_SOURCE_DIR`	env var	No	Optional path to a custom TVM source directory

Outputs

Name	Type	Description
`build/lib/libmlc_llm_static.a` (iOS)	Static library	Compiled MLC-LLM inference engine for iOS with Metal GPU support
`build/lib/*.a` (iOS)	Static libraries	Additional TVM runtime static libraries
`ios/MLCSwift/tvm_home` (iOS)	Symlink	Symbolic link to `3rdparty/tvm` for Swift package resolution
`build/output/` (Android)	Directory	Compiled shared libraries and JNI bindings for Android

Usage Examples

Basic Usage (iOS Device)

# Build for a physical iOS device (arm64)
cd ios
export MLC_LLM_SOURCE_DIR=..
bash prepare_libs.sh

iOS Simulator (Apple Silicon)

# Build for iOS Simulator on Apple Silicon Mac
cd ios
bash prepare_libs.sh --simulator --arch arm64

iOS Simulator (Intel)

# Build for iOS Simulator on Intel Mac
cd ios
bash prepare_libs.sh --simulator --arch x86_64

Basic Usage (Android)

# Build for Android (requires ANDROID_NDK to be set)
export ANDROID_NDK=/path/to/android-ndk
python android/mlc4j/prepare_libs.py --mlc-llm-source-dir /path/to/mlc-llm

Related Pages

Implements Principle

Principle:Mlc_ai_Mlc_llm_Mobile_Build_Environment

Environment Links

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment