MNN Rust Bindings

Safe Rust bindings for the MNN (Mobile Neural Network) inference library.

Overview

This crate provides a high-level, safe Rust interface to MNN, built on top of mnn-sys which provides the low-level FFI bindings.

Features

• Safe API: All unsafe operations are wrapped in safe Rust abstractions
• Zero-copy operations: Efficient tensor data transfer
• Platform support: CPU, Metal (macOS/iOS), CoreML, OpenCL, CUDA, and more
• Configurable backends: Choose the best backend for your platform

Usage

use mnn::{Interpreter, ScheduleConfig, ForwardType, BackendConfig, PrecisionMode, PowerMode};
use std::path::Path;

// Load model
let interpreter = Interpreter::from_file(Path::new("model.mnn"))?;

// Configure session
let mut config = ScheduleConfig::new();
config.set_type(ForwardType::Auto);

let mut backend_config = BackendConfig::new();
backend_config.set_precision_mode(PrecisionMode::High);
backend_config.set_power_mode(PowerMode::High);
config.set_backend_config(backend_config);

// Create session
let mut session = interpreter.create_session(config)?;

// Get input tensor
let mut input_tensor = interpreter.input(&mut session, "input")?;

// ... prepare input data ...

// Run inference
interpreter.run_session(&mut session)?;

// Get output
let output_tensor = interpreter.output(&session, "output")?;

Platform-Specific Features

macOS/iOS

Enable Metal and CoreML backends:

[dependencies]
mnn = { path = "crates/mnn", features = ["metal", "coreml"] }

Linux/Android

OpenCL backend is enabled by default on these platforms.

Building

This crate requires:

• CMake 3.10+
• C++11 compiler
• Platform-specific SDK (e.g., Metal SDK for macOS)

The MNN library is compiled from source during the build process.