Crate mnn_rs 

MNN Rust Bindings

Safe Rust bindings for Alibaba’s MNN (Mobile Neural Network) inference engine.

MNN is a highly efficient and lightweight deep learning inference framework. This crate provides idiomatic Rust bindings for running inference with MNN.

Features

• Safe API: All MNN operations are wrapped in safe Rust types
• Multiple Backends: CPU, CUDA, OpenCL, Vulkan, Metal support
• Async Support: Optional async API using tokio
• Cross-Platform: Windows, Linux, macOS, Android, iOS support

Quick Start

use mnn_rs::{Interpreter, ScheduleConfig, BackendType};

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

// Create a session
let config = ScheduleConfig::new()
    .backend(BackendType::CPU)
    .num_threads(4);

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

// Get input tensor
let input = session.get_input(None)?;

// Fill input with data (example)
// input.write(&my_data)?;

// Run inference
session.run()?;

// Get output
let output = session.get_output(None)?;

Backend Configuration

use mnn_rs::{ScheduleConfig, BackendType, MemoryMode, PrecisionMode};

// CPU with custom settings
let cpu_config = ScheduleConfig::new()
    .backend(BackendType::CPU)
    .num_threads(8)
    .memory_mode(MemoryMode::Low);

// GPU (auto-detect best backend)
let gpu_config = ScheduleConfig::new()
    .backend(BackendType::Auto)
    .precision_mode(PrecisionMode::Low);

Async API (requires “async” feature)

use mnn_rs::{AsyncInterpreter, ScheduleConfig};

#[tokio::main]
async fn main() -> Result<(), mnn_rs::MnnError> {
    let interpreter = AsyncInterpreter::from_file("model.mnn").await?;
    let mut session = interpreter.create_session(ScheduleConfig::default()).await?;

    session.run_async().await?;

    Ok(())
}

Re-exports

pub use mnn_rs_sys;

Modules

prelude

Prelude for common MNN types. Common types for MNN operations.

Structs

BackendCapabilities

Backend capabilities.

BackendConfig

Configuration for a compute backend.

Interpreter

A model interpreter that holds a loaded neural network model.

ScheduleConfig

Schedule configuration for creating sessions.

ScheduleConfigBuilder

Builder for creating schedule configurations.

Session

An inference session.

SessionGuard

A guard for ensuring session resources are properly managed.

Tensor

A multi-dimensional array for neural network operations.

TensorInfo

Information about a tensor’s shape and type.

TensorView

A view into a tensor’s data without ownership.

Enums

BackendType

Compute backend type.

DataFormat

Data format for tensors.

DataType

Data type for tensor elements.

Error

Main error type for MNN operations.

MemoryMode

Power usage mode.

MnnError

Main error type for MNN operations.

PowerMode

Power usage mode.

PrecisionMode

Precision mode for inference.

SessionMode

Session mode for controlling interpreter behavior.

Traits

TensorData

Trait for types that can be stored in a tensor.

Functions

available_backends

Get list of available backends on this system.

calculate_element_count

Calculate the total number of elements in a tensor.

calculate_tensor_size

Calculate the size of a tensor given its shape and element size.

convert_format

Convert between NHWC and NCHW formats.

is_backend_available

Check if a specific backend is available.

version

Get the MNN version string.

Type Aliases

MnnResult

Result type alias for MNN operations.