Crate edgefirst_tflite 

Ergonomic Rust API for TensorFlow Lite inference.

This crate provides a safe, idiomatic wrapper around the TFLite C API with support for hardware-accelerated inference via delegates, DMA-BUF zero-copy, and NPU preprocessing.

Quick Start

use edgefirst_tflite::{Library, Model, Interpreter};

let lib = Library::new()?;
let model = Model::from_file(&lib, "model.tflite")?;

let mut interpreter = Interpreter::builder(&lib)?
    .num_threads(4)
    .build(&model)?;

// Populate input tensors...
interpreter.invoke()?;

let outputs = interpreter.outputs()?;

Feature Flags

Feature	Description
dmabuf	DMA-BUF zero-copy inference via HAL Delegate API
camera_adaptor	NPU-accelerated format conversion
metadata	TFLite model metadata extraction
full	Enables all optional features

Re-exports

pub use delegate::Delegate;

pub use delegate::DelegateOptions;

pub use error::Error;

pub use error::StatusCode;

pub use interpreter::Interpreter;

pub use interpreter::InterpreterBuilder;

pub use library::Library;

pub use model::Model;

pub use tensor::QuantizationParams;

pub use tensor::Tensor;

pub use tensor::TensorMut;

pub use tensor::TensorType;

Modules

delegate

Delegate loading with configuration options.

error

Error types for the edgefirst-tflite crate.

interpreter

Interpreter and builder for TFLite model inference.

library

Safe wrapper around the TFLite shared-library handle.

model

Model loading for TFLite inference.

tensor

Type-safe tensor wrappers for the TensorFlow Lite C API.