This package contains high-level Rust bindings for [wasi-nn] system calls. It is similar in purpose to the [WASI bindings] but this package provides optional access to a system's machine learning functionality from WebAssembly.
NOTE: These bindings are experimental (use at your own risk) and subject to upstream changes in the wasi-nn specification.
Depend on this crate in your Cargo.toml:
toml
[dependencies]
wasi-nn = "0.6.0"
Use the wasi-nn APIs in your application, for example:
rust
use wasi_nn;
let graph = GraphBuilder::new(GraphEncoding::TensorflowLite, ExecutionTarget::CPU)
.build_from_files([model_path])?;
let mut ctx = graph.init_execution_context()?;
ctx.set_input(0, TensorType::F32, &input_dims, &input_buffer)?;
ctx.compute()?;
let output_num_bytes = ctx.get_output(0, &mut output_buffer)?;
Compile the application to WebAssembly:
shell script
cargo build --target=wasm32-wasi
Run the generated WebAssembly in a runtime supporting [wasi-nn], e.g., [Wasmtime].
To build this crate from source, use: cargo build --target wasm32-wasi.
This crate contains code (src/generated.rs) generated by
witx-bindgen. To
regenerate this code, run the following script:
console
$ scripts/regenerate-bindings-from-witx.sh
The [examples] demonstrate how to use wasi-nn from a Rust program.
This project is licensed under the Apache 2.0 license. See [LICENSE] for more details.
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in this project by you, as defined in the Apache-2.0 license, shall be licensed as above, without any additional terms or conditions.