Ortex

Ortex is a wrapper around ONNX Runtime (implemented as bindings to ort). Ortex leverages Nx.Serving to easily deploy ONNX models that run concurrently and distributed in a cluster. Ortex also provides a storage-only tensor implementation for ease of use.

ONNX models are a standard machine learning model format that can be exported from most ML libraries like PyTorch and TensorFlow. Ortex allows for easy loading and fast inference of ONNX models using different backends available to ONNX Runtime such as CUDA, TensorRT, Core ML, and ARM Compute Library.

Examples

TL;DR:

iex> model = Ortex.load("./models/resnet50.onnx")
#Ortex.Model<
  inputs: [{"input", "Float32", [nil, 3, 224, 224]}]
  outputs: [{"output", "Float32", [nil, 1000]}]>
iex> {output} = Ortex.run(model, Nx.broadcast(0.0, {1, 3, 224, 224}))
iex> output |> Nx.backend_transfer() |> Nx.argmax
#Nx.Tensor<
  s64
  499
>

Inspecting a model shows the expected inputs, outputs, data types, and shapes. Axes with nil represent a dynamic size.

To see more real world examples see the examples folder.

Serving

Ortex also implements Nx.Serving behaviour. To use it in your application's supervision tree consult the Nx.Serving docs.

iex> serving = Nx.Serving.new(Ortex.Serving, model)
iex> batch = Nx.Batch.stack([{Nx.broadcast(0.0, {3, 224, 224})}])
iex> {result} = Nx.Serving.run(serving, batch)
iex> result |> Nx.backend_transfer() |> Nx.argmax(axis: 1)
#Nx.Tensor<
  s64[1]
  [499]
>

Installation

Ortex can be installed by adding ortex to your list of dependencies in mix.exs:

def deps do
  [
    {:ortex, "~> 0.1.11"}
  ]
end

Precompiled NIFs

Starting with 0.1.11, Ortex ships precompiled NIF artifacts for common platforms. If you're on one of the supported targets you do not need a Rust toolchain — mix deps.get downloads a prebuilt binary directly.

Supported targets:

• aarch64-apple-darwin (Apple Silicon macOS)
• x86_64-apple-darwin (Intel macOS)
• x86_64-unknown-linux-gnu (Linux x86_64, glibc >= 2.35)
• aarch64-unknown-linux-gnu (Linux ARM64, glibc >= 2.35)
• x86_64-pc-windows-msvc (Windows x86_64, MSVC ABI)

onnxruntime is statically linked into the precompiled NIF, so the artifact is a single self-contained shared library — no separate runtime install required.

Source build

You'll need Rust and a working C toolchain to build from source. This happens automatically when:

• Your target tuple isn't in the precompiled list above (e.g. musl Linux, 32-bit ARM, FreeBSD).

• You explicitly request it with ORTEX_BUILD=true:

  ORTEX_BUILD=true mix deps.compile ortex --force

The source path is the same flow that's shipped since 0.1.0 — cargo build downloads onnxruntime and the resulting NIF is dropped into priv/native/. Useful for development on the NIF crate itself, or when you need a feature flag (cuda, tensorrt, coreml, directml) that the precompiled CPU-only artifact doesn't include.