Skip to content
NexisNexis Wiki

nexis-ml-rs (Rust)

nexis-ml-rs is a Python-free, single-binary ML engine for Nexis — the same job as nexis-ml, but as one downloadable binary with no Python or PyTorch toolchain required. It’s designed like a downloadable language server: fetch it, and the ML Lab panel can train and run models on a machine that has no Python at all.

It speaks the same NDJSON protocol and writes the same run store as the Python engine, so Nexis renders runs from either with zero changes — a Rust-produced run can be read by the Python nexis-ml runs.

Backends

Section titled “Backends”

Training runs on burn. Pick the backend with [train] device in train.toml:

  • auto — choose automatically.
  • cpu — burn’s ndarray backend.
  • gpu — burn’s wgpu backend (Vulkan / DX12 / Metal), with no vendor toolchain required.

Both backends include autodiff.

Models

Section titled “Models”

The model is declared in train.toml:

  • A CSV [data] path (or none → synthetic data) trains a variable-depth MLP.
  • A folder of class sub-folders trains a CNN over its images.

ONNX export is available for the tabular MLP (export --onnx), and Nexis can download the binary and detect it automatically.

Build & run

Section titled “Build & run”
Terminal window
cargo build --release            # produces target/release/nexis-ml(.exe)


nexis-ml --version               # → "nexis-ml 0.8.0" (Nexis-detectable)
nexis-ml env                     # one-line JSON capability report (backend: cpu|wgpu)
nexis-ml new tabular my-run      # scaffold a project (templates: tabular | image)
nexis-ml train my-run            # train; writes .nexis-ml/runs/<id>/
nexis-ml --nexis-protocol train my-run   # stream NDJSON protocol on stdout
nexis-ml export --onnx my-run    # train the MLP and write my-run/model.onnx
nexis-ml serve --run <id> my-run # inference loop: one JSON request per stdin line

When to use it

Section titled “When to use it”

Use nexis-ml-rs when you want a single ~31 MB binary and no Python dependency — for example on a fresh machine, a CI runner, or a container. When you want the full feature set (including the tiny GPT text template), use the Python engine nexis-ml. The ML Lab panel auto-detects whichever is installed.