WACS.WASI.NN.MLNet 0.2.4

WACS.WASI.NN.MLNet

ML.NET-flavored ONNX backend for WACS.WASI.NN. Implements IBackend for graph-encoding.onnx against Microsoft.ML.OnnxTransformer under an MLContext lifecycle — for embedders composing wasi-nn inference with the rest of an ML.NET pipeline (preprocessing transformers, custom predictors, IDataView / PredictionEngine).

For raw tensor inference with no pipeline integration, prefer WACS.WASI.NN.OnnxRuntime — it avoids the Microsoft.ML transitive surface (~70 MB lighter).

Install

dotnet add package WACS.WASI.NN.MLNet

The package's bin ships its NuGet transitives + RID-specific native libs (via <EnableDynamicLoading>true</EnableDynamicLoading>), so Assembly.LoadFrom resolves everything from the LoadFromContext probe.

CLI

# After dotnet build of this project's repo:
MLNET=$(realpath Wacs.WASI/Wacs.WASI.NN/Wacs.WASI.NN.MLNet/bin/Release/net8.0/Wacs.WASI.NN.MLNet.dll)

wacs run my.component.wasm --wasip2 --bind "$MLNET" -d ./models::/models

--bind auto-pulls the WASI.NN typed surface + DI sibling onto host-packages when the identity starts with Wacs.WASI.NN.. The Preview 2 DI scope wires the ML.NET-backed ORT into the DI bundle's WasiNNConfiguration.Backends[ONNX].

Embedder

Interpreter / one-line:

using Wacs.Core.Runtime;
using Wacs.WASI.NN;
using Wacs.WASI.NN.MLNet;
using Wacs.WASI.NN.Types;

var runtime = new WasmRuntime();
runtime.UseWasiNN(b => b.AddBackend(GraphEncoding.ONNX, new MLNetBackend()));

What it provides

• MLNetBackend : IBackend — wraps ORT under MLContext.Transforms.ApplyOnnxModel / IDataView, exposing the same LoadGraph(builders, target) / Compute(inputs) shape as the bare ORT backend. Embedders who want the ML.NET preprocessing surface get it; the rest of WACS doesn't notice
• WasiNNMLNetBindable : IBindable — parameterless adapter for --bind
• [assembly: WasiHostPackage]

Why ML.NET over bare ORT

Use this backend when:

• Your wasm guest is one stage in a longer ML.NET pipeline (custom transformers, preprocessing, IDataView consumers) and you want them composed in one host-side process
• You're already bringing in Microsoft.ML for adjacent work — the marginal cost of routing wasi-nn through MLContext is small
• You want MLContext.Log / structured ML.NET diagnostics around the inference call

For everything else — image classification, embeddings, encoder-only LLMs, raw tensor in / raw tensor out — bare WACS.WASI.NN.OnnxRuntime is lighter and simpler.

Backend choice

Documentation

• docs/WASI_NN_USAGE.md — unified usage guide (CLI flags, env vars, programmatic embedding, worked examples)
• docs/COMPONENT_CHAINING.md
• Wacs.WASI/Wacs.WASI.NN/README.md — backend matrix + package layout

License

Apache-2.0