Struct OnnxRuntime 

pub struct OnnxRuntime;

ONNX Runtime environment wrapper

Manages the global ONNX Runtime environment and provides helper methods for loading models.

Uses load-dynamic feature: the runtime library is loaded at runtime. The library is resolved in this order:

1. ORT_DYLIB_PATH environment variable (if set)
2. libs/onnxruntime/<LIB> relative to the executable
3. libs/onnxruntime/<LIB> relative to the current working directory

Implementations

impl OnnxRuntime

pub fn init() -> Result<Self>

Initialize the ONNX Runtime global environment.

Dynamically loads libonnxruntime.so at runtime. The library is searched in the order described on OnnxRuntime.

This should be called once at application startup, before creating any sessions.

pub fn load_model_with_threads<P: AsRef<Path>>( &self, path: P, intra_threads: usize, ) -> Result<Session>

Load an ONNX model with a specific number of intra-op threads.

Attempts to register GPU execution providers in platform-specific priority order, then falls back to CPU if none initialize. The CPU provider is always appended last so session creation never fails due to missing GPU drivers or runtime libraries.

Platform priority: Windows: DirectML (covers NVIDIA/AMD/Intel/Qualcomm via D3D12) Linux: CUDA (NVIDIA). ROCm could be added via a cargo flag later. macOS: CoreML (Apple Silicon + AMD on Intel Macs)

intra_threads applies to the CPU provider; GPU providers ignore it.