Expand description
Backend-independent model, session, named-I/O, and explicit binding contracts.
The default build contains no inference runtime. ONNX Runtime and hardware execution providers are additive features and must preserve the copy/device choices represented by these types.
Structs§
- IoBinding
- Inputs, requested output destinations, and completed named results.
- Mock
Inference Backend - Backend that never loads ONNX bytes and always uses
MockProfile. - Model
Info - Model-visible named input and output metadata.
- Named
Tensors - Ordered, uniquely named tensor collection.
- Onnx
Runtime Backend - ONNX Runtime backend configured explicitly for the CPU execution provider.
- Onnx
Runtime Session - Loaded ONNX Runtime CPU session.
- Preallocated
Output - Caller-owned mutable bytes for an explicitly bound CPU output.
- RunOptions
- Per-run host copy permissions.
- Session
Options - Runtime-independent session configuration.
- Tensor
Spec - One named model input or output contract.
Enums§
- AiError
- Errors shared by inference contracts and backend adapters.
- Copy
Policy - Whether a run may allocate and copy host tensor data.
- Dimension
- One model dimension, fixed, unconstrained, or symbolically dynamic.
- Graph
Optimization - Graph optimization level requested from a backend.
- Mock
Profile - Built-in mock model profiles selected through
ModelSource::Mock. - Model
Source - Model bytes, filesystem path, or built-in mock profile for session creation.
- Output
Binding - Explicit output destination for an I/O-bound run.
Traits§
- Inference
Backend - Stable interface implemented by inference engines.
- Model
Session - Loaded model session with named dynamic I/O.
Type Aliases§
- AiResult
- Result type for inference operations.