//clika-runtime/io.clika.runtime/OnnxModel

OnnxModel

[common]
class OnnxModel

An ONNX model opened from a file: the graph loads into memory and the external initializer payloads stay on disk until something consumes them; nothing is validated beyond the wire format, and compile is where operator support is decided. AutoCloseable: close releases the parsed model (a graph it compiled lives on).

OnnxModel.open("model.onnx").use { model ->
    model.compile(CompileOptions(sampleInputs = mapOf("x" to x))).use { graph ->
        graph.finalizeGraph()
        val y = graph.run(listOf(x)).single()
    }
}

Types

Name	Summary
Companion	[common] object Companion

Properties

Name	Summary
irVersion	[common] val irVersion: Long The file's ONNX IR version.
numInitializers	[common] val numInitializers: Long The initializer (weight) count.
numNodes	[common] val numNodes: Long The graph's node count.
opset	[common] val opset: Long The default (ai.onnx) opset version.

Functions

Name	Summary
close	[common] open fun close()
compile	[common] fun compile(options: CompileOptions = CompileOptions()): ModelGraph Build the executable form: every node parsed into its runtime operator (an unsupported one fails here, naming it), the weights bound, the shapes resolved per options. The graph comes back as built: optimize it when wanted, then ModelGraph.finalizeGraph it to run. The model is unchanged and reusable.
inputs	[common] fun inputs(): List<TensorSpec> The graph inputs as specs, declaration order (a dynamic dim reads as TensorSpec.DYNAMIC_DIM).
optimize	[common] fun optimize(): Int Run the ONNX rewrite passes to a fixed point (double casts collapsed, redundant quantize/dequantize pairs dropped, shape movement fused, constants folded, dead nodes removed); the number of passes run.
outputs	[common] fun outputs(): List<TensorSpec> The graph outputs as specs, declaration order.
save	[common] fun save(path: String) Assemble the model and write it to path (the payloads inline or in a sidecar per OnnxModelOptions.saveAsExternalData).