Package com.codedstream.otterstream.onnx

Class OnnxModelLoader

java.lang.Object

com.codedstream.otterstream.onnx.OnnxModelLoader

All Implemented Interfaces:

ModelLoader<InferenceSession>

public class OnnxModelLoader
extends Object
implements ModelLoader<InferenceSession>

ModelLoader implementation for loading ONNX models.

This loader handles the specifics of loading ONNX models from various sources (file system, input streams) and configuring ONNX Runtime sessions with optimized settings. It extracts model metadata and validates loaded models.

Loading Sources:

• File Path: Load models from local filesystem or network paths
• InputStream: Load models from memory streams or network streams

Session Configuration:

Supports configuration through ModelConfig model options:

• interOpThreads: Number of threads for inter-operation parallelism
• intraOpThreads: Number of threads for intra-operation parallelism
• optimizationLevel: "disable", "basic", "extended", or "all"
• useGpu: Boolean flag to enable GPU execution (CUDA)

Usage Example:

 OnnxModelLoader loader = new OnnxModelLoader();

 // Configure with options
 ModelConfig config = ModelConfig.builder()
     .modelPath("model.onnx")
     .modelOption("interOpThreads", 2)
     .modelOption("intraOpThreads", 4)
     .modelOption("optimizationLevel", "all")
     .modelOption("useGpu", true)
     .build();

 // Load model
 InferenceSession session = loader.loadModel(config);

 // Validate and get metadata
 if (loader.validateModel(session, config)) {
     ModelMetadata metadata = loader.getModelMetadata(session);
 }
 

Supported Formats:

This loader only supports ModelFormat.ONNX format.

Model Validation:

Validates that loaded models have at least one input and one output. More comprehensive validation can be added by extending this class.

Error Handling:

All loading failures throw ModelLoadException with detailed error messages and root causes.

Since:

1.0.0

Author:

Nestor Martourez, Sr Software and Data Streaming Engineer @ CodedStreams

See Also:

ModelLoader, InferenceSession, OnnxInferenceEngine

Constructor Summary

Constructors

Constructor	Description
OnnxModelLoader()

Method Summary

Modifier and Type	Method	Description
ModelMetadata	getModelMetadata(InferenceSession model)	Extracts metadata from a loaded ONNX model.
ModelFormat[]	getSupportedFormats()	Gets the model formats supported by this loader.
InferenceSession	loadModel(ModelConfig config)	Loads an ONNX model from the path specified in configuration.
InferenceSession	loadModel(InputStream inputStream, ModelConfig config)	Loads an ONNX model from an input stream.
boolean	validateModel(InferenceSession model, ModelConfig config)	Validates that a loaded ONNX model matches the configuration.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

OnnxModelLoader

public OnnxModelLoader()

Method Detail

loadModel

public InferenceSession loadModel(ModelConfig config)
                           throws ModelLoadException

Loads an ONNX model from the path specified in configuration.

Specified by:

loadModel in interface ModelLoader<InferenceSession>

Parameters:

config - model configuration containing path and options

Returns:

loaded InferenceSession instance

Throws:

ModelLoadException - if loading fails

loadModel

public InferenceSession loadModel(InputStream inputStream,
                                  ModelConfig config)
                           throws ModelLoadException

Loads an ONNX model from an input stream.

Useful for loading models from memory or network sources.

Specified by:

loadModel in interface ModelLoader<InferenceSession>

Parameters:

inputStream - stream containing ONNX model data

config - model configuration

Returns:

loaded InferenceSession instance

Throws:

ModelLoadException - if loading fails

validateModel

public boolean validateModel(InferenceSession model,
                             ModelConfig config)

Validates that a loaded ONNX model matches the configuration.

Basic validation checks that the model has at least one input and one output.

Specified by:

validateModel in interface ModelLoader<InferenceSession>

Parameters:

model - the loaded InferenceSession

config - model configuration (not used in basic validation)

Returns:

true if model has both inputs and outputs

getSupportedFormats

public ModelFormat[] getSupportedFormats()

Gets the model formats supported by this loader.

Specified by:

getSupportedFormats in interface ModelLoader<InferenceSession>

Returns:

array containing only ModelFormat.ONNX

getModelMetadata

public ModelMetadata getModelMetadata(InferenceSession model)

Extracts metadata from a loaded ONNX model.

Extracts input and output schemas from model metadata. If extraction fails, returns minimal metadata with empty schemas.

Specified by:

getModelMetadata in interface ModelLoader<InferenceSession>

Parameters:

model - the loaded InferenceSession

Returns:

model metadata including input/output schemas