OnnxFullModelAdapter<T>

Models & Types · AiDotNet.ModelLoading.Pretrained

Adapts an OnnxModel (an IOnnxModel that runs an ONNX graph through the ONNX runtime) to the IFullModel, ILanguageModel, and IIncrementalCausalLanguageModel contracts so ConfigureModel(PretrainedSource.Onnx(path)) supports full-context and KV-cached causal language model generation behind the AiModelResult facade.

Implements: IFullModel<T, Tensor<T>, Tensor<T>>, IIncrementalCausalLanguageModel<T>, ILanguageModel<T>

Constructors

Constructor Summary
OnnxFullModelAdapter(byte[] onnxBytes, OnnxModelOptions options) Creates an adapter over the ONNX graph in onnxBytes.

Properties

Property Summary
ILossFunction<T> DefaultLossFunction { get; }
OnnxModel<T> Model { get; } The wrapped ONNX runtime model.
bool SupportsParameterInitialization { get; } Always false: a frozen ONNX graph cannot be initialized from a flat vector.
int VocabularySize { get; }
long ParameterCount { get; } Always zero: the graph owns its weights, this adapter does not.

Methods

Method Summary
Dictionary<string, T> GetFeatureImportance()
IEnumerable<int> GetActiveFeatureIndices()
IFullModel<T, Tensor<T>, Tensor<T>> Clone()
IFullModel<T, Tensor<T>, Tensor<T>> DeepCopy()
IFullModel<T, Tensor<T>, Tensor<T>> WithParameters(Vector<T> parameters)
ModelMetadata<T> GetModelMetadata()
OnnxFullModelAdapter<T> FromFile(string path, OnnxModelOptions options) Creates an adapter over the ONNX graph at path.
Tensor<T> ForwardLogits(Tensor<T> tokenIds)
Tensor<T> Predict(Tensor<T> input)
Vector<T> AppendToken(int tokenId)
Vector<T> GetParameters() An empty vector, matching ParameterCount.
Vector<T> NextTokenLogits(IReadOnlyList<int> tokenIds)
Vector<T> SanitizeParameters(Vector<T> parameters)
Vector<T> StartSequence(IReadOnlyList<int> promptTokenIds)
bool IsFeatureUsed(int featureIndex)
byte[] Serialize()
void Deserialize(byte[] data)
void Dispose()
void LoadModel(string filePath)
void LoadState(Stream stream)
void ResetCache()
void SaveModel(string filePath)
void SaveState(Stream stream)
void SetActiveFeatureIndices(IEnumerable<int> featureIndices)
void SetParameters(Vector<T> parameters) Writes a parameter vector into this adapter.
void Train(Tensor<T> input, Tensor<T> expectedOutput)