OnnxMultimodalConfiguration
Options & Configuration · AiDotNet.Onnx
The effective host dimensions and immutable signatures of loaded multimodal graphs.
How It Works
This is separate from the caller’s requested options. Graph signatures do not infer internal patch size, layer counts, attention heads or vocabulary tables from I/O shapes.
Symbolic dimensions remain unknown here; the wrapper validates actual outputs when executing them. Tokenizer vocabulary size describes the supplied tokenizer, not a claim that an opaque graph contains a matching embedding table.
Properties
| Property | Summary |
|---|---|
IReadOnlyDictionary<OnnxModelRole, OnnxGraphSignature> Graphs { get; } |
Gets graph signatures by role. |
int EmbeddingDimension { get; } |
Gets the host’s validated embedding width. |
int ImageChannels { get; } |
Gets the host’s image channel count. |
int ImageSize { get; } |
Gets the host’s configured square-image size. |
int MaxSequenceLength { get; } |
Gets the host’s configured text-token context. |
int TokenizerVocabularySize { get; } |
Gets the supplied tokenizer’s vocabulary size, not an inferred graph vocabulary. |
int? NumFrames { get; } |
Gets the host’s selected frame count, or null for a non-video wrapper. |