InstructionTunedVLMOptions

Options & Configuration · AiDotNet.VisionLanguage.InstructionTuned

Base configuration options for instruction-tuned vision-language models.

Inheritance: ModelOptions → GenerativeVLMOptions → InstructionTunedVLMOptions

Known derived types: AquilaVLOptions, AriaOptions, Cambrian1Options, CogVLM2Options, CogVLMOptions, DeepSeekVL2Options, DeepSeekVLOptions, DragonflyOptions, Eagle25Options, EagleOptions, FuyuOptions, Gemma3Options, InternVL25Options, InternVL2Options, InternVL3Options, InternVLOptions, LLaVA15Options, LLaVANeXTOptions, LLaVAOneVision15Options, LLaVAOneVisionOptions, Llama32VisionOptions, MPLUGOwl2Options, MPLUGOwl3Options, MPLUGOwlOptions, MantisOptions, MayaOptions, MiniCPMVOptions, MiniCPMoOptions, MiniGPT4Options, MiniGPTv2Options, MolmoOptions, MonkeyOptions, MoondreamOptions, NVLMOptions, OvisOptions, Phi3VisionOptions, Phi4MultimodalOptions, PixtralLargeOptions, PixtralOptions, Qwen25VLOptions, Qwen2VLOptions, Qwen3VLOptions, QwenVLOptions, ReasoningVLMOptions, SmolVLMOptions, VILAOptions, VILAUOptions

For Beginners

These options configure the InstructionTuned model. Default values follow the original paper settings.

Constructors

Constructor Summary
InstructionTunedVLMOptions() Initializes a new instance with default values.
InstructionTunedVLMOptions(InstructionTunedVLMOptions other) Initializes a new instance by copying from another instance.

Properties

Property Summary
InstructionTunedArchitectureType InstructionArchitectureType { get; set; } Gets or sets the instruction-tuned architecture type.
int MaxVisualTokens { get; set; } Gets or sets the maximum number of visual tokens per image.
int ProjectionDim { get; set; } Gets or sets the MLP projection hidden dimension (for MLP connector models).
string LanguageModelName { get; set; } Gets or sets the language model backbone name.
string SystemPrompt { get; set; } Gets or sets the system prompt for chat mode.

Methods

Method Summary
void ValidateVisualSizing() Fail-fast validation of the size-related options before any layer is allocated.