VideoLLaMA2Options
Options & Configuration · AiDotNet.VisionLanguage.VideoLanguage
Configuration options for VideoLLaMA 2: spatial-temporal convolution for video tokens.
Inheritance: ModelOptions → GenerativeVLMOptions → VideoLanguageOptions → VideoLLaMA2Options
For Beginners
These options configure the VideoLLaMA2 model. Default values follow the original paper settings.
Constructors
| Constructor | Summary |
|---|---|
VideoLLaMA2Options() |
Initializes a new instance with default values. |
VideoLLaMA2Options(VideoLLaMA2Options other) |
Initializes a new instance by copying from another instance. |
Properties
| Property | Summary |
|---|---|
bool EnableSpatialTemporalConv { get; set; } |
Gets or sets whether to use spatial-temporal convolution for video token compression. |
double RoPETheta { get; set; } |
Gets or sets the Mistral rotary-position base frequency. |
int DecoderFfnDim { get; set; } |
Gets or sets the Mistral decoder SwiGLU intermediate width. |
int DecoderNumHeads { get; set; } |
Gets or sets the number of language-decoder query heads. |
int DecoderNumKeyValueHeads { get; set; } |
Gets or sets the number of language-decoder key/value heads. |
int PatchSize { get; set; } |
Gets or sets the square vision patch size. |
int STCKernelSize { get; set; } |
Gets or sets the uniform temporal/spatial STC Conv3D kernel size. |
int STCMlpDepth { get; set; } |
Gets or sets the number of linear layers in the STC readout MLP. |
int STCPadding { get; set; } |
Gets or sets the uniform temporal/spatial STC Conv3D padding. |
int STCStageDepth { get; set; } |
Gets or sets the number of RegNet blocks in each STC spatial stage. |
int STCStride { get; set; } |
Gets or sets the uniform temporal/spatial STC Conv3D stride. |
int VisionFfnDim { get; set; } |
Gets or sets the CLIP vision-transformer feed-forward width. |
int VisionNumHeads { get; set; } |
Gets or sets the number of CLIP vision-attention heads. |
string VisionEncoderName { get; set; } |
Gets or sets the vision encoder identifier. |