InternViTOptions
Options & Configuration · AiDotNet.VisionLanguage.Encoders
Configuration options for InternViT, the vision encoder used in the InternVL series.
Inheritance: ModelOptions → VisionEncoderOptions → InternViTOptions
How It Works
InternViT (Chen et al., 2024) is a 6B-parameter ViT designed for progressive alignment with LLMs. It uses dynamic resolution processing and pixel shuffle downsampling to handle images of varying sizes efficiently as part of the InternVL architecture.
Constructors
| Constructor | Summary |
|---|---|
InternViTOptions() |
Initializes a new instance with default values. |
InternViTOptions(InternViTOptions other) |
Initializes a new instance by copying from another instance. |
Properties
| Property | Summary |
|---|---|
bool Use3DRoPE { get; set; } |
Gets or sets whether to use 3D-RoPE for positional encoding. |
bool UseDynamicResolution { get; set; } |
Gets or sets whether to use dynamic resolution tiling. |
int MaxTiles { get; set; } |
Gets or sets the maximum number of tiles for dynamic resolution. |
int PixelShuffleRatio { get; set; } |
Gets or sets the pixel shuffle downsampling ratio. |