Struct VisionTransformerConfig 

pub struct VisionTransformerConfig {
    pub patch_embed: PatchEmbeddingConfig,
    pub encoder: EncoderStackConfig,
    pub num_classes: usize,
    pub use_class_token: bool,
    pub classifier_dropout: f64,
}

Configuration for Vision Transformer

Fields

patch_embed: PatchEmbeddingConfig

Patch embedding configuration

encoder: EncoderStackConfig

Transformer encoder stack configuration

num_classes: usize

Number of output classes

use_class_token: bool

Whether to use class token

classifier_dropout: f64

Dropout rate for classification head

Implementations

impl VisionTransformerConfig

pub fn new( image_size: usize, patch_size: usize, in_channels: usize, d_model: usize, n_heads: usize, d_ff: usize, n_layers: usize, num_classes: usize, ) -> Result<Self>

Create new Vision Transformer configuration

pub fn with_class_token(self, use_class_token: bool) -> Self

Builder: Set whether to use class token

pub fn with_classifier_dropout(self, dropout: f64) -> Self

Builder: Set classifier dropout

pub fn with_learned_position_encoding(self) -> Self

Builder: Set learned position encoding

pub fn with_pre_norm(self, pre_norm: bool) -> Self

Builder: Set whether to use pre-norm

pub fn with_dropout(self, dropout: f64) -> Self

Builder: Set dropout

pub fn seq_length(&self) -> usize

Get sequence length (num_patches + optional class token)

pub fn validate(&self) -> Result<()>

Validate configuration

Trait Implementations

impl Clone for VisionTransformerConfig

fn clone(&self) -> VisionTransformerConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for VisionTransformerConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.