Struct VisionConfig 

pub struct VisionConfig {

    pub depth: usize,
    pub hidden_size: usize,
    pub num_heads: usize,
    pub patch_size: usize,
    pub temporal_patch_size: usize,
    pub spatial_merge_size: usize,
    pub intermediate_size: usize,
    pub out_hidden_size: usize,
    pub deepstack_visual_indexes: Vec<usize>,
    pub image_pad_token_id: u32,
    pub video_pad_token_id: u32,
    pub max_pixels: Option<usize>,
    pub hidden_act: String,
    pub rms_norm_eps: f32,
    pub attention_bias: bool,
    pub projection_intermediate_size: usize,
    pub swiglu_limit: f32,
    pub rope_theta: f32,
    pub image_mean: [f32; 3],
    pub image_std: [f32; 3],
    pub min_image_tokens: usize,
    pub max_image_tokens: usize,
    pub block_major_patches: bool,
}

2026-09-26: Vision tower configuration: parse_vision_config for the Qwen family, and the GLM-5.3 parser, which also fills the fields the Qwen parser leaves at their defaults.

Fields

depth: usize

hidden_size: usize

num_heads: usize

patch_size: usize

temporal_patch_size: usize

spatial_merge_size: usize

intermediate_size: usize

out_hidden_size: usize

deepstack_visual_indexes: Vec<usize>

image_pad_token_id: u32

2026-09-26: The token that marks where vision embeddings are spliced into the text. 0 means undeclared: the engine then uses vision_encoder::IMAGE_PAD_TOKEN_ID, and the chat tokenizer the id of <|image_pad|>.

video_pad_token_id: u32

2026-09-26: The video-frame counterpart of Self::image_pad_token_id, with the same 0 rule (VIDEO_PAD_TOKEN_ID, <|video_pad|>).

max_pixels: Option<usize>

2026-09-26: The vision area bound in pixels. The parsers leave it None; serve_load sets it from resolve_vision_max_pixels before the model is built, and the vision loaders pass it to the encoder.

hidden_act: String

rms_norm_eps: f32

attention_bias: bool

projection_intermediate_size: usize

swiglu_limit: f32

rope_theta: f32

image_mean: [f32; 3]

image_std: [f32; 3]

min_image_tokens: usize

max_image_tokens: usize

block_major_patches: bool

2026-09-26: Patch tokens in 2x2-block-major order rather than raster; the preprocessor reads it (vision_preprocess). The GLM-5.3 parser sets it.

Implementations

impl VisionConfig

pub fn merger_input_size(&self) -> usize

Trait Implementations

impl Clone for VisionConfig

fn clone(&self) -> VisionConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for VisionConfig

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

Formats the value using the given formatter.

impl Default for VisionConfig

fn default() -> Self

Returns the “default value” for a type.