Struct VisionItem 

pub struct VisionItem {
    pub groups: Vec<Vec<f32>>,
    pub grid_h: usize,
    pub grid_w: usize,
}

2026-09-25: One vision item: a still image, or a video, ready for the encoder.

groups holds the temporal groups: one for a still image (image), and grid_t for a video, each group a full grid_h × grid_w patch plane built from temporal_patch_size consecutive frames. All groups share the item’s one grid, so the encoder takes each group as an image of that grid.

The group count lives in the item rather than in a parallel vector, so the pad run (pad_count) and the encoder grids both read it from t_len.

Fields

groups: Vec<Vec<f32>>

2026-09-25: Per temporal group: [grid_h * grid_w, C * temporal_patch_size * patch^2].

grid_h: usize

2026-09-25: Pre-merge patch grid, shared by all of this item’s groups.

grid_w: usize

Implementations

impl VisionItem

pub fn image(pixels: Vec<f32>, grid_h: usize, grid_w: usize) -> Self

2026-09-25: A still image: one temporal group.

pub fn t_len(&self) -> usize

2026-09-25: Temporal extent, in groups; at least 1.

pub fn pad_count(&self, spatial_merge_size: usize) -> usize

2026-09-25: Merged tokens this item occupies in the prompt: t_len times the spatially merged grid. The chat request builder uses it as the length of the item’s pad run.

Trait Implementations

impl Clone for VisionItem

fn clone(&self) -> VisionItem

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for VisionItem

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

Formats the value using the given formatter.

impl PartialEq for VisionItem

fn eq(&self, other: &VisionItem) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for VisionItem