Enum ModelSize 

pub enum ModelSize {
    Tiny,
    Small,
    Base,
}

Named ViT model-size variants, matching the standard ViT paper dimensions.

Memory figures assume fp32, N = 2448 patches, chunk = 64, and cover attention score/weight tensors for one transformer layer (the practical backward-pass peak). Total GPU memory is 3–5× higher once weights, activations, and Adam optimizer states are included.

Size	d_model	heads	~params	per-layer bwd B=16	per-layer bwd B=4
Tiny	192	3	~11 M	2.1 GB	0.5 GB
Small	384	6	~44 M	4.4 GB	1.1 GB
Base	768	12	~205 M	17.5 GB ✗	2.2 GB

Recommended --batch-size per preset (WGPU / Metal, 16 GB device):

• tiny: up to 16 — comfortable; per-layer bwd ≈ 2.1 GB
• small: up to 8 — comfortable; per-layer bwd ≈ 2.2 GB
• base: up to 4 — per-layer bwd ≈ 2.2 GB; total ≈ 10 GB

Variants

Tiny

ViT-Ti: d=192, depth=12, heads=3, mlp=768. Fits in ~2 GB VRAM.

Small

ViT-S: d=384, depth=12, heads=6, mlp=1536. Fits in ~6 GB VRAM.

Base

ViT-B: d=768, depth=12, heads=12, mlp=3072. Requires ≥ 16 GB VRAM.

Implementations

impl ModelSize

pub fn d_model(self) -> usize

Return the transformer hidden dimension for this size.

pub fn depth(self) -> usize

Return the number of transformer layers.

pub fn num_heads(self) -> usize

Return the number of attention heads.

pub fn mlp_dim(self) -> usize

Return the MLP hidden dimension (4 × d_model).

pub fn sensor_encoder_config(self) -> SensorEncoderConfig

Build a SensorEncoderConfig for this size with sensible defaults.

pub fn text_encoder_config(self) -> TextEncoderConfig

Build a TextEncoderConfig for this size.

pub fn sensorlm_config(self) -> SensorLMConfig

Build a complete SensorLMConfig for this size.

pub fn approx_params(self) -> &'static str

Human-readable approximate parameter count for both towers combined.

Trait Implementations

impl Clone for ModelSize

fn clone(&self) -> ModelSize

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ModelSize

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

Formats the value using the given formatter.

impl Default for ModelSize

fn default() -> ModelSize

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for ModelSize

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl PartialEq for ModelSize

fn eq(&self, other: &ModelSize) -> 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 Serialize for ModelSize

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl Copy for ModelSize

impl Eq for ModelSize

impl StructuralPartialEq for ModelSize