Struct ClipEncoder 

pub struct ClipEncoder<B: Backend> { /* private fields */ }

CLIP vision encoder for DeepSeek-OCR.

Implements the CLIP-Large architecture (or configurable variants) for global semantic understanding of visual features. Processes patch embeddings through a stack of transformer blocks with pre-LayerNorm and Quick GELU activation.

Architecture Details

• Input: RGB images [B, 3, H, W]
• Patch embedding: Splits image into non-overlapping patches
• Class token: Learnable token prepended to sequence
• Position encoding: Added to all tokens
• Transformer: Stack of attention + FFN blocks
• Output: Sequence of embeddings [B, num_patches+1, hidden_size]

Default Configuration (CLIP-Large)

• 1024-dimensional embeddings
• 24 transformer layers
• 16 attention heads per layer
• 4096-dimensional feed-forward hidden layer
• 14×14 patch size on 224×224 images (256 patches)

Example

let config = ClipConfig::large();
let encoder = ClipEncoder::new(&config, &device);

let image = Tensor::zeros(, &device);[1]
let features = encoder.forward(image, None);
// features shape:  (256 patches + 1 class token)[1]

Implementations

impl<B: Backend> ClipEncoder<B>

pub fn new(cfg: &ClipConfig, device: &B::Device) -> Self

Creates a new CLIP encoder.

Initializes all layers including embeddings, pre-LayerNorm, and transformer blocks according to the provided configuration.

Arguments

• cfg - CLIP configuration specifying model architecture
• device - Device for tensor allocation (CPU/GPU)

Returns

Fully initialized CLIP encoder ready for inference

pub fn forward( &self, x: Tensor<B, 4>, _patch_embeds: Option<Tensor<B, 4>>, ) -> Tensor<B, 3>

Forward pass through CLIP encoder.

Processes input images through patch embedding, position encoding, and transformer layers to produce semantic visual features.

Arguments

• x - Input image tensor of shape [B, 3, H, W]
• _patch_embeds - Optional pre-computed patch embeddings (currently unused, reserved for future integration with SAM features)

Returns

Feature tensor of shape [B, num_patches+1, hidden_size] where the first token (index 0) is the class token and remaining tokens are patch features.

Note

The _patch_embeds parameter is currently ignored. Full integration with SAM features would require a projection layer to match embedding dimensions.

pub fn forward_features( &self, x: Tensor<B, 4>, patch_embeds: Option<Tensor<B, 4>>, ) -> Tensor<B, 3>

Forward pass excluding class token.

Convenience method that runs the encoder and strips the class token from the output sequence, returning only patch feature embeddings.

Arguments

• x - Input image tensor of shape [B, 3, H, W]
• patch_embeds - Optional pre-computed patch embeddings

Returns

Patch features of shape [B, num_patches, hidden_size] with the class token removed (excludes first token in sequence).

Trait Implementations

impl<B> AutodiffModule<B> for ClipEncoder<B>

where B: AutodiffBackend + Backend, <B as AutodiffBackend>::InnerBackend: Backend,

type InnerModule = ClipEncoder<<B as AutodiffBackend>::InnerBackend>

Inner module without auto-differentiation.

fn valid(&self) -> Self::InnerModule

Get the same module, but on the inner backend without auto-differentiation.

impl<B: Backend> Clone for ClipEncoder<B>

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<B: Debug + Backend> Debug for ClipEncoder<B>

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

Formats the value using the given formatter.

impl<B: Backend> Display for ClipEncoder<B>

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

Formats the value using the given formatter.

impl<B: Backend> Module<B> for ClipEncoder<B>

type Record = ClipEncoderRecord<B>

Type to save and load the module.

fn load_record(self, record: Self::Record) -> Self

Load the module state from a record.

fn into_record(self) -> Self::Record

Convert the module into a record containing the state.

fn num_params(&self) -> usize

Get the number of parameters the module has, including all of its sub-modules.

fn visit<Visitor: ModuleVisitor<B>>(&self, visitor: &mut Visitor)

Visit each tensor parameter in the module with a visitor.

fn map<Mapper: ModuleMapper<B>>(self, mapper: &mut Mapper) -> Self

Map each tensor parameter in the module with a mapper.

fn collect_devices(&self, devices: Devices<B>) -> Devices<B>

Return all the devices found in the underneath module tree added to the given vector without duplicates.

fn to_device(self, device: &B::Device) -> Self

Move the module and all of its sub-modules to the given device.

fn fork(self, device: &B::Device) -> Self

Fork the module and all of its sub-modules to the given device.

fn devices(&self) -> Vec<<B as Backend>::Device>

Return all the devices found in the underneath module tree without duplicates.

fn no_grad(self) -> Self

Each tensor in the module tree will not require grad.

fn save_file<FR, PB>( self, file_path: PB, recorder: &FR, ) -> Result<(), RecorderError>

where FR: FileRecorder<B>, PB: Into<PathBuf>,

Save the module to a file using the provided file recorder.

fn load_file<FR, PB>( self, file_path: PB, recorder: &FR, device: &<B as Backend>::Device, ) -> Result<Self, RecorderError>

where FR: FileRecorder<B>, PB: Into<PathBuf>,

Load the module from a file using the provided file recorder.

fn quantize_weights(self, quantizer: &mut Quantizer) -> Self

Quantize the weights of the module.

impl<B: Backend> ModuleDisplay for ClipEncoder<B>

fn format(&self, passed_settings: DisplaySettings) -> String

Formats the module with provided display settings.

fn custom_settings(&self) -> Option<DisplaySettings>

Custom display settings for the module.

fn custom_content(&self, _content: Content) -> Option<Content>

Custom attributes for the module.

impl<B: Backend> ModuleDisplayDefault for ClipEncoder<B>

fn content(&self, content: Content) -> Option<Content>

Attributes of the module used for display purposes.

fn num_params(&self) -> usize

Gets the number of the parameters of the module.