Struct EmbeddingHandle 

pub struct EmbeddingHandle {
    pub name: String,
    pub module: Option<PyObject>,
    pub trainable: bool,
    pub dim: usize,
    pub vocab_size: usize,
}

Handle to a registered embedding module.

Wraps either a trainable nn.Embedding or a frozen torch.Tensor. Created via CompiledProgram.register_embedding() in Python.

Fields

name: String

Unique name matching the nn() declaration

module: Option<PyObject>

The PyTorch nn.Embedding or tensor

trainable: bool

Whether gradients flow through this embedding

dim: usize

Embedding vector dimension (second axis of weight matrix)

vocab_size: usize

Number of embedding entries (first axis of weight matrix)

Implementations

impl EmbeddingHandle

pub fn new(name: String, trainable: bool, dim: usize, vocab_size: usize) -> Self

Create a new embedding handle.

pub fn has_module(&self) -> bool

Check if the PyTorch module/tensor has been set.

pub fn set_module(&mut self, module: PyObject)

Set the PyTorch module/tensor.

pub fn module(&self) -> Option<&PyObject>

Get a reference to the PyTorch module/tensor.

Trait Implementations

impl Debug for EmbeddingHandle

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

Formats the value using the given formatter.