kaira.losses.text.Word2VecLoss

Inheritance diagram for Word2VecLoss

class kaira.losses.text.Word2VecLoss(embedding_dim, vocab_size, n_negatives=5)

Bases: BaseLoss

Word2Vec Loss Module.

This module implements the negative sampling loss used in Word2Vec.

Methods

__init__	Initialize the Word2VecLoss module.
forward	Forward pass through the Word2VecLoss module.

__init__(embedding_dim, vocab_size, n_negatives=5)

Initialize the Word2VecLoss module.

Parameters:

• embedding_dim (int) – Dimensionality of embeddings.

• vocab_size (int) – Size of vocabulary.

• n_negatives (int) – Number of negative samples. Default is 5.

forward(input_idx: Tensor, output_idx: Tensor) → Tensor

Forward pass through the Word2VecLoss module.

Parameters:

• input_idx (torch.Tensor) – Input word indices.

• output_idx (torch.Tensor) – Output context word indices.

Returns:

The Word2Vec loss.

Return type:

torch.Tensor