quaterion.loss.softmax_loss module

class SoftmaxLoss(embedding_size: int, num_groups: int, temperature: float = 0.05)[source]

Bases: GroupLoss

Regular cross-entropy loss.

An implementation of softmax with dot product. It is designed to work with the base GroupLoss.

Parameters:

• embedding_size – Output dimension of the encoder.

• num_groups – Number of groups in the dataset.

• temperature – Temperature value to divide logits, defaults to 0.05

forward(embeddings: Tensor, groups: LongTensor) → Tensor[source]

Compute loss value.

Parameters:

• embeddings – shape: (batch_size, vector_length) - Output embeddings from the encoder

• groups – shape: (batch_size,) - Group ids, associated with embeddings

Returns:

Tensor – zero-size tensor, loss value

training: bool