Method normalize

src/dect_verbalizer.py:199–211

Given logits regarding the entire vocabulary, return the probs over the label words set. Args: logits (:obj:`Tensor`): The logits over the entire vocabulary. Returns: :obj:`Tensor`: The logits over the label words set.

(self, logits: torch.Tensor)

        return label_logits
    
    def normalize(self, logits: torch.Tensor) -> torch.Tensor:
        """
        Given logits regarding the entire vocabulary, return the probs over the label words set.
        
        Args:
            logits (:obj:`Tensor`): The logits over the entire vocabulary.

        Returns:
            :obj:`Tensor`: The logits over the label words set.
        
        """
        batch_size = logits.shape[0]
        return F.softmax(logits.reshape(batch_size, -1), dim=-1).reshape(*logits.shape)


    def aggregate(self, label_words_logits: torch.Tensor) -> torch.Tensor: