Last-token pooling — returns the last non-padding token embedding. Uses attentionMask to find the last real token in each sequence.
Last-token pooling — returns the last non-padding token embedding. Uses attentionMask to find the last real token in each sequence.