Trainer predict or evulate does not return softmax or sigmoid value

``````class BertPrefixForSequenceClassification(BertPreTrainedModel):
# method
return SequenceClassifierOutput(
loss=loss,
logits=logits,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
)
``````

While evaluating/ predicting with Trainer, the return value is not processed by sigmoid or softmax function. Is it necessary to add the sigmoid or softmax function in evaluating and predicting?

If you need it, yes.