relari-ai/continuous-eval

Add Softmax Output for DebertaAnswerScores batch_calculate method

Closed this issue · 1 comments

In DebertaAnswerScores, we need to ensure consistency in the output of the calculate and batch_calculate methods. In current implementation, the calculate method returns normalized probabilities, while the batch_calculate return logits values.
image

yisz commented

Thanks for catching this @satriowputra !