Open IcanDoItL opened 3 months ago
the example examples/training/quora_duplicate_questions/training_multi-task-learning.py Contrastive Loss, MultipleNegativesRankingLoss is not of the same magnitude. How to ensure its effectiveness? Is there any reference material available
the example examples/training/quora_duplicate_questions/training_multi-task-learning.py Comparing losses also uses BatchSamplers NO_DUPLICATES, Will it affect the effectiveness?
the example examples/training/quora_duplicate_questions/training_multi-task-learning.py Contrastive Loss, MultipleNegativesRankingLoss is not of the same magnitude. How to ensure its effectiveness? Is there any reference material available