training all-mpnet-base-v2 with MultipleNegativesRankingLoss for sentence embedding.
according to the docs here - it’s fine for a pair of sentences,
but when setting the evaluator it requires ‘scores’ column (which does not exist in current ds), doc here.
Where I can set a default score to all pairs in ds to 1, I wonder if that’s the best way to handle it, or if another evaluator should be chosen (if so, then why? all others seems unsuitable to me)
please notice I’m not using model.fit but a trainer, so same question goes to the dev_evaluator that goes into the sentence_transformer_trainer object.
Thank you