Input links cannot all be predicted

### Description After training the model, I planned to have the model predict 1955 links, and it turned out that the model only predicted 1840 links and gave up on predicting 115 links. And return <115 triples containing invalid keys skipped! You can use `ScoringBasedEmbeddingModel.get_invalid_keys` or `DataIndexer.get_invalid_keys` to find out which keys are invalid.> I tested the dataset and there are no invalid inputs. ### Actual Behavior Only some of the links were predicted ### Expected Behavior Forecast All Links ### Steps to Reproduce