I am using the Flair library in python for a Fake News Detection problem, using an RNN-based algorithm. The code does not give errors, but the training phase is excessively slow and also at the end of that phase the metrics used for evaluation (accuracy, recall, f1-score) take value 0.
The function used for the training phase is ModelTrainer.train(//various parameters). The dataset has undergone pre-processing, so I don't think there are any errors in the data, having also removed the empty sentences with Flair's filter_empty_sentence() function.