Inappropriate Contents in the dataset
#5
by xingpeng001 - opened
Does anyone notice that there are many sexual contents in the training data? Look at id 51e52e10-0ef1-4e9e-8dfe-88036876fd50, 5ad00753-fd17-4f52-8497-d24ff949dbd5, ... Maybe using a filter would be better, if it is not your intention to train a model with some specific knowledge
Hi @xingpeng001 , we do train on these datasets as they are pulled from the web. They do go through quality filtering, topic classification, etc. so these examples are representatively tagged as adult_content in our datasets.