Is it possible to have a seperate negatively weighted dataset in training

Essentially while training having a dataset subset that has a negative loss objective while still also having the main dataset the close thing to this right now is the negative tokens/captions where I suppose you could add only negative tag to the images that you want to use as your negative and then apply a negative caption weight i dont know if this works the same however

No files, tests, or entry points are named. Start by locating how training datasets, negative captions or tokens, and loss objectives are handled, then determine whether a separate negatively weighted subset can coexist with the main dataset. Done should mean the requested behavior is clearly specified and its effect during training is verified.

Written by the indexing model from the issue text.