Hello thank you very much for your work, it is very inspiring to me. But I have some questions about the code and the corresponding part of the paper. In your paper, "By requiring the model to first predict the appropriate token set, we force it to align the corresponding tokens with each task." However, in the code i can't see the part that indicates training the task token.
Hello thank you very much for your work, it is very inspiring to me. But I have some questions about the code and the corresponding part of the paper. In your paper, "By requiring the model to first predict the appropriate token set, we force it to align the corresponding tokens with each task." However, in the code i can't see the part that indicates training the task token.