Doubt about the TARGET parameter in the dataset

First I would like to thank you every single person that worked to make this dataset available.

I am studying the repository as well as the papers about FLAN. And I couldn't understand the `target` parameter.

Do you guys use such parameter to fine tune the language model in supervised way? Similar to what would be a reinforcement learning with human feedback?

For example, when you ask the question to the language models it uses the `target` value to understand if its response is right or wrong based on the `target` value? Thus, it can learn its mistakes from the `target` value?

In general my question is - when does the parameter `target` comes into place when training the model using the FLAN dataset?

Read the repository documentation and the FLAN papers referenced in the issue to determine how the dataset's target parameter is used during training. Done means documenting a clear explanation of when target is consumed and how it relates, if at all, to supervised fine-tuning or reinforcement learning.

Written by the indexing model from the issue text.