Datasets:

bigcode

/

the-stack

like 662

I would like to ask for the clarification on the following term of use.

you agree to update your own version of The Stack to the most recent usable version specified by the maintainers

Suppose that the new version of the stack is released;
I was wondering if the model trained on old (unusable) dataset is still usable, i.e., if one needs to retrain the model from scratch as soon as the new version of dataset is released.
For example, what happens to the current public LLMs? Do they need to be retrained?