jasonkena/

Eurus-2-RL-Data-privileged

A privileged variant of PRIME-RL/Eurus-2-RL-Data, built for privileged-information RL experiments (giving the value network / process reward model access to a worked reference solution z that the policy never sees).

The train split adds two columns; every other column is preserved verbatim from the base Eurus-2-RL-Data:

column description

reference_solution the full worked reference… See the full description on the dataset page: https://huggingface.co/datasets/jasonkena/Eurus-2-RL-Data-privileged.