While previous research has advocated for constraining policy optimization, our study introduces a novel approach to enhance the reward model's

MolGroup achieves this by utilizing a routing mechanismoptimized through abi-level optimization framework.

In the reminder of the paper we will try whenever possible to formally introduce functions by clearly stating domain and co-domain.

We propose a novel loss function called Energy Discrepancy (ED) which does not rely on the computation of scores or expensive Markov chain Monte Carlo.