Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Reward Weighted Classifier-Free Guidance as Policy Improvement in Autoregressive Models

About

Consider an auto-regressive model that produces outputs x (e.g., answers to questions, molecules) each of which can be summarized by an attribute vector y (e.g., helpfulness vs. harmlessness, or bio-availability vs. lipophilicity). An arbitrary reward function r(y) encodes tradeoffs between these properties. Typically, tilting the model's sampling distribution to increase this reward is done at training time via reinforcement learning. However, if the reward function changes, re-alignment requires re-training. In this paper, we show that a reward weighted classifier-free guidance (RCFG) can act as a policy improvement operator in this setting, approximating tilting the sampling distribution by the Q function. We apply RCFG to molecular generation, demonstrating that it can optimize novel reward functions at test time. Finally, we show that using RCFG as a teacher and distilling into the base policy to serve as a warm start significantly speeds up convergence for standard RL.

Alexander Peysakhovich, William Berman• 2026

Related benchmarks

TaskDatasetResultRank
Molecular Property Optimizationlipophilic efficiency
Normalized Reward0.6
10
Molecular Property Optimizationdrug like
Normalized Reward0.45
10
Molecular Property Optimizationfragment like
Normalized Reward34
10
Molecular Property Optimizationgpcr like
Normalized Reward54
10
Molecular Property Optimizationkinase like
Normalized Reward57
10
Molecular Property Optimizationlow mw high potency
Normalized Reward0.45
10
Molecular Property OptimizationQED maximize
Normalized Reward62
10
Molecular Property Optimizationoral bioavailable
Normalized Reward0.36
10
Molecular Property Optimizationsoluble permeable
Normalized Reward0.39
10
Molecular Property Optimizationeasy synth druglike
Normalized Reward0.62
10
Showing 10 of 24 rows

Other info

Follow for update