Alignment of large language models with constrained learning
This work addresses the challenge of ensuring safe and aligned behavior in large language models for AI safety applications, representing an incremental improvement over existing constrained alignment methods.
The paper tackles the problem of aligning large language models with constraints by maximizing a primary reward while satisfying secondary utility constraints, and demonstrates that their dual-based alignment method can find an optimal constrained policy up to a parametrization gap, as validated on the PKU-SafeRLHF dataset.
We study the problem of computing an optimal large language model (LLM) policy for a constrained alignment problem, where the goal is to maximize a primary reward objective while satisfying constraints on secondary utilities. Despite the popularity of Lagrangian-based LLM policy search in constrained alignment, iterative primal-dual methods often fail to converge, and non-iterative dual-based methods do not achieve optimality in the LLM parameter space. To address these challenges, we employ Lagrangian duality to develop an iterative dual-based alignment method that alternates between updating the LLM policy via Lagrangian maximization and updating the dual variable via dual descent. In theory, we characterize the primal-dual gap between the primal value in the distribution space and the dual value in the LLM parameter space. We further quantify the optimality gap of the learned LLM policies at near-optimal dual variables with respect to both the objective and the constraint functions. These results prove that dual-based alignment methods can find an optimal constrained LLM policy, up to an LLM parametrization gap. We demonstrate the effectiveness and merits of our approach through extensive experiments conducted on the PKU-SafeRLHF dataset.