Distributionally Robust Token Optimization in RLHF

Large Language Models (LLMs) tend to respond correctly to prompts that align well with the data they were trained and fine-tuned on. Yet, small shifts in wording, format, or language can trigger surprisingly large failures, especially on multi-step reasoning problems. To address this problem, we propose a $\text{Distributionally Robust Token Optimization (DRTO)}$ approach, which combines token-level Reinforcement Learning from Human Feedback (RLHF) with $\text{Distributionally Robust Optimization (DRO)}$. DRTO constructs f-divergence ambiguity sets over span-level actor losses, providing a principled way to emphasize difficult response segments during policy optimization. Empirically, DRTO enhances consistency under distribution shifts in multiple reasoning benchmarks among different tasks.