cs.LGNov 19, 2026

Decision-Focused Reinforcement Learning for LLM Post-Training

Authors: Haotian Zhai, Leqi Liu

Organizations: The University of Texas at Austin

Abstract

Large language models (LLMs) have shown promise as predictors across various tasks, benefiting from their ability to use information from unstructured context. However, when predictions serve as parameters of downstream optimization problems, higher prediction accuracy does not necessarily translate into lower decision costs. We propose decision-focused reinforcement learning (DFRL), a framework for post-training LLMs to predict optimization parameters using negative downstream decision regret as reward. DFRL trains LLM predictors to minimize the costs of the decisions induced by their predictions, without requiring downstream convexity or differentiability, and integrates readily into existing RL post-training frameworks. To address expensive reward evaluation, we develop two approaches. Heuristic warm-starting uses inexpensive approximate rewards during an initial phase before switching to refinement with exact rewards. Reward correction combines inexpensive estimates with randomized exact evaluations, preserving the expected policy gradient while reducing the expected number of exact solves per update. Experiments on three synthetic optimization problems demonstrate improvements over prediction-loss baselines, while a larger knapsack experiment shows faster early improvement under short training budgets. On inventory planning with H&M transactions and product descriptions, our best configuration, in which the LLM reasons before predicting, reduces regret by 31.4% relative to the SFT baseline and 12.4% relative to the strongest evaluated conventional forecasting baseline.

Figures & tables

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Sep 29, 2026

Unlocking the Critic: Reward-Free Policy Optimization for LLM Post-Training

Recent approaches to reinforcement learning (RL) post-training for large language models increasingly remove the critic to reduce training instability and memory overhead. Even where a critic is trained, it is discarded once training ends, although it has learned to predict outcomes. We revisit this trend and show that a pretrained critic's ability to predict future outcomes can make it a valuable asset for efficient long-horizon reasoning. First, we find that instability in critic-based RL for long chain-of-thought reasoning is largely an optimization artifact: keeping policy updates small and low in variance restores stable convergence. Second, a well-pretrained critic estimates the posterior probability of eventual success from later trajectory states and unfinished prefixes. Its predictions provide outcome-derived, dense, per-prefix learning signals that, during policy optimization, require neither completed rollouts, step-level annotations, nor external reward labels. Building on this insight, we introduce Reward-Free Policy Optimization (RFPO), which repurposes a single calibrated, frozen critic as a rollout-level reward, a value baseline for generalized advantage estimation, and a success forecaster for unfinished prefixes. We further show that binarizing the debiased score stops the policy from exploiting the critic's length bias. Binarized, RFPO matches supervised PPO without a single label in the training loop, while cutting compute and memory overhead. This makes RFPO well suited to long-horizon reasoning tasks, where outcomes arrive late and generation dominates cost: because rollouts can be rewarded before they finish, training no longer has to pay for waiting on every trajectory to complete. Our findings challenge the prevailing critic-free paradigm and establish critic-based, reward-free optimization as a scalable and computationally efficient path for LLM post-training.
Jun 2, 2026

RL Excursions during Pre-Training: Re-examining Policy Optimization for LLM training

The standard LLM training pipeline applies reinforcement learning (RL) only after pre-training and supervised fine-tuning (SFT). We question this status quo by training a LLM from scratch and applying RL, SFT, and SFT followed by RL directly to intermediate pre-training checkpoints. We find that RL is effective very early, and often matches the full SFT→\toRL pipeline early as well. Through experiments on harder problems, we find that targeted pre-training data composition is a strong lever for RL effectiveness, even more so than model scale. Beyond reasoning accuracy, applying RL directly to base checkpoints expands the model's distribution; the sharpening effect reported in recent work arises only when RL follows SFT. The general capabilities of the model remain essentially unchanged by RL, while they degrade following SFT. Finally, we merge RL and SFT objectives by parallel averaging, which outperforms across all other training methods discussed, across metrics, while preserving general capabilities. Together, these results suggest that LLM training might benefit from an expanded use of RL.
Jun 2, 2026

Efficient Hyperparameter Optimization for LLM Reinforcement Learning

Reinforcement learning (RL) for large language models (LLMs) is highly sensitive to hyperparameter configurations, making hyperparameter optimization (HPO) essential yet computationally expensive. Existing multi-fidelity HPO methods remain inefficient for LLM RL due to the massive model scale and resource-intensive training cycles. In this paper, we propose Joint Fidelity Hyperparameter Optimization (JF-HPO), which simultaneously adapts both model size and training budget as fidelity. JF-HPO is empowered by: (i) it leverages a small proxy model of the target LLM for efficient training and evaluation in each HPO trial; (ii) it integrates carefully designed early-stopping strategies based on training dynamics; (iii) it introduces an efficient checkpointing mechanism to eliminate redundant computations. Compared with existing HPO methods, JF-HPO significantly improves the computational efficiency of each trial (up to 14.9 times), while achieving better or competitive predictive accuracy under the same time budget. Notably, compared with utilizing hyperparameter configurations from the VeRL Recipe, JF-HPO delivers performance improvements ranging from 5.8% to 111.6%.