Vector Policy Optimization: Training for Diversity Improves Test-Time Search

• URL: http://arxiv.org/abs/2605.22817v1
• Date: Thu, 21 May 2026 17:59:26 GMT
• Title: Vector Policy Optimization: Training for Diversity Improves Test-Time Search
• Abstract summary: We propose Vector Policy Optimization (VPO), an RL algorithm that explicitly trains policies to anticipate diverse downstream reward functions.<n>VPO exploits that rewards are often vector-valued in practice, like per-test-case correctness in code generation or, say, multiple different user personas or reward models.
• Score: 25.300820670341427
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Language models must now generalize out of the box to novel environments and work inside inference-scaling search procedures, such as AlphaEvolve, that select rollouts with a variety of task-specific reward functions. Unfortunately, the standard paradigm of LLM post-training optimizes a pre-specified scalar reward, often leading current LLMs to produce low-entropy response distributions and thus to struggle at displaying the diversity that inference-time search will require. We propose Vector Policy Optimization (VPO), an RL algorithm that explicitly trains policies to anticipate diverse downstream reward functions and to produce diverse solutions. VPO exploits that rewards are often vector-valued in practice, like per-test-case correctness in code generation or, say, multiple different user personas or reward models. VPO is essentially a drop-in replacement for the GRPO advantage estimator, but it trains the LLM to output a set of solutions where individual solutions specialize to different trade-offs in the vector reward space. Across four tasks, VPO matches or beats the strongest scalar RL baselines on test-time search (e.g. pass@k and best@k), with the gap widening as the search budget grows. For evolutionary search, VPO models unlock problems that GRPO models cannot solve at all. As test-time search becomes more standardized, optimizing for diversity may need to become the default post-training objective.

Related papers

• ASAP: Agent-System Co-Design for Wall-Clock-Centered Auto HPO Research for ML Experiments [45.577182866886126]
  We present ASAP, an agent-system co-design for HPO.<n>We show how ASAP integrates a diverse pool of inductive-biased generalizations and to select among their proposals each round.<n>We also show how ASAP re-architects the loop to reduce end-to-end wall-clock while preserving regret quality.
  arXiv  Detail & Related papers  (2026-06-23T22:00:26Z)
• Breaking $\ extit{Winner-Takes-All}$: Cooperative Policy Optimization Improves Diverse LLM Reasoning [53.42577591449649]
  Group Cooperative Policy Optimization shifts the training paradigm from rollout competition to team cooperation.<n>GCPO replaces independent rollout scoring with team-level credit assignment.<n>It redistributes the collective team reward to each single rollout according to its average marginal contribution to the team.
  arXiv  Detail & Related papers  (2026-05-12T03:20:24Z)
• Poly-EPO: Training Exploratory Reasoning Models [62.82992914206963]
  We present a framework for post-training language models (LMs) that explicitly encourages optimistic exploration and promotes a synergy between exploration and exploitation.<n>We propose Polychromic Exploratory Policy Optimization (Poly-EPO), which instantiates this framework with an objective that explicitly synergizes exploration and exploitation.
  arXiv  Detail & Related papers  (2026-04-19T22:54:19Z)
• ShapE-GRPO: Shapley-Enhanced Reward Allocation for Multi-Candidate LLM Training [17.817168693366916]
  Existing reinforcement learning post-training paradigms typically assign the same set-level scalar reward to every candidate in the set.<n>This leads to noisy training signals where poor candidates free-ride on the high reward produced by a single strong peer, resulting in suboptimal exploration.<n>By leveraging the permutation-invariant nature of set-level utility, we derive a Shapley-enhanced formulation from cooperative game theory to decompose set-level rewards into granular, candidate-specific signals.
  arXiv  Detail & Related papers  (2026-03-31T15:24:15Z)
• GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization [133.27496265096445]
  We show how to apply Group Relative Policy Optimization under multi-reward setting without examining its suitability.<n>We then introduce Group reward-Decoupled Normalization Policy Optimization (GDPO), a new policy optimization method to resolve these issues.<n>GDPO consistently outperforms GRPO, demonstrating its effectiveness and generalizability for multi-reward reinforcement learning optimization.
  arXiv  Detail & Related papers  (2026-01-08T18:59:24Z)
• Group-Aware Reinforcement Learning for Output Diversity in Large Language Models [8.356950556877612]
  Large Language Models (LLMs) often suffer from mode collapse, repeatedly generating the same few completions even when many valid answers exist.<n>We introduce Group-Aware Policy Optimization (GAPO), a simple extension of the popular Group Relative Policy Optimization (GRPO) that computes rewards over the group as a whole.<n>We demonstrate GAPO using a frequency-aware reward function that encourages uniform sampling over valid LLM completions, and show that GAPO-trained models produce valid and more diverse model responses.
  arXiv  Detail & Related papers  (2025-11-16T13:42:55Z)
• FlowRL: Matching Reward Distributions for LLM Reasoning [69.88820066093798]
  We propose FlowRL: matching the full reward distribution via flow balancing instead of maximizing rewards in large language model (LLM) reinforcement learning (RL)<n>We transform scalar rewards into a normalized target distribution using a learnable partition function, and then minimize the reverse KL divergence between the policy and the target distribution.
  arXiv  Detail & Related papers  (2025-09-18T17:56:36Z)
• Reinforcing Multi-Turn Reasoning in LLM Agents via Turn-Level Reward Design [35.544075583073685]
  We present the first systematic study of textitturn-level reward design for multi-turn RL algorithms and agent applications.<n>We conduct case studies on multi-turn reasoning-augmented search agents, where we carefully design two types of turn-level rewards: verifiable and LLM-as-judge.<n>Our experiments on multi-turn search tasks demonstrate that incorporating well-designed turn-level rewards enables RL algorithms to significantly outperform baseline methods with trajectory-level rewards.
  arXiv  Detail & Related papers  (2025-05-17T04:09:46Z)
• Not All Rollouts are Useful: Down-Sampling Rollouts in LLM Reinforcement Learning [55.15106182268834]
  Reinforcement learning with verifiable rewards (RLVR) has emerged as the leading approach for enhancing reasoning capabilities in large language models.<n>It faces a fundamental compute and memory asymmetry: rollout generation is embarrassingly parallel and memory-light, whereas policy updates are communication-heavy and memory-intensive.<n>We introduce PODS (Policy Optimization with Down-Sampling), which decouples rollout generation from policy updates by training only on a strategically selected subset of rollouts.
  arXiv  Detail & Related papers  (2025-04-18T17:49:55Z)
• VinePPO: Refining Credit Assignment in RL Training of LLMs [66.80143024475635]
  We propose VinePPO, a straightforward approach that leverages the flexibility of language environments to compute unbiased Monte Carlo-based estimates.<n>Our method consistently outperforms PPO and other baselines across MATH and GSM8K datasets in less wall-clock time.
  arXiv  Detail & Related papers  (2024-10-02T15:49:30Z)
• From $r$ to $Q^*$: Your Language Model is Secretly a Q-Function [50.812404038684505]
  We show that we can derive DPO in the token-level MDP as a general inverse Q-learning algorithm, which satisfies the Bellman equation. We discuss applications of our work, including information elicitation in multi-turn dialogue, reasoning, agentic applications and end-to-end training of multi-model systems.
  arXiv  Detail & Related papers  (2024-04-18T17:37:02Z)
• Relative Preference Optimization: Enhancing LLM Alignment through Contrasting Responses across Identical and Diverse Prompts [95.09994361995389]
  Relative Preference Optimization (RPO) is designed to discern between more and less preferred responses derived from both identical and related prompts. RPO has demonstrated a superior ability to align large language models with user preferences and to improve their adaptability during the training process.
  arXiv  Detail & Related papers  (2024-02-12T22:47:57Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.