Pith. sign in

REVIEW 7 cited by

Offline Reinforcement Learning for LLM Multi-Step Reasoning

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2412.16145 v2 pith:5RPKB5NP submitted 2024-12-20 cs.LG cs.AIcs.CL

classification cs.LGcs.AIcs.CL
keywords reasoningmulti-stepofflinetaskslearningreinforcementassignmentavailable
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Improving the multi-step reasoning ability of large language models (LLMs) with offline reinforcement learning (RL) is essential for quickly adapting them to complex tasks. While Direct Preference Optimization (DPO) has shown promise in aligning LLMs with human preferences, it is less suitable for multi-step reasoning tasks because (1) DPO relies on paired preference data, which is not readily available for multi-step reasoning tasks, and (2) it treats all tokens uniformly, making it ineffective for credit assignment in multi-step reasoning tasks, which often come with sparse reward. In this work, we propose OREO (Offline Reasoning Optimization), an offline RL method for enhancing LLM multi-step reasoning. Building on insights from previous works of maximum entropy reinforcement learning, it jointly learns a policy model and value function by optimizing the soft Bellman Equation. We show in principle that it reduces the need to collect pairwise data and enables better credit assignment. Empirically, OREO surpasses existing offline learning methods on multi-step reasoning benchmarks, including mathematical reasoning tasks (GSM8K, MATH) and embodied agent control (ALFWorld). The approach can be extended to a multi-iteration framework when additional resources are available. Furthermore, the learned value function can be leveraged to guide the tree search for free, which can further boost performance during test time.

Discussion (0). Sign in to comment.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. LeAct: Learning to Reason from Expert Actions

    cs.LG 2026-07 conditional novelty 6.0 of 10

    An AI can learn to reason by sampling explanations for an expert's actions and keeping only the ones that help it predict those actions.

  2. Unleashing Embodied Task Planning Ability in LLMs via Reinforcement Learning

    cs.CL 2025-06 conditional novelty 6.0 of 10

    A 7B LLM trained with sparse completion rewards and a GRPO-style algorithm reaches state-of-the-art on ALFWorld and ScienceWorld.

  3. Think Clearly: Improving Reasoning via Redundant Token Pruning

    cs.AI 2025-06 conditional novelty 6.0 of 10

    A training-free test-time method prunes low-attention reasoning tokens from the KV cache, guided by an injected end-of-thinking token, and reports accuracy gains on math competition benchmarks.

  4. CRAFT: Learn the Schema, Execute the Plan

    cs.AI 2026-06 conditional novelty 5.0 of 10

    CRAFT, a two-stage post-training recipe that strips schema documentation from prompts and uses execution-grounded reinforcement learning, reports improved enterprise coding-agent quality at roughly 9x lower input-token cost.

  5. Reinforcement Learning in hyperbolic space for multi-step reasoning

    cs.LG 2025-07 reject novelty 4.0 of 10

    Hyperbolic transformer policies are claimed to beat vanilla transformer policies by 32-45% on a handful of reasoning and control problems, but the evidence is too weak to support the claim.

  6. A Technical Survey of Reinforcement Learning Techniques for Large Language Models

    cs.AI 2025-07 conditional novelty 3.0 of 10

    A survey of RL methods for LLMs that organizes the field by reward modeling, feedback source, and optimization strategy, with benchmark tables favoring a scalar-regression UNA variant over DPO and KTO in offline alignment.

  7. Response-Level Rewards Are All You Need for Online Reinforcement Learning in LLMs: A Mathematical Perspective

    cs.LG 2025-06 conditional novelty 3.0 of 10

    Under the assumption that the response reward equals the discounted sum of token rewards, response-level rewards suffice for unbiased token-level policy gradients in LLMs.

Pith tools