pith. sign in

arxiv: 2505.07686 · v2 · pith:CB5VYPTNnew · submitted 2025-05-12 · 💻 cs.AI · cs.LG

S-GRPO: Early Exit via Reinforcement Learning in Reasoning Models

classification 💻 cs.AI cs.LG
keywords reasoningmodelss-grpoexitearlygenerationgrouplearning
0
0 comments X
read the original abstract

As Test-Time Scaling emerges as an active research focus in the large language model community, advanced post-training methods increasingly emphasize extending chain-of-thought (CoT) generation length, thereby enhancing reasoning capabilities to approach Deepseek R1-like reasoning models. However, recent studies reveal that reasoning models (even Qwen3) consistently exhibit excessive thought redundancy in CoT generation. This overthinking issue arises from the inherent limitations of conventional outcome-reward reinforcement learning, which systematically overlooks the regulation of intermediate reasoning processes. This paper introduces Serial-Group Decaying-Reward Policy Optimization (S-GRPO), a novel reinforcement learning paradigm that enables models to implicitly evaluate the sufficiency of intermediate reasoning steps, thereby facilitating early exit in CoT generation. Unlike GRPO, which samples multiple possible reasoning paths in parallel (parallel group), S-GRPO only samples one reasoning path and serially selects multiple temporal positions from the path to exit thinking and directly generate answers (serial group). For correct answers within a serial group, rewards gradually decrease based on the exit positions along the reasoning path from front to back. This design encourages the model to produce more accurate and concise thoughts, while also incentivizing early thinking termination when appropriate. Empirical evaluations demonstrate that S-GRPO is compatible with state-of-the-art reasoning models, including Qwen3 and Deepseek-distill. Across diverse benchmarks such as GSM8K, AIME 2024, AMC 2023, MATH-500, and GPQA Diamond, S-GRPO achieves a substantial reduction in sequence length (35.4% - 61.1%) while simultaneously improving accuracy (absolute 0.72% - 6.08%).

This paper has not been read by Pith yet.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 18 Pith papers

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

  1. Reinforcement Learning from Rich Feedback with Distributional DAgger

    cs.LG 2026-06 unverdicted novelty 7.0

    DistIL applies distributional DAgger with forward cross-entropy to achieve monotonic policy improvement and better Pass@N from rich feedback in RL for reasoning tasks.

  2. Post Reasoning: Improving the Performance of Non-Thinking Models at No Cost

    cs.AI 2026-05 conditional novelty 7.0

    Post-Reasoning boosts LLM accuracy by reversing the usual answer-after-reasoning order, delivering mean relative gains of 17.37% across 117 model-benchmark pairs with zero extra cost.

  3. Near-Future Policy Optimization

    cs.LG 2026-04 unverdicted novelty 7.0

    NPO uses a policy's own near-future checkpoint as auxiliary trajectories to maximize effective learning signal S = Q/V, improving performance from 57.88 to 63.15 on Qwen3-VL-8B-Instruct with GRPO while accelerating co...

  4. Generate, Filter, Control, Replay: A Comprehensive Survey of Rollout Strategies for LLM Reinforcement Learning

    cs.LG 2026-04 unverdicted novelty 7.0

    This survey introduces the Generate-Filter-Control-Replay (GFCR) taxonomy to structure rollout pipelines for RL-based post-training of reasoning LLMs.

  5. Self-Distilled RLVR

    cs.LG 2026-04 unverdicted novelty 7.0

    RLSD mixes self-distillation for token-level policy difference magnitudes with RLVR for reliable update directions from response correctness to reach higher convergence and better training stability.

  6. ThoughtFold: Folding Reasoning Chains via Introspective Preference Learning

    cs.AI 2026-06 unverdicted novelty 6.0

    ThoughtFold applies introspective redundancy detection within correct CoT trajectories to create sub-trajectory spectra, then uses masked preference optimization to penalize redundant explorations, yielding 56% token ...

  7. Learning to Solve, Forgetting to Retain: Correct-Set Turnover in RLVR

    cs.LG 2026-06 unverdicted novelty 6.0

    RLVR exhibits correct-set turnover where solved problems regress during training, and a periodic review mechanism exploiting a repair-window principle improves retention and performance over baselines.

  8. Selective Off-Policy Reference Tuning with Plan Guidance

    cs.AI 2026-05 unverdicted novelty 6.0

    SORT turns all-wrong prompts into selective learning signals by weighting tokens more predictable under plan guidance from reference solutions, improving over GRPO on reasoning benchmarks especially for weaker models.

  9. Implicit Compression Regularization: Concise Reasoning via Internal Shorter Distributions in RL Post-Training

    cs.AI 2026-05 unverdicted novelty 6.0

    ICR creates a virtual shorter distribution from shortest correct on-policy responses to regularize RL post-training toward concise yet accurate reasoning, improving the accuracy-length Pareto frontier on math and know...

  10. Co-Evolving Policy Distillation

    cs.LG 2026-04 unverdicted novelty 6.0

    CoPD integrates multiple expert capabilities by running parallel RLVR training with bidirectional online policy distillation among experts, outperforming mixed RLVR and sequential OPD while surpassing domain-specific ...

  11. GRPO-VPS: Enhancing Group Relative Policy Optimization with Verifiable Process Supervision for Effective Reasoning

    cs.LG 2026-04 unverdicted novelty 6.0

    GRPO-VPS improves GRPO by using segment-wise conditional probabilities of the correct answer to supply process-level feedback, yielding up to 2.6-point accuracy gains and 13.7% shorter reasoning on math tasks.

  12. Entropy After </Think> for reasoning model early exiting

    cs.LG 2025-09 unverdicted novelty 6.0

    Entropy After </Think> (EAT) enables early exiting in reasoning LLMs by tracking entropy stabilization after a </think> token, cutting token use 12-22% on MATH500 and AIME2025 with no accuracy loss.

  13. Trust Region On-Policy Distillation

    cs.LG 2026-05 unverdicted novelty 5.0

    TrOPD stabilizes on-policy distillation for LLMs with trust-region learning, outlier estimation, and off-policy guidance, outperforming prior OPD methods on reasoning and code benchmarks.

  14. Selective Off-Policy Reference Tuning with Plan Guidance

    cs.AI 2026-05 unverdicted novelty 5.0

    SORT converts all-failed reasoning prompts into selective, structure-aware training signals by weighting tokens according to how much a reference-derived plan increases their probability.

  15. TinyJudge: Unverifiable Constraint Alignment via Lightweight Specialist Ensembles

    cs.CL 2026-04 conditional novelty 5.0

    Distilling large LLM judge capabilities into an ensemble of tiny specialist models for soft-constraint evaluation achieves a 10% performance gain and 3x speedup in RLVR training.

  16. Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models

    cs.CL 2025-03 accept novelty 5.0

    A survey organizing techniques to achieve efficient reasoning in LLMs by shortening chain-of-thought outputs.

  17. EasyVideoR1: Easier RL for Video Understanding

    cs.CV 2026-04 unverdicted novelty 4.0

    EasyVideoR1 delivers an optimized RL pipeline for video understanding in large vision-language models, achieving 1.47x throughput gains and aligned results on 22 benchmarks.

  18. A Survey of Reinforcement Learning for Large Reasoning Models

    cs.CL 2025-09 accept novelty 3.0

    A survey compiling RL methods, challenges, data resources, and applications for enhancing reasoning in large language models and large reasoning models since DeepSeek-R1.