RL Is Neither a Panacea Nor a Mirage: Understanding Supervised vs. Reinforcement Learning Fine-Tuning for LLMs

• URL: http://arxiv.org/abs/2508.16546v1
• Date: Fri, 22 Aug 2025 17:10:37 GMT
• Title: RL Is Neither a Panacea Nor a Mirage: Understanding Supervised vs. Reinforcement Learning Fine-Tuning for LLMs
• Authors: Hangzhan Jin, Sicheng Lv, Sifan Wu, Mohammad Hamdaqa,
• Abstract summary: Training large language models (LLMs) from scratch is increasingly impractical, making post-training methods such as supervised fine-tuning central to modern practice.<n>Using an out-of-distribution (OOD) variant of the 24-point card game and new spectrum-based diagnostics, we revisit how these two stages reshape model representation and OOD performance.
• Score: 1.772462140285781
• License: http://creativecommons.org/publicdomain/zero/1.0/
• Abstract: Training large language models (LLMs) from scratch is increasingly impractical, making post-training methods such as supervised fine-tuning (SFT) and reinforcement-learning fine-tuning (RL-FT, e.g., PPO) central to modern practice. Using an out-of-distribution (OOD) variant of the 24-point card game and new spectrum-based diagnostics, we revisit how these two stages reshape model representation and OOD performance. Our key findings are- (1) RL-FT can restore much of the OOD performance loss from SFT (e.g., Llama-11B 8.97% to 15.38%, Qwen-7B 17.09% to 19.66%). But when SFT induces severe overfitting and a clear distribution shift, RL-FT cannot fully recover OOD performance. (2) Direction shifts of singular vectors matter more than singular value magnitudes. These shifts concentrate on directions linked to the largest and smallest singular values, leaving the bulk spectrum intact. (3) Low-rank and shallow recovery is effective: restoring singular vector directions for the top 20% of values or first 25% of layers recovers 70-80% of OOD performance. (4) Stronger SFT checkpoints enable better recovery by RL, while overfitted ones resist restoration. These results reconcile prior reports of RL superior OOD performance: RL primarily counteracts SFT-induced directional drift rather than finding new solutions. Our spectrum-aware analysis highlights inexpensive recovery knobs low-rank UV merging and shallow-layer resets that practitioners can use before costly RL fine-tuning.

Related papers

• Good SFT Optimizes for SFT, Better SFT Prepares for Reinforcement Learning [8.550698116833123]
  Post-training of reasoning LLMs typically consists of an offline SFT stage followed by an online reinforcement learning stage.<n>We show that, after identical RL training, models from stronger SFT checkpoints can significantly underperform those from weaker ones.<n>We propose PEAR, an SFT-stage method that corrects this mismatch and better prepares the model for RL.
  arXiv  Detail & Related papers  (2026-02-01T06:53:45Z)
• On the Non-decoupling of Supervised Fine-tuning and Reinforcement Learning in Post-training [10.433802085981046]
  Post-training of large language models routinely interleaves supervised fine-tuning (SFT) with reinforcement learning (RL)<n>We show that RL increases SFT loss under SFT optimality and that SFT lowers the reward achieved by RL.<n> Experiments on Qwen3-0.6B confirm the predicted degradation, verifying that SFT and RL cannot be separated without loss of prior performance in the post-training.
  arXiv  Detail & Related papers  (2026-01-12T10:14:09Z)
• Trust-Region Adaptive Policy Optimization [82.09255251747818]
  Post-training methods play an important role in improving large language models' (LLMs) complex reasoning abilities.<n>We introduce TRAPO, a framework that interleavesSupervised Fine-Tuning (SFT) and Reinforcement Learning (RL) within each training instance.<n>Experiments on five mathematical reasoning benchmarks show that TRAPO consistently surpasses standard SFT, RL, and SFT-then-RL pipelines.
  arXiv  Detail & Related papers  (2025-12-19T14:37:07Z)
• Quagmires in SFT-RL Post-Training: When High SFT Scores Mislead and What to Use Instead [20.446287312285648]
  We study whether high SFT scores translate to improved performance after RL.<n>We find high SFT scores can be biased toward simpler or more homogeneous data and are not reliably predictive of subsequent RL gains or scaled-up post-training effectiveness.<n>We study alternative metrics and identify generalization loss on held-out reasoning examples and Pass@large k performance to provide strong proxies for the RL outcome.
  arXiv  Detail & Related papers  (2025-10-02T02:57:00Z)
• RL Fine-Tuning Heals OOD Forgetting in SFT [35.01074051556079]
  We investigate the evolution and mechanism behind the synergy ofSupervised Fine-Tuning and Reinforcement Learning.<n>Our findings re-identify the roles of SFT and RL in the two-stage fine-tuning and discover the rotation of singular vectors as the key mechanism.
  arXiv  Detail & Related papers  (2025-09-08T21:40:41Z)
• Shuffle-R1: Efficient RL framework for Multimodal Large Language Models via Data-centric Dynamic Shuffle [53.239242017802056]
  Reinforcement learning (RL) has emerged as an effective post-training paradigm for enhancing the reasoning capabilities of multimodal large language model (MLLM)<n>However, current RL pipelines often suffer from training inefficiencies caused by two underexplored issues: Advantage Collapsing and Rollout Silencing.<n>We propose Shuffle-R1, a simple yet principled framework that improves RL fine-tuning efficiency by dynamically restructuring trajectory sampling and batch composition.
  arXiv  Detail & Related papers  (2025-08-07T17:53:47Z)
• When Imitation Learning Outperforms Reinforcement Learning in Surgical Action Planning [38.040478801450774]
  We present the first comprehensive comparison of imitation learning (IL) and reinforcement learning (RL) for surgical action planning on CholecT50.<n>Our DARIL baseline achieves 34.6% action triplet recognition mAP and 33.6% next frame prediction mAP with smooth planning degradation to 29.2% at 10-second horizons.<n>Our analysis reveals that distribution matching on expert-annotated test sets systematically favors IL over potentially valid RL policies.
  arXiv  Detail & Related papers  (2025-07-07T13:49:57Z)
• Beyond Accuracy: Dissecting Mathematical Reasoning for LLMs Under Reinforcement Learning [93.00629872970364]
  Reinforcement learning (RL) has become the dominant paradigm for improving the performance of language models on complex reasoning tasks.<n>We introduce SPARKLE, a fine-grained analytic framework to dissect the effects of RL across three key dimensions.<n>We study whether difficult problems -- those yielding no RL signals and mixed-quality reasoning traces -- can still be effectively used for training.
  arXiv  Detail & Related papers  (2025-06-05T07:53:59Z)
• How Much Backtracking is Enough? Exploring the Interplay of SFT and RL in Enhancing LLM Reasoning [6.92510069380188]
  We investigate the dynamics between SFT and RL on eight reasoning tasks.<n>Short CoT sequences used in SFT as a warm-up do have moderate contribution to RL training, compared with cold-start RL.<n>We find that longer CoT with backtracks generally induce better and more stable RL training.
  arXiv  Detail & Related papers  (2025-05-30T06:49:00Z)
• Towards Revealing the Effectiveness of Small-Scale Fine-tuning in R1-style Reinforcement Learning [4.967941028121525]
  R1-style Reinforcement Learning (RL) significantly enhances Large Language Models' reasoning capabilities.<n>We found that small-scale SFT has substantial influence on RL but shows poor efficiency.<n>We propose textbfRe-distillation, a technique that aims to boost the effectiveness of small-scale distillation.
  arXiv  Detail & Related papers  (2025-05-23T14:55:22Z)
• AceReason-Nemotron: Advancing Math and Code Reasoning through Reinforcement Learning [50.02117478165099]
  We show that large-scale reinforcement learning can significantly enhance the reasoning capabilities of strong, small- and mid-sized models.<n>We propose a simple yet effective approach: first training on math-only prompts, then on code-only prompts.
  arXiv  Detail & Related papers  (2025-05-22T08:50:47Z)
• OpenVLThinker: Complex Vision-Language Reasoning via Iterative SFT-RL Cycles [91.88062410741833]
  We introduce OpenVLThinker, one of the first open-source large vision-language models (LVLMs) to exhibit sophisticated chain-of-thought reasoning.<n>We show that OpenVLThinker-7B consistently advances performance across six benchmarks demanding mathematical and general reasoning.
  arXiv  Detail & Related papers  (2025-03-21T17:52:43Z)
• Learning to Prune Deep Neural Networks via Reinforcement Learning [64.85939668308966]
  PuRL is a deep reinforcement learning based algorithm for pruning neural networks. It achieves sparsity and accuracy comparable to current state-of-the-art methods.
  arXiv  Detail & Related papers  (2020-07-09T13:06:07Z)

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.