You Only Need Minimal RLVR Training: Extrapolating LLMs via Rank-1 Trajectories

• URL: http://arxiv.org/abs/2605.21468v1
• Date: Wed, 20 May 2026 17:53:20 GMT
• Title: You Only Need Minimal RLVR Training: Extrapolating LLMs via Rank-1 Trajectories
• Abstract summary: We show thatReinforcement learning with verifiable rewards (RLVR) is extremely low-rank and highly predictable.<n>We propose a simple and compute-efficient method RELEX, which estimates the rank-1 subspace from a short observation window.<n>Remarkably, RELEX is able to extrapolate far beyond the observation window at no training cost.
• Score: 23.542887618146988
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Reinforcement learning with verifiable rewards (RLVR) has become a dominant paradigm for improving reasoning in large language models (LLMs), yet the underlying geometry of the resulting parameter trajectories remains underexplored. In this work, we demonstrate that RLVR weight trajectories are extremely low-rank and highly predictable. Specifically, we find that the majority of downstream performance gains are captured by a rank-1 approximation of the parameter deltas, where the magnitude of this projection evolves near-linearly with training steps. Motivated by this, we propose a simple and compute-efficient method RELEX (REinforcement Learning EXtrapolation), which estimates the rank-1 subspace from a short observation window and extrapolates future checkpoints via linear regression, with no learned model required. Across three models (i.e., Qwen2.5-Math-1.5B, Qwen3-4B-Base, and Qwen3-8B-Base), RELEX produces checkpoints that match or exceed RLVR performance on both in-domain and out-of-domain benchmarks, requiring as few as 15% steps of full RLVR training. Remarkably, RELEX is able to extrapolate far beyond the observation window at no training cost, predicting checkpoints up to 10-20$\times$ beyond the observed prefix with continued improvement (e.g., observe only the first 50 steps and extrapolate to 1000 steps). Our ablation analysis confirms the minimalist sufficiency of RELEX: neither increasing the subspace rank nor employing non-linear modeling yields further gains in extrapolation. Finally, we show that RELEX's success stems from a "denoising" effect: by projecting updates onto the rank-1 subspace, the model discards stochastic optimization noise that would otherwise degrade performance during extrapolation. Our code is available at https://github.com/weizhepei/RELEX.

Related papers

• On the Implicit Reward Overfitting and the Low-rank Dynamics in RLVR [51.935533482549545]
  We show that RLVR may exhibit implicit reward overfitting to the training dataset.<n>The model can achieve satisfactory performance on the test set even when its rewards remain relatively low during the training process.
  arXiv  Detail & Related papers  (2026-05-07T16:30:28Z)
• Low-rank Optimization Trajectories Modeling for LLM RLVR Acceleration [88.7332439088792]
  scaling reinforcement learning with verifiable rewards (RLVR) for large language models (LLMs) has emerged as an effective training paradigm.<n>We propose the textbfNonlinear textbfExtrapolation of low-rank trajectories (textbfNExt)<n>Our method reduces computational overhead by approximately 37.5% while remaining compatible with a wide range of RLVR algorithms and tasks.
  arXiv  Detail & Related papers  (2026-04-13T13:28:12Z)
• REAL: Regression-Aware Reinforcement Learning for LLM-as-a-Judge [83.2858110368572]
  We propose textbfREAL (underlineREgression-underlineAware Reinforcement underlineLThought), a principled RL framework designed to optimize regression rewards.<n>We show that REAL consistently outperforms both regression-aware SFT baselines and standard RL methods.
  arXiv  Detail & Related papers  (2026-03-17T21:19:08Z)
• Not All Steps are Informative: On the Linearity of LLMs' RLVR Training [14.59942263367421]
  Reinforcement learning with verifiable rewards (RLVR) has become a central component of large language model (LLM) post-training.<n>We investigate whether future model states can be predicted from intermediate checkpoints via extrapolation, avoiding continued expensive training.<n>We show that Weight Extrapolation produces models with performance comparable to standard RL training while requiring significantly less computation.
  arXiv  Detail & Related papers  (2026-01-08T03:06:18Z)
• Masked-and-Reordered Self-Supervision for Reinforcement Learning from Verifiable Rewards [13.064343544668283]
  We propose MR-RLVR (Masked-and-Reordered RLVR), which constructs process-level self-supervised rewards via "masked-then-fill" and "step reordering"<n>We implement MR-RLVR on Qwen2.5-3B and DeepSeek-R1-Distill-Qwen-1.5B, and evaluate on AIME24, AIME25, AMC23, and MATH500.
  arXiv  Detail & Related papers  (2025-11-21T18:23:04Z)
• The Path Not Taken: RLVR Provably Learns Off the Principals [85.41043469428365]
  We show that sparsity is a surface artifact of a model-conditioned optimization bias.<n>We mechanistically explain these dynamics with a Three-Gate Theory.<n>We provide a parameter-level characterization of RLVR's learning dynamics.
  arXiv  Detail & Related papers  (2025-11-11T18:49:45Z)
• Reinforcement Learning for Reasoning in Large Language Models with One Training Example [117.86853102104256]
  We show that reinforcement learning with verifiable reward using one training example (1-shot RLVR) is effective in incentivizing the math reasoning capabilities of large language models (LLMs)<n>We identify some interesting phenomena during 1-shot RLVR, including cross-category generalization, increased frequency of self-reflection, and sustained test performance improvement.
  arXiv  Detail & Related papers  (2025-04-29T09:24:30Z)
• Understanding R1-Zero-Like Training: A Critical Perspective [73.25430192337235]
  We critically examine R1-Zero-like training by analyzing its two core components: base models and RL.<n>We investigate a wide range of base models, including DeepSeek-V3-Base, to understand how pretraining characteristics influence RL performance.<n>We present a minimalist R1-Zero recipe that achieves 43.3% accuracy on AIME 2024 with a 7B base model.
  arXiv  Detail & Related papers  (2025-03-26T17:59:14Z)

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.