RELEX: Extrapolating LLM RLVR Training via Rank-1 Parameter Trajectories
This paper demonstrates that RLVR weight update trajectories are extremely low-rank and near-linearly predictable, with a rank-1 approximation capturing most downstream performance gains. The authors propose RELEX, a compute-efficient method that observes a short training window, estimates the rank-1 subspace, and extrapolates future checkpoints via linear regression—requiring no additional training. Evaluated on Qwen2.5-Math-1.5B, Qwen3-4B-Base, and Qwen3-8B-Base, RELEX matches or exceeds full RLVR performance using as few as 15% of training steps, and can extrapolate up to 10–20× beyond the observed prefix. The authors attribute the method's effectiveness to a denoising effect from rank-1 projection that discards stochastic optimization noise.
Related guides (3)
Related events (8)
RePro: Retrospective Progress-Aware Self-Refinement for LLM Agent Training
Researchers introduce RePro (Retrospective Progress-Aware Training), a framework addressing the gap between step-wise RL optimization and metacognitive task-progress awareness in LLM agents. The approach uses a forward-then-reflect rollout paradigm where agents execute actions online and then retrospectively assess step-wise progress given the completed trajectory and known outcome. Evaluated on WebShop, ALFWorld, and Sokoban, RePro achieves up to 12% absolute success rate gains over baseline Qwen-family models without requiring continuous external supervision.
ExpRL: RL-based mid-training using human QA data as reward scaffolds for LLM reasoning
ExpRL proposes an automated approach to LLM mid-training that replaces manually curated reasoning traces with large corpora of human-written QA data used as reward scaffolds rather than imitation targets. Reference solutions are hidden from the policy and used only to construct problem-specific grading rubrics, enabling dense process-level rewards that reinforce partial progress and intermediate reasoning steps. On challenging math reasoning benchmarks, ExpRL outperforms SFT, sparse-reward GRPO, and self-distillation as an RL initialization strategy, with additional mixed-domain experiments suggesting broader applicability.
LongTraceRL: Reinforcement Learning for Long-Context Reasoning via Search Agent Trajectories and Rubric Rewards
LongTraceRL is a new RL training framework for improving long-context reasoning in LLMs, addressing limitations of existing RLVR methods. It constructs challenging training data using multi-hop questions from knowledge graph random walks and tiered distractors derived from search agent trajectories (high-confusability: read but uncited; low-confusability: seen but unopened). A rubric reward provides entity-level process supervision along reasoning chains, applied only to correct responses to prevent reward hacking. Experiments across three LLMs (4B–30B parameters) on five long-context benchmarks show consistent improvements over strong baselines.
Bebop: MTP with rejection sampling and TV loss achieves 1.8x RL training speedup
Researchers introduce Bebop, a framework for integrating Multi-Token Prediction (MTP) into large-scale RL training pipelines for LLMs. The work identifies that MTP acceptance rates degrade during RL due to entropy fluctuations, and proposes probabilistic rejection sampling plus a novel end-to-end Total Variation (TV) loss that directly optimizes multi-step acceptance rates, achieving up to 95% acceptance rates and 25% extra inference throughput gains. Applied to Qwen3.5, Qwen3.6, and Qwen3.7 models, the method yields up to 1.8x end-to-end acceleration in async RL training. The approach eliminates the need for costly online MTP updating by using pre-RL MTP training with the proposed objectives.
Reroute: Training-free recoverable visual token routing for vision-language models
A new arXiv preprint proposes Reroute, a training-free plug-in that replaces the standard rank-and-remove visual token pruning paradigm in VLMs with a recoverable routing mechanism. Instead of permanently discarding low-ranked tokens, Reroute defers them to re-enter the candidate pool at later decoder stages, addressing the problem that token importance shifts across decoder depth. Evaluated on LLaVA-1.5 and Qwen backbones augmented with FastV, PDrop, and Nüwa pruning methods, Reroute improves grounding performance under aggressive token reduction without sacrificing general VQA accuracy. The approach preserves the theoretical compute and KV-cache budget of the underlying pruning method.
Extrapolative Weight Averaging Reveals Correctness-Efficiency Frontiers in Code RL
This paper investigates whether extrapolative weight averaging of RL-trained checkpoints can extend Pareto frontiers between competing objectives (correctness vs. computational efficiency) without additional training. Starting from a shared initialization, the authors train checkpoints under nested unit-test coverage regimes for competitive programming tasks, revealing a correctness-efficiency frontier where higher-coverage rewards reduce optimization failures but increase correctness failures. Extrapolation beyond trained endpoints produces complementary policies that, when ensembled, improve pass@250 on LCB/hard by 3.3% over the best single checkpoint at matched sample budget. Results hold across 7B and 32B model scales and three inference settings: pure reasoning, tool use, and agentic coding.
QwQ-32B: Scaling Reinforcement Learning for Enhanced Reasoning
Alibaba's Qwen team releases QwQ-32B, a 32-billion parameter model trained with scaled Reinforcement Learning to improve reasoning capabilities beyond conventional pretraining and post-training methods. The release draws explicit comparison to DeepSeek R1's cold-start and multi-stage RL training approach. The model is available via Qwen Chat, Hugging Face, ModelScope, and a demo interface. This represents Qwen's exploration of RL scalability as a path to enhanced LLM intelligence.
MAST: Mechanism-guided selective unlearning for RLVR-trained reasoning models
Researchers introduce MAST (Mechanism-Aligned Selective Targeting), a method for selectively unlearning capabilities induced by reinforcement learning from verifiable rewards (RLVR) in language models while minimizing collateral damage to retained knowledge. The approach ranks attention-projection tensors by off-principal energy and gradient coupling to identify a targeted subset for update, rather than applying full-parameter gradient ascent. Evaluated on Qwen2.5-Math-1.5B and Qwen3-1.7B-Base, MAST achieves statistically significant forgetting on target MATH problems while preserving GSM8K performance, whereas full-parameter unlearning collapses retained capabilities. The method generalizes across seeds and unlearning objectives (NPO/SimNPO).