Pion modifies Muon's Newton-Schulz iterations into a controllable high-pass filter that anchors dominant singular values at 1 while suppressing noisy tails, outperforming Muon and AdamW in VLA and RLVR regimes.
Canonical reference
Tapered off-policy REINFORCE: Stable and efficient reinforcement learning for LLMs
Canonical reference. 80% of citing Pith papers cite this work as background.
13
Pith papers citing it
Background
80% of classified citations
abstract
We propose a new algorithm for fine-tuning large language models using reinforcement learning. Tapered Off-Policy REINFORCE (TOPR) uses an asymmetric, tapered variant of importance sampling to speed up learning while maintaining stable learning dynamics, even without the use of KL regularization. TOPR can be applied in a fully offline fashion, allows the handling of positive and negative examples in a unified framework, and benefits from the implementational simplicity that is typical of Monte Carlo algorithms. We demonstrate the effectiveness of our approach with a series of experiments on the GSM8K and MATH reasoning benchmarks, finding performance gains for training both a model for solution generation and as a generative verifier. We show that properly leveraging positive and negative examples alike in the off-policy regime simultaneously increases test-time accuracy and training data efficiency, all the while avoiding the ``wasted inference'' that comes with discarding negative examples. We find that this advantage persists over multiple iterations of training and can be amplified by dataset curation techniques, enabling us to match 70B-parameter model performance with 8B language models. As a corollary to this work, we find that REINFORCE's baseline parameter plays an important and unexpected role in defining dataset composition in the presence of negative examples, and is consequently critical in driving off-policy performance.
citation-role summary
method 3
background 2
citation-polarity summary
representative citing papers
Freshness-Aware PER augments prioritized experience replay with exponential age decay based on effective sample size to enable successful reuse of trajectories in LLM and VLM reinforcement learning, outperforming on-policy baselines on agentic tasks.
Replacing the importance sampling ratio with a stop-gradient self-anchored ratio for positive advantages yields unclipped, REINFORCE-equivalent gradients that improve exploration without training instability.
Theoretical analysis of RLVR update dynamics leads to ACPO, an adaptive clipping method that outperforms DAPO and CISPO on reasoning benchmarks with 3B and 7B models.
Rollout-level advantage-prioritized experience replay for GRPO recycles high-advantage individual rollouts with age eviction and fresh-anchored batches to outperform standard GRPO on math benchmarks, with gains increasing with model size.
ERPD decouples aggressive off-policy optimization on fixed trajectories from trust-region distillation to achieve comparable or better LLM performance with substantially smaller KL divergence.
Mu-GRPO enables substantially more off-policy GRPO training for LLMs via relaxed clipping and negative-advantage veto in large staged batches, matching standard GRPO performance at ~2x training speed.
Missing old logits in async agentic RL entangle discrepancy and staleness terms in PPO off-policy correction; exact acquisition methods and revised PPO-EWMA restore decoupled updates with reported gains in speed and performance.
Survey mapping RL techniques onto LLM training and highlighting gaps in value-based, off-policy, and bootstrapping methods.
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.
R²VPO uses ratio-variance regularization as a distributional soft brake on policy updates, claiming better performance than PPO on math reasoning and robotic control without hard clipping.
Introduces polychromic objectives adapted into PPO via vine sampling and modified advantages, showing higher success rates and better coverage under perturbations on BabyAI, Minigrid, and algorithmic tasks.
An expository book that systematically presents RLHF methods, from reward modeling to direct alignment algorithms, aimed at readers with quantitative backgrounds.
citing papers explorer
-
Rethinking Muon Beyond Pretraining: Spectral Failures and High-Pass Remedies for VLA and RLVR
Pion modifies Muon's Newton-Schulz iterations into a controllable high-pass filter that anchors dominant singular values at 1 while suppressing noisy tails, outperforming Muon and AdamW in VLA and RLVR regimes.
-
Freshness-Aware Prioritized Experience Replay for LLM/VLM Reinforcement Learning
Freshness-Aware PER augments prioritized experience replay with exponential age decay based on effective sample size to enable successful reuse of trajectories in LLM and VLM reinforcement learning, outperforming on-policy baselines on agentic tasks.
-
UP: Unbounded Positive Asymmetric Optimization for Breaking the Exploration-Stability Dilemma
Replacing the importance sampling ratio with a stop-gradient self-anchored ratio for positive advantages yields unclipped, REINFORCE-equivalent gradients that improve exploration without training instability.
-
What are Key Factors for Updates in RL for LLM Reasoning?
Theoretical analysis of RLVR update dynamics leads to ACPO, an adaptive clipping method that outperforms DAPO and CISPO on reasoning benchmarks with 3B and 7B models.
-
Rollout-Level Advantage-Prioritized Experience Replay for GRPO
Rollout-level advantage-prioritized experience replay for GRPO recycles high-advantage individual rollouts with age eviction and fresh-anchored batches to outperform standard GRPO on math benchmarks, with gains increasing with model size.
-
Extreme Region Policy Distillation
ERPD decouples aggressive off-policy optimization on fixed trajectories from trust-region distillation to achieve comparable or better LLM performance with substantially smaller KL divergence.
-
How Off-Policy Can GRPO Be? Mu-GRPO for Efficient LLM Reinforcement Learning
Mu-GRPO enables substantially more off-policy GRPO training for LLMs via relaxed clipping and negative-advantage veto in large staged batches, matching standard GRPO performance at ~2x training speed.
-
Missing Old Logits in Asynchronous Agentic RL: Semantic Mismatch and Repair Methods for Off-Policy Correction
Missing old logits in async agentic RL entangle discrepancy and staleness terms in PPO off-policy correction; exact acquisition methods and revised PPO-EWMA restore decoupled updates with reported gains in speed and performance.
-
Modularized Reinforcement Learning on LLMs: From MDP Creation to Exploration and Learning
Survey mapping RL techniques onto LLM training and highlighting gaps in value-based, off-policy, and bootstrapping methods.
-
Trust Region On-Policy Distillation
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.
-
Ratio-Variance Regularized Policy Optimization
R²VPO uses ratio-variance regularization as a distributional soft brake on policy updates, claiming better performance than PPO on math reasoning and robotic control without hard clipping.
-
Polychromic Objectives for Reinforcement Learning
Introduces polychromic objectives adapted into PPO via vine sampling and modified advantages, showing higher success rates and better coverage under perturbations on BabyAI, Minigrid, and algorithmic tasks.
-
Reinforcement Learning from Human Feedback
An expository book that systematically presents RLHF methods, from reward modeling to direct alignment algorithms, aimed at readers with quantitative backgrounds.