REVIEW 44 cited by
Pass@k Training for Adaptively Balancing Exploration and Exploitation of Large Reasoning Models
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Pass@k Training for Adaptively Balancing Exploration and Exploitation of Large Reasoning Models
read the original abstract
Reinforcement learning with verifiable rewards (RLVR), which typically adopts Pass@1 as the reward, has faced the issues in balancing exploration and exploitation, causing policies to prefer conservative actions, converging to a local optimum. Identifying an appropriate reward metric is therefore crucial. Regarding the prior work, although Pass@k has been used in evaluation, its connection to LLM exploration ability in RLVR remains largely overlooked. To investigate this, we first use Pass@k as the reward to train the policy model (i.e., $\textbf{Pass@k Training}$), and observe the improvement on its exploration ability. Next, we derive an analytical solution for the advantage of Pass@k Training, leading to an efficient and effective process. Building on this, our analysis reveals that exploration and exploitation are not inherently conflicting objectives, while they can mutually enhance each other. Moreover, Pass@k Training with analytical derivation essentially involves directly designing the advantage function. Inspired by this, we preliminarily explore the advantage design for RLVR, showing promising results and highlighting a potential future direction.
Forward citations
Cited by 44 Pith papers
-
DecompRL: Solving Harder Problems by Learning Modular Code Generation
DecompRL is an RL method that learns modular code decomposition for LLMs, enabling exponential candidate generation via recombination to solve harder coding problems with lower GPU cost.
-
OrderGrad: Optimizing Beyond the Mean with Order-Statistic Policy Gradient Estimation
OrderGrad supplies unbiased likelihood-ratio and reparameterization gradient estimators for finite-sample L-statistics by applying a rank-based reward transformation usable in standard policy-gradient updates.
-
Exploiting Verification-Generation Gap: Test-Time Reinforcement Learning with Confidence-Conditioned Verification
TTRL-CoCoV is a confidence-conditioned test-time RL framework that selectively applies verification to address pseudo-label errors and diversity collapse, yielding +9.8% Pass@1 and +18.7% Pass@16 gains over prior TTRL...
-
Residual Skill Optimization for Text-to-SQL Ensembles
Residual skill optimization creates complementary Text-to-SQL agents by training each new skill on prior ensemble failures, yielding accuracy gains on Spider2-Lite and transfer to other dialects and tasks.
-
Finite-Time Regret Analysis of Retry-Aware Bandits
ReMax achieves the first sublinear finite-time regret bound for Gaussian bandits with M=2 by deriving an expected-improvement balance condition for its optimal sampling distribution and separating saturation from unde...
-
Finite-Time Regret Analysis of Retry-Aware Bandits
ReMax achieves the first sublinear regret bound for Gaussian rewards at M=2 by characterizing the optimal sampling distribution via an expected-improvement balance condition and separating saturation from underestimat...
-
Breaking $\textit{Winner-Takes-All}$: Cooperative Policy Optimization Improves Diverse LLM Reasoning
GCPO shifts RLVR from rollout competition to team cooperation by assigning advantages via marginal contributions to a determinant-based coverage volume over semantic embeddings, yielding higher accuracy and solution d...
-
Breaking $\textit{Winner-Takes-All}$: Cooperative Policy Optimization Improves Diverse LLM Reasoning
GCPO uses team-level credit assignment via determinant volume over reward-weighted semantic embeddings to promote non-redundant correct reasoning paths, improving both accuracy and diversity in LLM training.
-
Rebellious Student: Reversing Teacher Signals for Reasoning Exploration with Self-Distilled RLVR
RLRT augments GRPO by reinforcing tokens on correct student rollouts that the teacher would not have predicted, outperforming standard self-distillation and exploration baselines on Qwen3 models.
-
ResRL: Boosting LLM Reasoning via Negative Sample Projection Residual Reinforcement Learning
ResRL decouples shared semantics between positive and negative responses in LLM reinforcement learning via SVD-based projection residuals, outperforming baselines including NSR by up to 9.4% on math reasoning benchmarks.
-
Uniform-Correct Policy Optimization: Breaking RLVR's Indifference to Diversity
UCPO modifies GRPO with a uniformity penalty over correct solutions to prevent diversity collapse in RLVR, yielding up to 10% higher Pass@64 on AIME24 and 45% more equation-level diversity.
-
Selective Expert Guidance for Effective and Diverse Exploration in Reinforcement Learning of LLMs
MENTOR selectively mixes expert and policy token distributions at high-entropy decision points, improving RLVR accuracy and diversity on math and out-of-domain benchmarks compared with full-trajectory imitation.
-
Spectral Rewiring for Exploration, Purification, and Model Merging
Subspace-Aligned Rewiring projects RL weight updates onto the base model’s SVD basis, retaining a compact rewiring matrix that preserves reasoning and improves exploration and multi-domain merging.
-
Beyond the Mean: Multi-Moment Policy Optimization for LLM Reasoning
MMPO jointly minimizes the first T moments of the per-problem failure-probability distribution — equivalent to minimizing expected truncated time to first success — and reports average gains over single-moment RL obje...
-
Instruction-Conditioned Exploration with Asymmetric Reinforcement Learning and Self-Distillation
Adding behavioral strategy hints to RL training prompts, then distilling only correct hint-conditioned answers into the unconditioned model, lifts Qwen3-1.7B math pass@1 by 5.0% over DAPO.
-
Ring-Zero: Scaling Zero RL to a Trillion Parameters for Emergent Reasoning
Scaling zero-RL training to a 1T-parameter model yields competitive math performance and spontaneous cognitive behaviors such as self-verification and structured formatting.
-
Ring-Zero: Scaling Zero RL to a Trillion Parameters for Emergent Reasoning
A trillion-parameter model trained purely on reward feedback reaches competitive math scores and produces structured, self-verifying reasoning traces, but the claim that these behaviors emerge without hand-crafted rul...
-
Don't Let Gains FADE: Breaking Down Policy Gradient Weights in RL
FADE is a self-adapting advantage for policy-gradient RL that reads training dynamics to balance positive/negative gradient mass and difficulty focus, yielding faster peak performance and better accuracy-diversity tra...
-
On-Policy Self-Distillation with Sampled Demonstrations Reduces Output Diversity
On-policy self-distillation with sampled demonstrations reduces rollout diversity by amplifying existing probability gaps in the base model, unlike ideal RL which preserves ratios among correct outputs.
-
Self-Improvement Can Self-Regress: The Rise-and-Collapse Failure Mode of LLM Self-Training
REINFORCE self-training on competitive programming tasks exhibits robust rise-then-collapse in pass@1; CARE, ES, and GRPO mitigate it in model-size-dependent ways across Qwen-2.5-3B/7B and a Gemma pilot.
-
REVES: REvision and VErification--Augmented Training for Test-Time Scaling
REVES augments LLM post-training by decoupling revision and verification signals from successful multi-step trajectories, reporting +6.5 point gains on LiveCodeBench over RL baselines.
-
On Advantage Estimates for Max@K Policy Gradients
Proposes MaxPO using a Leave-Two-Out baseline for centered unbiased advantages in max@K policy gradients, with a unified derivation of finite-batch estimators.
-
Retry Policy Gradients in Continuous Action Spaces
ReMAC applies pathwise estimators to retry objectives in continuous RL, reshaping gradients to increase policy entropy and matching SAC performance without explicit regularization.
-
Multi-Step Likelihood-Ratio Correction for Reinforcement Learning with Verifiable Rewards
NFPO augments the PPO surrogate with N-step forward traces to bridge local approximations and exact policy gradients, delivering tighter policy-improvement bounds and improved results on reasoning benchmarks.
-
Beyond Mode Collapse: Distribution Matching for Diverse Reasoning
DMPO approximates forward KL minimization in on-policy RL by aligning the policy to a group-level reward-proportional target distribution, yielding 9-12% relative gains over GRPO on NP-Bench and smaller gains on math ...
-
SAGE: Shaping Anchors for Guided Exploration in RLVR of LLMs
SAGE reshapes the reverse-KL anchor via guide function q(x,y) for controllable empirical support expansion, yielding gains in both pass@1 and pass@k on math reasoning benchmarks.
-
ResRL: Boosting LLM Reasoning via Negative Sample Projection Residual Reinforcement Learning
ResRL boosts LLM reasoning by modulating negative gradients with SVD-based projection residuals from negative samples, outperforming NSR by 9.4% Avg@16 on math benchmarks while preserving diversity across 12 tasks.
-
SPS: Steering Probability Squeezing for Better Exploration in Reinforcement Learning for Large Language Models
SPS interleaves RL and IRL to counteract probability squeezing in LLM reasoning trajectories, improving Pass@k on five benchmarks while identifying an empirical upper bound on multi-sample performance.
-
From $P(y|x)$ to $P(y)$: Investigating Reinforcement Learning in Pre-train Space
PreRL applies reward-driven updates to P(y) in pre-train space, uses Negative Sample Reinforcement to prune bad reasoning paths and boost reflection, and combines with standard RL in Dual Space RL to outperform baseli...
-
Compress the Easy, Explore the Hard: Difficulty-Aware Entropy Regularization for Efficient LLM Reasoning
CEEH selectively applies entropy regularization to hard questions and a shortest-correct-length penalty to easy ones, reducing reasoning length while preserving accuracy across six math benchmarks.
-
Towards Generalizable Reasoning: Group Causal Counterfactual Policy Optimization for LLM Reasoning
Group Causal Counterfactual Policy Optimization trains LLMs on generalizable reasoning by defining episodic rewards for counterfactual robustness and transferability then optimizing the policy with token-level advantages.
-
Beyond the Sampled Token: Preserving Candidate Support in RLVR
SimKO, an asymmetric top-K token-level regularizer for RLVR, improves pass@K on math/logic benchmarks across K=1..256 without lowering pass@1 relative to GRPO.
-
Emergent Slow Thinking in LLMs as Inverse Tree Freezing
RLVR drives a concept network in LLMs through nucleation and freezing into inverse trees that support slow thinking, and intervening with brief SFT at peak frustration outperforms standard RLVR while post-freeze SFT c...
-
Outcome-based Exploration for LLM Reasoning
Outcome-based exploration bonuses (UCB-Con and Batch) improve pass@1 and pass@32 for LLM math reasoning while slowing diversity collapse, supported by a bandit model with a strong generalization assumption.
-
The Landscape of Agentic Reinforcement Learning for LLMs: A Survey
Survey that defines agentic RL for LLMs via POMDPs, introduces a taxonomy of planning/tool-use/memory/reasoning capabilities and domains, and compiles open environments from over 500 papers.
-
Explorative Modeling: Unlocking a Third Pretraining Axis and End-to-End Generation
Training on the best of K generated candidates improves image, video, and language generative models, with the reported gains growing with scale and enabling single-pass end-to-end generation.
-
SFT Overtraining Predicts Rank Inversion via Entropy Collapse Under RLVR
SFT depth increases pre-RL pass@1 but can cause entropy collapse that inverts GRPO outcomes on Qwen models via reduced group advantage variance.
-
How You Begin is How You Reason: Driving Exploration in RLVR via Prefix-Tuned Priors
IMAX trains soft prefixes with an InfoMax reward to drive diverse exploration in RLVR, yielding up to 11.60% gains in Pass@4 over standard RLVR across model scales.
-
PACEvolve++: Improving Test-time Learning for Evolutionary Search Agents
PACEvolve++ uses a phase-adaptive reinforcement learning advisor to decouple hypothesis selection from execution in LLM-driven evolutionary search, delivering faster convergence than prior frameworks on load balancing...
-
On the Implicit Reward Overfitting and the Low-rank Dynamics in RLVR
RLVR exhibits implicit reward overfitting to training data and optimizes heavy-tailed singular spectra with rank-1 focus on reasoning capability.
-
MCPO: Mastery-Consolidated Policy Optimization for Large Reasoning Models
MCPO fixes vanishing training signals and shrinking weights in GRPO by using a hinge-KL regularizer on mastered prompts and prioritizing majority-correct prompts, yielding higher pass@1 and pass@k on math tasks.
-
TaPR: Test-Aware Policy Refinement for Feedback-Conditioned Code Generation
Dense per-turn pass-ratio rewards improve multi-turn code repair (Pass@3) without changing first-attempt accuracy, across six models on LiveCodeBench.
-
Entropy-KL Divergence-based Token Masking: A Novel Approach for Selective Fine-tuning of Large Language Models
EKSFT masks high-entropy or high-KL tokens in low-data SFT to preserve pre-trained distribution and improve downstream RL performance on math reasoning tasks.
-
A Survey of Reinforcement Learning for Large Reasoning Models
A survey compiling RL methods, challenges, data resources, and applications for enhancing reasoning in large language models and large reasoning models since DeepSeek-R1.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.