{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:OEOH5MG6DJODA2K45PBHR3HXE6","short_pith_number":"pith:OEOH5MG6","schema_version":"1.0","canonical_sha256":"711c7eb0de1a5c30695cebc278ecf7278ce183451ed9b91ea0cde980ff548bb4","source":{"kind":"arxiv","id":"2508.10751","version":1},"attestation_state":"computed","paper":{"title":"Pass@k Training for Adaptively Balancing Exploration and Exploitation of Large Reasoning Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Guang Shi, Qinghao Ye, Wayne Xin Zhao, Xiaobo Qin, Youbin Wu, Yue Ling, Zhipeng Chen","submitted_at":"2025-08-14T15:34:47Z","abstract_excerpt":"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"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2508.10751","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-08-14T15:34:47Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"9ad8bb18d9b51d321fd2f9abbafaac22075102ef8aaf1102cd36cc9257369a24","abstract_canon_sha256":"fa2bc6130e3f106ebda35cda844334556bddd6baa39ab5971d1b4736e4f5f948"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:54:02.005090Z","signature_b64":"OWE6E7UAWmhtvrqLbcxiFMUZodfjLgWcYjawr00s+m2Im8QhN6krVaAxFHtKl1OtKKxSiEl8olv9LyiSxTRbBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"711c7eb0de1a5c30695cebc278ecf7278ce183451ed9b91ea0cde980ff548bb4","last_reissued_at":"2026-07-05T11:54:02.004742Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:54:02.004742Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Pass@k Training for Adaptively Balancing Exploration and Exploitation of Large Reasoning Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Guang Shi, Qinghao Ye, Wayne Xin Zhao, Xiaobo Qin, Youbin Wu, Yue Ling, Zhipeng Chen","submitted_at":"2025-08-14T15:34:47Z","abstract_excerpt":"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"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.10751","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2508.10751/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2508.10751","created_at":"2026-07-05T11:54:02.004798+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.10751v1","created_at":"2026-07-05T11:54:02.004798+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.10751","created_at":"2026-07-05T11:54:02.004798+00:00"},{"alias_kind":"pith_short_12","alias_value":"OEOH5MG6DJOD","created_at":"2026-07-05T11:54:02.004798+00:00"},{"alias_kind":"pith_short_16","alias_value":"OEOH5MG6DJODA2K4","created_at":"2026-07-05T11:54:02.004798+00:00"},{"alias_kind":"pith_short_8","alias_value":"OEOH5MG6","created_at":"2026-07-05T11:54:02.004798+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":33,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26091","citing_title":"On-Policy Self-Distillation with Sampled Demonstrations Reduces Output Diversity","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2606.21090","citing_title":"Self-Improvement Can Self-Regress: The Rise-and-Collapse Failure Mode of LLM Self-Training","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2606.18910","citing_title":"REVES: REvision and VErification--Augmented Training for Test-Time Scaling","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01490","citing_title":"Don't Let Gains FADE: Breaking Down Policy Gradient Weights in RL","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2606.18487","citing_title":"SFT Overtraining Predicts Rank Inversion via Entropy Collapse Under RLVR","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2607.02390","citing_title":"DecompRL: Solving Harder Problems by Learning Modular Code Generation","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05888","citing_title":"Retry Policy Gradients in Continuous Action Spaces","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06096","citing_title":"OrderGrad: Optimizing Beyond the Mean with Order-Statistic Policy Gradient Estimation","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06080","citing_title":"On Advantage Estimates for Max@K Policy Gradients","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03608","citing_title":"Exploiting Verification-Generation Gap: Test-Time Reinforcement Learning with Confidence-Conditioned Verification","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20854","citing_title":"Finite-Time Regret Analysis of Retry-Aware Bandits","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29303","citing_title":"Entropy-KL Divergence-based Token Masking: A Novel Approach for Selective Fine-tuning of Large Language Models","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21792","citing_title":"Residual Skill Optimization for Text-to-SQL Ensembles","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20854","citing_title":"Finite-Time Regret Analysis of Retry-Aware Bandits","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20865","citing_title":"Multi-Step Likelihood-Ratio Correction for Reinforcement Learning with Verifiable Rewards","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11461","citing_title":"Breaking $\\textit{Winner-Takes-All}$: Cooperative Policy Optimization Improves Diverse LLM Reasoning","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18864","citing_title":"SAGE: Shaping Anchors for Guided Exploration in RLVR of LLMs","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19461","citing_title":"Beyond Mode Collapse: Distribution Matching for Diverse Reasoning","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2509.02547","citing_title":"The Landscape of Agentic Reinforcement Learning for LLMs: A Survey","ref_index":68,"is_internal_anchor":false},{"citing_arxiv_id":"2509.23629","citing_title":"Emergent Slow Thinking in LLMs as Inverse Tree Freezing","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2509.08827","citing_title":"A Survey of Reinforcement Learning for Large Reasoning Models","ref_index":83,"is_internal_anchor":false},{"citing_arxiv_id":"2602.06475","citing_title":"Towards Generalizable Reasoning: Group Causal Counterfactual Policy Optimization for LLM Reasoning","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11461","citing_title":"Breaking $\\textit{Winner-Takes-All}$: Cooperative Policy Optimization Improves Diverse LLM Reasoning","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10781","citing_title":"Rebellious Student: Reversing Teacher Signals for Reasoning Exploration with Self-Distilled RLVR","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08817","citing_title":"How You Begin is How You Reason: Driving Exploration in RLVR via Prefix-Tuned Priors","ref_index":7,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OEOH5MG6DJODA2K45PBHR3HXE6","json":"https://pith.science/pith/OEOH5MG6DJODA2K45PBHR3HXE6.json","graph_json":"https://pith.science/api/pith-number/OEOH5MG6DJODA2K45PBHR3HXE6/graph.json","events_json":"https://pith.science/api/pith-number/OEOH5MG6DJODA2K45PBHR3HXE6/events.json","paper":"https://pith.science/paper/OEOH5MG6"},"agent_actions":{"view_html":"https://pith.science/pith/OEOH5MG6DJODA2K45PBHR3HXE6","download_json":"https://pith.science/pith/OEOH5MG6DJODA2K45PBHR3HXE6.json","view_paper":"https://pith.science/paper/OEOH5MG6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.10751&json=true","fetch_graph":"https://pith.science/api/pith-number/OEOH5MG6DJODA2K45PBHR3HXE6/graph.json","fetch_events":"https://pith.science/api/pith-number/OEOH5MG6DJODA2K45PBHR3HXE6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OEOH5MG6DJODA2K45PBHR3HXE6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OEOH5MG6DJODA2K45PBHR3HXE6/action/storage_attestation","attest_author":"https://pith.science/pith/OEOH5MG6DJODA2K45PBHR3HXE6/action/author_attestation","sign_citation":"https://pith.science/pith/OEOH5MG6DJODA2K45PBHR3HXE6/action/citation_signature","submit_replication":"https://pith.science/pith/OEOH5MG6DJODA2K45PBHR3HXE6/action/replication_record"}},"created_at":"2026-07-05T11:54:02.004798+00:00","updated_at":"2026-07-05T11:54:02.004798+00:00"}