{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:FDDXSKF6SAA3EEKD3RVOOZHJEP","short_pith_number":"pith:FDDXSKF6","schema_version":"1.0","canonical_sha256":"28c77928be9001b21143dc6ae764e923eaaf67fc6c9ecaca73c562ac467e0d13","source":{"kind":"arxiv","id":"2506.09340","version":1},"attestation_state":"computed","paper":{"title":"RePO: Replay-Enhanced Policy Optimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chaochao Lu, Chao Yang, Siheng Li, Wai Lam, Zhanhui Zhou","submitted_at":"2025-06-11T02:44:10Z","abstract_excerpt":"Reinforcement learning (RL) is vital for optimizing large language models (LLMs). Recent Group Relative Policy Optimization (GRPO) estimates advantages using multiple on-policy outputs per prompt, leading to high computational costs and low data efficiency. To address this, we introduce Replay-Enhanced Policy Optimization (RePO), which leverages diverse replay strategies to retrieve off-policy samples from a replay buffer, allowing policy optimization based on a broader and more diverse set of samples for each prompt. Experiments on five LLMs across seven mathematical reasoning benchmarks demo"},"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":"2506.09340","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-11T02:44:10Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"935258b1e806958e8c467395dcf5c9001b51e807ae7a4367b84f2d60b0a7c17b","abstract_canon_sha256":"ae23b7e43999a8cf40ef16971c81022045485c4d134871e69566ede0e1083866"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:19:35.059252Z","signature_b64":"wymrpj4do9yIQlZxgLmD/3UP1LpsesamkmmQQPTzMroza/6B4qX8aIwzcdHYs/NcnDdS9bq9PLayHg1JfuT/Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"28c77928be9001b21143dc6ae764e923eaaf67fc6c9ecaca73c562ac467e0d13","last_reissued_at":"2026-07-05T11:19:35.058788Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:19:35.058788Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RePO: Replay-Enhanced Policy Optimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chaochao Lu, Chao Yang, Siheng Li, Wai Lam, Zhanhui Zhou","submitted_at":"2025-06-11T02:44:10Z","abstract_excerpt":"Reinforcement learning (RL) is vital for optimizing large language models (LLMs). Recent Group Relative Policy Optimization (GRPO) estimates advantages using multiple on-policy outputs per prompt, leading to high computational costs and low data efficiency. To address this, we introduce Replay-Enhanced Policy Optimization (RePO), which leverages diverse replay strategies to retrieve off-policy samples from a replay buffer, allowing policy optimization based on a broader and more diverse set of samples for each prompt. Experiments on five LLMs across seven mathematical reasoning benchmarks demo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.09340","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/2506.09340/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":"2506.09340","created_at":"2026-07-05T11:19:35.058844+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.09340v1","created_at":"2026-07-05T11:19:35.058844+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.09340","created_at":"2026-07-05T11:19:35.058844+00:00"},{"alias_kind":"pith_short_12","alias_value":"FDDXSKF6SAA3","created_at":"2026-07-05T11:19:35.058844+00:00"},{"alias_kind":"pith_short_16","alias_value":"FDDXSKF6SAA3EEKD","created_at":"2026-07-05T11:19:35.058844+00:00"},{"alias_kind":"pith_short_8","alias_value":"FDDXSKF6","created_at":"2026-07-05T11:19:35.058844+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":12,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21943","citing_title":"Modularized Reinforcement Learning on LLMs: From MDP Creation to Exploration and Learning","ref_index":105,"is_internal_anchor":false},{"citing_arxiv_id":"2606.18216","citing_title":"Zone of Proximal Policy Optimization: Teacher in Prompts, Not Gradients","ref_index":89,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04560","citing_title":"Rollout-Level Advantage-Prioritized Experience Replay for GRPO","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03087","citing_title":"Learning to Solve, Forgetting to Retain: Correct-Set Turnover in RLVR","ref_index":114,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01281","citing_title":"RLVR without Ineffective Samples: Group Prioritized Off-Policy Optimization for LLM Reasoning","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30420","citing_title":"Experience Augmented Policy Optimization for LLM Reasoning","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2604.28005","citing_title":"Kernelized Advantage Estimation: From Nonparametric Statistics to LLM Reasoning","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.04142","citing_title":"OP-GRPO: Efficient Off-Policy GRPO for Flow-Matching Models","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2604.28005","citing_title":"Kernelized Advantage Estimation: From Nonparametric Statistics to LLM Reasoning","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07941","citing_title":"Large Language Model Post-Training: A Unified View of Off-Policy and On-Policy Learning","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02913","citing_title":"Generate, Filter, Control, Replay: A Comprehensive Survey of Rollout Strategies for LLM Reinforcement Learning","ref_index":68,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20733","citing_title":"Near-Future Policy Optimization","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FDDXSKF6SAA3EEKD3RVOOZHJEP","json":"https://pith.science/pith/FDDXSKF6SAA3EEKD3RVOOZHJEP.json","graph_json":"https://pith.science/api/pith-number/FDDXSKF6SAA3EEKD3RVOOZHJEP/graph.json","events_json":"https://pith.science/api/pith-number/FDDXSKF6SAA3EEKD3RVOOZHJEP/events.json","paper":"https://pith.science/paper/FDDXSKF6"},"agent_actions":{"view_html":"https://pith.science/pith/FDDXSKF6SAA3EEKD3RVOOZHJEP","download_json":"https://pith.science/pith/FDDXSKF6SAA3EEKD3RVOOZHJEP.json","view_paper":"https://pith.science/paper/FDDXSKF6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.09340&json=true","fetch_graph":"https://pith.science/api/pith-number/FDDXSKF6SAA3EEKD3RVOOZHJEP/graph.json","fetch_events":"https://pith.science/api/pith-number/FDDXSKF6SAA3EEKD3RVOOZHJEP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FDDXSKF6SAA3EEKD3RVOOZHJEP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FDDXSKF6SAA3EEKD3RVOOZHJEP/action/storage_attestation","attest_author":"https://pith.science/pith/FDDXSKF6SAA3EEKD3RVOOZHJEP/action/author_attestation","sign_citation":"https://pith.science/pith/FDDXSKF6SAA3EEKD3RVOOZHJEP/action/citation_signature","submit_replication":"https://pith.science/pith/FDDXSKF6SAA3EEKD3RVOOZHJEP/action/replication_record"}},"created_at":"2026-07-05T11:19:35.058844+00:00","updated_at":"2026-07-05T11:19:35.058844+00:00"}