{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:BAOAZKMSHAEGMVLOIX4SDHVBUD","short_pith_number":"pith:BAOAZKMS","schema_version":"1.0","canonical_sha256":"081c0ca992380866556e45f9219ea1a0f44e0d78683dc9393bf4feab1943dbe3","source":{"kind":"arxiv","id":"2505.22257","version":2},"attestation_state":"computed","paper":{"title":"Revisiting Group Relative Policy Optimization: Insights into On-Policy and Off-Policy Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Apoorva Nitsure, Brian Belgodere, Jerret Ross, Jesus Rios, Jiri Navratil, Kristjan Greenewald, Mattia Rigotti, Nicolas Dupuis, Youssef Mroueh","submitted_at":"2025-05-28T11:42:33Z","abstract_excerpt":"We revisit Group Relative Policy Optimization (GRPO) in both on-policy and off-policy optimization regimes. Our motivation comes from recent work on off-policy Proximal Policy Optimization (PPO), which improves training stability, sampling efficiency, and memory usage. In addition, a recent analysis of GRPO suggests that estimating the advantage function with off-policy samples could be beneficial. Building on these observations, we adapt GRPO to the off-policy setting. We show that both on-policy and off-policy GRPO objectives yield an improvement in the reward. This result motivates the use "},"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":"2505.22257","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-28T11:42:33Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"f44620a377f93f7d887c27928c0b3760c0c042d36705abf892d2499a2bce4796","abstract_canon_sha256":"e61fd9ff5b55146dc80aaf37d96ad35ff461d9e794eb1fe108314e4789bd371b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:12:34.119993Z","signature_b64":"7WY3S/qOmk6xs39wewkvuoS6C+x/YuCfO3ygzzmoNIf5n50HFuoSR8Df7If5s60FpYe/KpWiII9UyMDmBB4LAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"081c0ca992380866556e45f9219ea1a0f44e0d78683dc9393bf4feab1943dbe3","last_reissued_at":"2026-07-05T11:12:34.119481Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:12:34.119481Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Revisiting Group Relative Policy Optimization: Insights into On-Policy and Off-Policy Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Apoorva Nitsure, Brian Belgodere, Jerret Ross, Jesus Rios, Jiri Navratil, Kristjan Greenewald, Mattia Rigotti, Nicolas Dupuis, Youssef Mroueh","submitted_at":"2025-05-28T11:42:33Z","abstract_excerpt":"We revisit Group Relative Policy Optimization (GRPO) in both on-policy and off-policy optimization regimes. Our motivation comes from recent work on off-policy Proximal Policy Optimization (PPO), which improves training stability, sampling efficiency, and memory usage. In addition, a recent analysis of GRPO suggests that estimating the advantage function with off-policy samples could be beneficial. Building on these observations, we adapt GRPO to the off-policy setting. We show that both on-policy and off-policy GRPO objectives yield an improvement in the reward. This result motivates the use "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.22257","kind":"arxiv","version":2},"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/2505.22257/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":"2505.22257","created_at":"2026-07-05T11:12:34.119550+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.22257v2","created_at":"2026-07-05T11:12:34.119550+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.22257","created_at":"2026-07-05T11:12:34.119550+00:00"},{"alias_kind":"pith_short_12","alias_value":"BAOAZKMSHAEG","created_at":"2026-07-05T11:12:34.119550+00:00"},{"alias_kind":"pith_short_16","alias_value":"BAOAZKMSHAEGMVLO","created_at":"2026-07-05T11:12:34.119550+00:00"},{"alias_kind":"pith_short_8","alias_value":"BAOAZKMS","created_at":"2026-07-05T11:12:34.119550+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":8,"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":133,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03604","citing_title":"Beyond the Literal: Decomposing Pragmatic Intent in Multimodal Meme Understanding","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30789","citing_title":"Smaller Models are Natural Explorers for Policy-Level Diversity in GRPO","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15951","citing_title":"From Failure to Feedback: Group Revision Unlocks Hard Cases in Object-Level Grounding","ref_index":55,"is_internal_anchor":false},{"citing_arxiv_id":"2510.17881","citing_title":"POPI: Personalizing LLMs via Optimized Natural Language Preference Inference","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2604.04937","citing_title":"Pramana: Fine-Tuning Large Language Models for Epistemic Reasoning through Navya-Nyaya","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12160","citing_title":"PubSwap: Public-Data Off-Policy Coordination for Federated RLVR","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06755","citing_title":"Gradient Extrapolation-Based Policy Optimization","ref_index":22,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BAOAZKMSHAEGMVLOIX4SDHVBUD","json":"https://pith.science/pith/BAOAZKMSHAEGMVLOIX4SDHVBUD.json","graph_json":"https://pith.science/api/pith-number/BAOAZKMSHAEGMVLOIX4SDHVBUD/graph.json","events_json":"https://pith.science/api/pith-number/BAOAZKMSHAEGMVLOIX4SDHVBUD/events.json","paper":"https://pith.science/paper/BAOAZKMS"},"agent_actions":{"view_html":"https://pith.science/pith/BAOAZKMSHAEGMVLOIX4SDHVBUD","download_json":"https://pith.science/pith/BAOAZKMSHAEGMVLOIX4SDHVBUD.json","view_paper":"https://pith.science/paper/BAOAZKMS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.22257&json=true","fetch_graph":"https://pith.science/api/pith-number/BAOAZKMSHAEGMVLOIX4SDHVBUD/graph.json","fetch_events":"https://pith.science/api/pith-number/BAOAZKMSHAEGMVLOIX4SDHVBUD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BAOAZKMSHAEGMVLOIX4SDHVBUD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BAOAZKMSHAEGMVLOIX4SDHVBUD/action/storage_attestation","attest_author":"https://pith.science/pith/BAOAZKMSHAEGMVLOIX4SDHVBUD/action/author_attestation","sign_citation":"https://pith.science/pith/BAOAZKMSHAEGMVLOIX4SDHVBUD/action/citation_signature","submit_replication":"https://pith.science/pith/BAOAZKMSHAEGMVLOIX4SDHVBUD/action/replication_record"}},"created_at":"2026-07-05T11:12:34.119550+00:00","updated_at":"2026-07-05T11:12:34.119550+00:00"}