{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:SO4VX2BG5HVLKWUQYK6RYAE6BY","short_pith_number":"pith:SO4VX2BG","schema_version":"1.0","canonical_sha256":"93b95be826e9eab55a90c2bd1c009e0e3a3f6c589ad1fd17792e0e2f54cd3637","source":{"kind":"arxiv","id":"2506.21495","version":1},"attestation_state":"computed","paper":{"title":"Bridging Offline and Online Reinforcement Learning for LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Angelica Chen, Ilia Kulikov, Jack Lanchantin, Janice Lan, Jason E Weston, Jing Xu, Ping Yu, Sainbayar Sukhbaatar, Swarnadeep Saha, Tianlu Wang, Weizhe Yuan, Xian Li","submitted_at":"2025-06-26T17:25:49Z","abstract_excerpt":"We investigate the effectiveness of reinforcement learning methods for finetuning large language models when transitioning from offline to semi-online to fully online regimes for both verifiable and non-verifiable tasks. Our experiments cover training on verifiable math as well as non-verifiable instruction following with a set of benchmark evaluations for both. Across these settings, we extensively compare online and semi-online Direct Preference Optimization and Group Reward Policy Optimization objectives, and surprisingly find similar performance and convergence between these variants, whic"},"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":"2506.21495","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-06-26T17:25:49Z","cross_cats_sorted":[],"title_canon_sha256":"3455f0f4217bb737f3663ed1c5ab44d5a0812b7a2d48babf48d4523fe5f24973","abstract_canon_sha256":"98c352dfcdf8386948fca0ff7a9a10f44fe6cc0544d07a6e70c2f6f2fcd70865"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:27:46.924920Z","signature_b64":"xHolkQpIGY93P5zljpHIWeIlJS9Cmnq3CEf1Qqw/eAT5NbUdzGNLvHoIHHjlP20FG73DAypaISb+A8VygknqBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"93b95be826e9eab55a90c2bd1c009e0e3a3f6c589ad1fd17792e0e2f54cd3637","last_reissued_at":"2026-07-05T11:27:46.924461Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:27:46.924461Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Bridging Offline and Online Reinforcement Learning for LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Angelica Chen, Ilia Kulikov, Jack Lanchantin, Janice Lan, Jason E Weston, Jing Xu, Ping Yu, Sainbayar Sukhbaatar, Swarnadeep Saha, Tianlu Wang, Weizhe Yuan, Xian Li","submitted_at":"2025-06-26T17:25:49Z","abstract_excerpt":"We investigate the effectiveness of reinforcement learning methods for finetuning large language models when transitioning from offline to semi-online to fully online regimes for both verifiable and non-verifiable tasks. Our experiments cover training on verifiable math as well as non-verifiable instruction following with a set of benchmark evaluations for both. Across these settings, we extensively compare online and semi-online Direct Preference Optimization and Group Reward Policy Optimization objectives, and surprisingly find similar performance and convergence between these variants, whic"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.21495","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/2506.21495/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":"2506.21495","created_at":"2026-07-05T11:27:46.924524+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.21495v1","created_at":"2026-07-05T11:27:46.924524+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.21495","created_at":"2026-07-05T11:27:46.924524+00:00"},{"alias_kind":"pith_short_12","alias_value":"SO4VX2BG5HVL","created_at":"2026-07-05T11:27:46.924524+00:00"},{"alias_kind":"pith_short_16","alias_value":"SO4VX2BG5HVLKWUQ","created_at":"2026-07-05T11:27:46.924524+00:00"},{"alias_kind":"pith_short_8","alias_value":"SO4VX2BG","created_at":"2026-07-05T11:27:46.924524+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":9,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25996","citing_title":"Autodata: An agentic data scientist to create high quality synthetic data","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2606.25996","citing_title":"Autodata: An agentic data scientist to create high quality synthetic data","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2606.21943","citing_title":"Modularized Reinforcement Learning on LLMs: From MDP Creation to Exploration and Learning","ref_index":97,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05464","citing_title":"Step-by-Step Optimization-like Reasoning in LLMs over Expanding Search Spaces","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03698","citing_title":"Multi$^2$: Hierarchical Multi-Agent Decision-Making with LLM-Based Agents in Interactive Environments","ref_index":72,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30642","citing_title":"LeVo 2: Stable and Melodious Song Generation via Hierarchical Representation Modeling and Progressive Post-Training","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30789","citing_title":"Smaller Models are Natural Explorers for Policy-Level Diversity in GRPO","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2509.21267","citing_title":"Task-Dependent Evaluation of LLM Output Homogenization: A Taxonomy-Guided Framework","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18530","citing_title":"OGER: A Robust Offline-Guided Exploration Reward for Hybrid Reinforcement Learning","ref_index":13,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SO4VX2BG5HVLKWUQYK6RYAE6BY","json":"https://pith.science/pith/SO4VX2BG5HVLKWUQYK6RYAE6BY.json","graph_json":"https://pith.science/api/pith-number/SO4VX2BG5HVLKWUQYK6RYAE6BY/graph.json","events_json":"https://pith.science/api/pith-number/SO4VX2BG5HVLKWUQYK6RYAE6BY/events.json","paper":"https://pith.science/paper/SO4VX2BG"},"agent_actions":{"view_html":"https://pith.science/pith/SO4VX2BG5HVLKWUQYK6RYAE6BY","download_json":"https://pith.science/pith/SO4VX2BG5HVLKWUQYK6RYAE6BY.json","view_paper":"https://pith.science/paper/SO4VX2BG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.21495&json=true","fetch_graph":"https://pith.science/api/pith-number/SO4VX2BG5HVLKWUQYK6RYAE6BY/graph.json","fetch_events":"https://pith.science/api/pith-number/SO4VX2BG5HVLKWUQYK6RYAE6BY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SO4VX2BG5HVLKWUQYK6RYAE6BY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SO4VX2BG5HVLKWUQYK6RYAE6BY/action/storage_attestation","attest_author":"https://pith.science/pith/SO4VX2BG5HVLKWUQYK6RYAE6BY/action/author_attestation","sign_citation":"https://pith.science/pith/SO4VX2BG5HVLKWUQYK6RYAE6BY/action/citation_signature","submit_replication":"https://pith.science/pith/SO4VX2BG5HVLKWUQYK6RYAE6BY/action/replication_record"}},"created_at":"2026-07-05T11:27:46.924524+00:00","updated_at":"2026-07-05T11:27:46.924524+00:00"}