{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:4GCOYZFZKY4JLXFUMWNYUPBFWN","short_pith_number":"pith:4GCOYZFZ","schema_version":"1.0","canonical_sha256":"e184ec64b9563895dcb4659b8a3c25b35e1bdb80e2bc3882843768df484ec9a2","source":{"kind":"arxiv","id":"2605.31455","version":1},"attestation_state":"computed","paper":{"title":"DRIFT: Decoupled Rollouts and Importance-Weighted Fine-Tuning for Efficient Multi-Turn Optimization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Chengwei Qin, Jian Mu, Tianyi Lin, Yao Shu, Zhongxiang Dai","submitted_at":"2026-05-29T15:49:13Z","abstract_excerpt":"Large language models are increasingly deployed in multi-turn interactive settings where users or environments can iteratively provide lightweight feedback. Unfortunately, optimizing such behavior presents a sharp dilemma in practice: online reinforcement learning is able to effectively address multi-turn dynamics but is prohibitively expensive due to the cost of generating full correction trajectories at every update, whereas offline supervised fine-tuning (SFT) is efficient but suffers from distribution shift and behavioral collapse. To this end, we novelly propose DRIFT (Decoupled Rollouts "},"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":"2605.31455","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-29T15:49:13Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"b2d63acda36d4a0cc8c3f5ad55d71f135467ab38106a1da22b98f7cffec9b052","abstract_canon_sha256":"a8f397e96a0c0d3d8285a62b1443bff8ac93e54d2daf9ee1ad6f8a26c51bcea0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-01T02:04:06.752471Z","signature_b64":"PDxPG5G91rLNuVZUJwwLFy3WObxZDhl/QlZPpekOWhmOfGN0fWsVelnFmd16fYy38rmcD0z6CqSxJ1GV699iDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e184ec64b9563895dcb4659b8a3c25b35e1bdb80e2bc3882843768df484ec9a2","last_reissued_at":"2026-06-01T02:04:06.751600Z","signature_status":"signed_v1","first_computed_at":"2026-06-01T02:04:06.751600Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DRIFT: Decoupled Rollouts and Importance-Weighted Fine-Tuning for Efficient Multi-Turn Optimization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Chengwei Qin, Jian Mu, Tianyi Lin, Yao Shu, Zhongxiang Dai","submitted_at":"2026-05-29T15:49:13Z","abstract_excerpt":"Large language models are increasingly deployed in multi-turn interactive settings where users or environments can iteratively provide lightweight feedback. Unfortunately, optimizing such behavior presents a sharp dilemma in practice: online reinforcement learning is able to effectively address multi-turn dynamics but is prohibitively expensive due to the cost of generating full correction trajectories at every update, whereas offline supervised fine-tuning (SFT) is efficient but suffers from distribution shift and behavioral collapse. To this end, we novelly propose DRIFT (Decoupled Rollouts "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.31455","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/2605.31455/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":"2605.31455","created_at":"2026-06-01T02:04:06.751748+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.31455v1","created_at":"2026-06-01T02:04:06.751748+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.31455","created_at":"2026-06-01T02:04:06.751748+00:00"},{"alias_kind":"pith_short_12","alias_value":"4GCOYZFZKY4J","created_at":"2026-06-01T02:04:06.751748+00:00"},{"alias_kind":"pith_short_16","alias_value":"4GCOYZFZKY4JLXFU","created_at":"2026-06-01T02:04:06.751748+00:00"},{"alias_kind":"pith_short_8","alias_value":"4GCOYZFZ","created_at":"2026-06-01T02:04:06.751748+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4GCOYZFZKY4JLXFUMWNYUPBFWN","json":"https://pith.science/pith/4GCOYZFZKY4JLXFUMWNYUPBFWN.json","graph_json":"https://pith.science/api/pith-number/4GCOYZFZKY4JLXFUMWNYUPBFWN/graph.json","events_json":"https://pith.science/api/pith-number/4GCOYZFZKY4JLXFUMWNYUPBFWN/events.json","paper":"https://pith.science/paper/4GCOYZFZ"},"agent_actions":{"view_html":"https://pith.science/pith/4GCOYZFZKY4JLXFUMWNYUPBFWN","download_json":"https://pith.science/pith/4GCOYZFZKY4JLXFUMWNYUPBFWN.json","view_paper":"https://pith.science/paper/4GCOYZFZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.31455&json=true","fetch_graph":"https://pith.science/api/pith-number/4GCOYZFZKY4JLXFUMWNYUPBFWN/graph.json","fetch_events":"https://pith.science/api/pith-number/4GCOYZFZKY4JLXFUMWNYUPBFWN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4GCOYZFZKY4JLXFUMWNYUPBFWN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4GCOYZFZKY4JLXFUMWNYUPBFWN/action/storage_attestation","attest_author":"https://pith.science/pith/4GCOYZFZKY4JLXFUMWNYUPBFWN/action/author_attestation","sign_citation":"https://pith.science/pith/4GCOYZFZKY4JLXFUMWNYUPBFWN/action/citation_signature","submit_replication":"https://pith.science/pith/4GCOYZFZKY4JLXFUMWNYUPBFWN/action/replication_record"}},"created_at":"2026-06-01T02:04:06.751748+00:00","updated_at":"2026-06-01T02:04:06.751748+00:00"}