{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:7U74AWXTCTGR5JJCZLQZQ55GLD","short_pith_number":"pith:7U74AWXT","schema_version":"1.0","canonical_sha256":"fd3fc05af314cd1ea522cae19877a658d61cc0f2cc15a8971e0fccc707130861","source":{"kind":"arxiv","id":"2601.12784","version":2},"attestation_state":"computed","paper":{"title":"StaleFlow: Staleness-Aware Data Management for Mitigating Data Skewness in Fully Disaggregated RL Post-Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Bin Cui, Fangcheng Fu, Haoyang Li, Jie Jiang, Lefeng Wang, Sheng Lin, Xiaodong Ji, Yanfeng Zhao, Yuming Zhou","submitted_at":"2026-01-19T07:31:32Z","abstract_excerpt":"Reinforcement learning (RL) post-training has become pivotal for enhancing the capabilities of modern large models. A recent trend is to develop RL systems with a fully disaggregated architecture, which decouples the three RL phases (rollout, reward, and training) onto separate resources and executes them asynchronously. However, two critical data-level concerns arise: (1) asynchronous execution leads to data staleness in trajectories (the data generated by rollout) as the model parameters used in rollout may not be up to date, which impairs RL convergence; and (2) the length variation of traj"},"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":"2601.12784","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2026-01-19T07:31:32Z","cross_cats_sorted":[],"title_canon_sha256":"26db37a4e316c63d56dba07e43d1f2ba065b2bfa99887034ce42aa577f3d2f80","abstract_canon_sha256":"e7d27e4b8048726d3a7e5403f04d961ce03faccc72225f1d7114945f6224245e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-04T02:09:17.786476Z","signature_b64":"elii+6X15a0ucHiA2CFftynhM4YSZLkbRhNuZPmB/ogy1J7PA5R/I/bLu8TIz3vrZRuOSuQbYNDJIcVibUqKDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fd3fc05af314cd1ea522cae19877a658d61cc0f2cc15a8971e0fccc707130861","last_reissued_at":"2026-08-04T02:09:17.784545Z","signature_status":"signed_v1","first_computed_at":"2026-08-04T02:09:17.784545Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"StaleFlow: Staleness-Aware Data Management for Mitigating Data Skewness in Fully Disaggregated RL Post-Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Bin Cui, Fangcheng Fu, Haoyang Li, Jie Jiang, Lefeng Wang, Sheng Lin, Xiaodong Ji, Yanfeng Zhao, Yuming Zhou","submitted_at":"2026-01-19T07:31:32Z","abstract_excerpt":"Reinforcement learning (RL) post-training has become pivotal for enhancing the capabilities of modern large models. A recent trend is to develop RL systems with a fully disaggregated architecture, which decouples the three RL phases (rollout, reward, and training) onto separate resources and executes them asynchronously. However, two critical data-level concerns arise: (1) asynchronous execution leads to data staleness in trajectories (the data generated by rollout) as the model parameters used in rollout may not be up to date, which impairs RL convergence; and (2) the length variation of traj"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2601.12784","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/2601.12784/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":"2601.12784","created_at":"2026-08-04T02:09:17.786129+00:00"},{"alias_kind":"arxiv_version","alias_value":"2601.12784v2","created_at":"2026-08-04T02:09:17.786129+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.12784","created_at":"2026-08-04T02:09:17.786129+00:00"},{"alias_kind":"pith_short_12","alias_value":"7U74AWXTCTGR","created_at":"2026-08-04T02:09:17.786129+00:00"},{"alias_kind":"pith_short_16","alias_value":"7U74AWXTCTGR5JJC","created_at":"2026-08-04T02:09:17.786129+00:00"},{"alias_kind":"pith_short_8","alias_value":"7U74AWXT","created_at":"2026-08-04T02:09:17.786129+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2606.11867","citing_title":"Harnessing Routing Foresight for Micro-step-level MoE load balancing in RL Post-training","ref_index":26,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7U74AWXTCTGR5JJCZLQZQ55GLD","json":"https://pith.science/pith/7U74AWXTCTGR5JJCZLQZQ55GLD.json","graph_json":"https://pith.science/api/pith-number/7U74AWXTCTGR5JJCZLQZQ55GLD/graph.json","events_json":"https://pith.science/api/pith-number/7U74AWXTCTGR5JJCZLQZQ55GLD/events.json","paper":"https://pith.science/paper/7U74AWXT"},"agent_actions":{"view_html":"https://pith.science/pith/7U74AWXTCTGR5JJCZLQZQ55GLD","download_json":"https://pith.science/pith/7U74AWXTCTGR5JJCZLQZQ55GLD.json","view_paper":"https://pith.science/paper/7U74AWXT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2601.12784&json=true","fetch_graph":"https://pith.science/api/pith-number/7U74AWXTCTGR5JJCZLQZQ55GLD/graph.json","fetch_events":"https://pith.science/api/pith-number/7U74AWXTCTGR5JJCZLQZQ55GLD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7U74AWXTCTGR5JJCZLQZQ55GLD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7U74AWXTCTGR5JJCZLQZQ55GLD/action/storage_attestation","attest_author":"https://pith.science/pith/7U74AWXTCTGR5JJCZLQZQ55GLD/action/author_attestation","sign_citation":"https://pith.science/pith/7U74AWXTCTGR5JJCZLQZQ55GLD/action/citation_signature","submit_replication":"https://pith.science/pith/7U74AWXTCTGR5JJCZLQZQ55GLD/action/replication_record"}},"created_at":"2026-08-04T02:09:17.786129+00:00","updated_at":"2026-08-04T02:09:17.786129+00:00"}