{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:WIXV5TB2YWVE42FQHGLWPOP6BU","short_pith_number":"pith:WIXV5TB2","schema_version":"1.0","canonical_sha256":"b22f5ecc3ac5aa4e68b0399767b9fe0d31e7350250a3973baf4f4b1099ca1164","source":{"kind":"arxiv","id":"2509.08721","version":1},"attestation_state":"computed","paper":{"title":"Sharing is Caring: Efficient LM Post-Training with Collective RL Experience Sharing","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.MA"],"primary_cat":"cs.LG","authors_text":"Austin Virts, Ben Fielding, Christopher Nies, Diogo Ortega, Edward Phillip Flores Nu\\~no, Gabriel Passamani Andrade, Harry Grieve, Jari Kolehmainen, Jeffrey Amico, John Donaghy, Matthew J. Wright, Semih Kara, Shikhar Rastogi, Tristin Forbus, Yihua Lou","submitted_at":"2025-09-10T16:14:20Z","abstract_excerpt":"Post-training language models (LMs) with reinforcement learning (RL) can enhance their complex reasoning capabilities without supervised fine-tuning, as demonstrated by DeepSeek-R1-Zero. However, effectively utilizing RL for LMs requires significant parallelization to scale-up inference, which introduces non-trivial technical challenges (e.g. latency, memory, and reliability) alongside ever-growing financial costs. We present Swarm sAmpling Policy Optimization (SAPO), a fully decentralized and asynchronous RL post-training algorithm. SAPO is designed for decentralized networks of heterogenous "},"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":"2509.08721","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-09-10T16:14:20Z","cross_cats_sorted":["cs.MA"],"title_canon_sha256":"772544a6e7044ce0d99e7ed1c24ec4c90e5a281902e3429d646e14610af267e0","abstract_canon_sha256":"5edc3c20b7b6f0db258a292f8bbd197c2a6dcd414fef6874a5bd19ecef9fd6c4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:08:34.918454Z","signature_b64":"KHEZZYDEAhEPu5a3LuM1sAV6Qz5jMTMmhamjWGU3fK72QuU35wQcd0qIMQo9Bs6x3ovr1YcNbxJWMm/gBHMFDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b22f5ecc3ac5aa4e68b0399767b9fe0d31e7350250a3973baf4f4b1099ca1164","last_reissued_at":"2026-07-05T12:08:34.917927Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:08:34.917927Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Sharing is Caring: Efficient LM Post-Training with Collective RL Experience Sharing","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.MA"],"primary_cat":"cs.LG","authors_text":"Austin Virts, Ben Fielding, Christopher Nies, Diogo Ortega, Edward Phillip Flores Nu\\~no, Gabriel Passamani Andrade, Harry Grieve, Jari Kolehmainen, Jeffrey Amico, John Donaghy, Matthew J. Wright, Semih Kara, Shikhar Rastogi, Tristin Forbus, Yihua Lou","submitted_at":"2025-09-10T16:14:20Z","abstract_excerpt":"Post-training language models (LMs) with reinforcement learning (RL) can enhance their complex reasoning capabilities without supervised fine-tuning, as demonstrated by DeepSeek-R1-Zero. However, effectively utilizing RL for LMs requires significant parallelization to scale-up inference, which introduces non-trivial technical challenges (e.g. latency, memory, and reliability) alongside ever-growing financial costs. We present Swarm sAmpling Policy Optimization (SAPO), a fully decentralized and asynchronous RL post-training algorithm. SAPO is designed for decentralized networks of heterogenous "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.08721","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/2509.08721/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":"2509.08721","created_at":"2026-07-05T12:08:34.917987+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.08721v1","created_at":"2026-07-05T12:08:34.917987+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.08721","created_at":"2026-07-05T12:08:34.917987+00:00"},{"alias_kind":"pith_short_12","alias_value":"WIXV5TB2YWVE","created_at":"2026-07-05T12:08:34.917987+00:00"},{"alias_kind":"pith_short_16","alias_value":"WIXV5TB2YWVE42FQ","created_at":"2026-07-05T12:08:34.917987+00:00"},{"alias_kind":"pith_short_8","alias_value":"WIXV5TB2","created_at":"2026-07-05T12:08:34.917987+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.22537","citing_title":"F-TIS: Harnessing Diverse Models in Collaborative GRPO","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2604.02372","citing_title":"Backdoor Attacks on Decentralised Post-Training","ref_index":1,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WIXV5TB2YWVE42FQHGLWPOP6BU","json":"https://pith.science/pith/WIXV5TB2YWVE42FQHGLWPOP6BU.json","graph_json":"https://pith.science/api/pith-number/WIXV5TB2YWVE42FQHGLWPOP6BU/graph.json","events_json":"https://pith.science/api/pith-number/WIXV5TB2YWVE42FQHGLWPOP6BU/events.json","paper":"https://pith.science/paper/WIXV5TB2"},"agent_actions":{"view_html":"https://pith.science/pith/WIXV5TB2YWVE42FQHGLWPOP6BU","download_json":"https://pith.science/pith/WIXV5TB2YWVE42FQHGLWPOP6BU.json","view_paper":"https://pith.science/paper/WIXV5TB2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.08721&json=true","fetch_graph":"https://pith.science/api/pith-number/WIXV5TB2YWVE42FQHGLWPOP6BU/graph.json","fetch_events":"https://pith.science/api/pith-number/WIXV5TB2YWVE42FQHGLWPOP6BU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WIXV5TB2YWVE42FQHGLWPOP6BU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WIXV5TB2YWVE42FQHGLWPOP6BU/action/storage_attestation","attest_author":"https://pith.science/pith/WIXV5TB2YWVE42FQHGLWPOP6BU/action/author_attestation","sign_citation":"https://pith.science/pith/WIXV5TB2YWVE42FQHGLWPOP6BU/action/citation_signature","submit_replication":"https://pith.science/pith/WIXV5TB2YWVE42FQHGLWPOP6BU/action/replication_record"}},"created_at":"2026-07-05T12:08:34.917987+00:00","updated_at":"2026-07-05T12:08:34.917987+00:00"}