{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:JLLBNGL3I6PJLIUNZPQFZUDJ5W","short_pith_number":"pith:JLLBNGL3","schema_version":"1.0","canonical_sha256":"4ad616997b479e95a28dcbe05cd069ed9559cad25392cff5bc85b760ed6d6792","source":{"kind":"arxiv","id":"2508.05387","version":3},"attestation_state":"computed","paper":{"title":"Echo: Decoupling Inference and Training for Large-Scale RL Alignment on Heterogeneous Swarms","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alfred Long, Changyuan Fan, Eric Yang, Jie Xiao, Lynn Ai, Qingnan Ren, Rymon Yu, Shaoduo Gan, Yuchen Zhang","submitted_at":"2025-08-07T13:37:04Z","abstract_excerpt":"Modern RL-based post-training for large language models (LLMs) co-locate trajectory sampling and policy optimisation on the same GPU cluster, forcing the system to switch between inference and training workloads. This serial context switching violates the single-program-multiple-data (SPMD) assumption underlying today's distributed training systems. We present Echo, the RL system that cleanly decouples these two phases across heterogeneous \"inference\" and \"training\" swarms while preserving statistical efficiency. Echo introduces two lightweight synchronization protocols: a sequential pull mode"},"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":"2508.05387","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2025-08-07T13:37:04Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"18716e606101a0b791c44d4607f446857833e4ae8fce2ac70d0fc78c26cb7daf","abstract_canon_sha256":"489e2138227851a42101a028e905d75abe212c6ed2d35ae79ffe56aceeed0cb2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:52:23.489077Z","signature_b64":"KyAgBf2yVqcMcIpJs/sIKRu7pSzZEt0ggjSsFdXqRzZQztZkReQuTOqJ+UGVXGsergIVUU7UCuZXxdL96w8kBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4ad616997b479e95a28dcbe05cd069ed9559cad25392cff5bc85b760ed6d6792","last_reissued_at":"2026-07-05T11:52:23.488493Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:52:23.488493Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Echo: Decoupling Inference and Training for Large-Scale RL Alignment on Heterogeneous Swarms","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alfred Long, Changyuan Fan, Eric Yang, Jie Xiao, Lynn Ai, Qingnan Ren, Rymon Yu, Shaoduo Gan, Yuchen Zhang","submitted_at":"2025-08-07T13:37:04Z","abstract_excerpt":"Modern RL-based post-training for large language models (LLMs) co-locate trajectory sampling and policy optimisation on the same GPU cluster, forcing the system to switch between inference and training workloads. This serial context switching violates the single-program-multiple-data (SPMD) assumption underlying today's distributed training systems. We present Echo, the RL system that cleanly decouples these two phases across heterogeneous \"inference\" and \"training\" swarms while preserving statistical efficiency. Echo introduces two lightweight synchronization protocols: a sequential pull mode"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.05387","kind":"arxiv","version":3},"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/2508.05387/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":"2508.05387","created_at":"2026-07-05T11:52:23.488572+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.05387v3","created_at":"2026-07-05T11:52:23.488572+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.05387","created_at":"2026-07-05T11:52:23.488572+00:00"},{"alias_kind":"pith_short_12","alias_value":"JLLBNGL3I6PJ","created_at":"2026-07-05T11:52:23.488572+00:00"},{"alias_kind":"pith_short_16","alias_value":"JLLBNGL3I6PJLIUN","created_at":"2026-07-05T11:52:23.488572+00:00"},{"alias_kind":"pith_short_8","alias_value":"JLLBNGL3","created_at":"2026-07-05T11:52:23.488572+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/JLLBNGL3I6PJLIUNZPQFZUDJ5W","json":"https://pith.science/pith/JLLBNGL3I6PJLIUNZPQFZUDJ5W.json","graph_json":"https://pith.science/api/pith-number/JLLBNGL3I6PJLIUNZPQFZUDJ5W/graph.json","events_json":"https://pith.science/api/pith-number/JLLBNGL3I6PJLIUNZPQFZUDJ5W/events.json","paper":"https://pith.science/paper/JLLBNGL3"},"agent_actions":{"view_html":"https://pith.science/pith/JLLBNGL3I6PJLIUNZPQFZUDJ5W","download_json":"https://pith.science/pith/JLLBNGL3I6PJLIUNZPQFZUDJ5W.json","view_paper":"https://pith.science/paper/JLLBNGL3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.05387&json=true","fetch_graph":"https://pith.science/api/pith-number/JLLBNGL3I6PJLIUNZPQFZUDJ5W/graph.json","fetch_events":"https://pith.science/api/pith-number/JLLBNGL3I6PJLIUNZPQFZUDJ5W/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JLLBNGL3I6PJLIUNZPQFZUDJ5W/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JLLBNGL3I6PJLIUNZPQFZUDJ5W/action/storage_attestation","attest_author":"https://pith.science/pith/JLLBNGL3I6PJLIUNZPQFZUDJ5W/action/author_attestation","sign_citation":"https://pith.science/pith/JLLBNGL3I6PJLIUNZPQFZUDJ5W/action/citation_signature","submit_replication":"https://pith.science/pith/JLLBNGL3I6PJLIUNZPQFZUDJ5W/action/replication_record"}},"created_at":"2026-07-05T11:52:23.488572+00:00","updated_at":"2026-07-05T11:52:23.488572+00:00"}