{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:PMFV2DU26T6YLVUTL2P7OYUKDN","short_pith_number":"pith:PMFV2DU2","schema_version":"1.0","canonical_sha256":"7b0b5d0e9af4fd85d6935e9ff7628a1b5a4d97990d5248f4f29e68463e8d1c48","source":{"kind":"arxiv","id":"2503.10377","version":1},"attestation_state":"computed","paper":{"title":"SPPO:Efficient Long-sequence LLM Training via Adaptive Sequence Pipeline Parallel Offloading","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Peng Sun, Qiaoling Chen, Shenggui Li, Tianwei Zhang, Wei Gao, Yonggang Wen","submitted_at":"2025-03-13T13:55:22Z","abstract_excerpt":"In recent years, Large Language Models (LLMs) have exhibited remarkable capabilities, driving advancements in real-world applications. However, training LLMs on increasingly long input sequences imposes significant challenges due to high GPU memory and computational demands. Existing solutions face two key limitations: (1) memory reduction techniques, such as activation recomputation and CPU offloading, compromise training efficiency; (2) distributed parallelism strategies require excessive GPU resources, limiting the scalability of input sequence length.\n  To address these gaps, we propose Ad"},"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":"2503.10377","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2025-03-13T13:55:22Z","cross_cats_sorted":[],"title_canon_sha256":"72d26bf3debf7dfd8ef15b3d17434280d9380cf584ec9d02bc480c6ab02f6359","abstract_canon_sha256":"d52f504b882c3263e5afcff4d9b5f6c5dfb15766440a01e3b325f4457b9ea238"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:30:44.106833Z","signature_b64":"A4X1++5/R8IX1a85EJMKkJx1f/Z947q32AY5U6YK0UQTe0JmhRsqO0eQq5AZfOr27HyK/BAikMBTsOv7qF3xBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7b0b5d0e9af4fd85d6935e9ff7628a1b5a4d97990d5248f4f29e68463e8d1c48","last_reissued_at":"2026-07-05T10:30:44.106194Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:30:44.106194Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SPPO:Efficient Long-sequence LLM Training via Adaptive Sequence Pipeline Parallel Offloading","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Peng Sun, Qiaoling Chen, Shenggui Li, Tianwei Zhang, Wei Gao, Yonggang Wen","submitted_at":"2025-03-13T13:55:22Z","abstract_excerpt":"In recent years, Large Language Models (LLMs) have exhibited remarkable capabilities, driving advancements in real-world applications. However, training LLMs on increasingly long input sequences imposes significant challenges due to high GPU memory and computational demands. Existing solutions face two key limitations: (1) memory reduction techniques, such as activation recomputation and CPU offloading, compromise training efficiency; (2) distributed parallelism strategies require excessive GPU resources, limiting the scalability of input sequence length.\n  To address these gaps, we propose Ad"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.10377","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/2503.10377/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":"2503.10377","created_at":"2026-07-05T10:30:44.106266+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.10377v1","created_at":"2026-07-05T10:30:44.106266+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.10377","created_at":"2026-07-05T10:30:44.106266+00:00"},{"alias_kind":"pith_short_12","alias_value":"PMFV2DU26T6Y","created_at":"2026-07-05T10:30:44.106266+00:00"},{"alias_kind":"pith_short_16","alias_value":"PMFV2DU26T6YLVUT","created_at":"2026-07-05T10:30:44.106266+00:00"},{"alias_kind":"pith_short_8","alias_value":"PMFV2DU2","created_at":"2026-07-05T10:30:44.106266+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2509.21275","citing_title":"InfiniPipe: Elastic Pipeline Parallelism for Efficient Variable-Length Long-Context LLM Training","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07569","citing_title":"HexiSeq: Accommodating Long Context Training of LLMs over Heterogeneous Hardware","ref_index":4,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PMFV2DU26T6YLVUTL2P7OYUKDN","json":"https://pith.science/pith/PMFV2DU26T6YLVUTL2P7OYUKDN.json","graph_json":"https://pith.science/api/pith-number/PMFV2DU26T6YLVUTL2P7OYUKDN/graph.json","events_json":"https://pith.science/api/pith-number/PMFV2DU26T6YLVUTL2P7OYUKDN/events.json","paper":"https://pith.science/paper/PMFV2DU2"},"agent_actions":{"view_html":"https://pith.science/pith/PMFV2DU26T6YLVUTL2P7OYUKDN","download_json":"https://pith.science/pith/PMFV2DU26T6YLVUTL2P7OYUKDN.json","view_paper":"https://pith.science/paper/PMFV2DU2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.10377&json=true","fetch_graph":"https://pith.science/api/pith-number/PMFV2DU26T6YLVUTL2P7OYUKDN/graph.json","fetch_events":"https://pith.science/api/pith-number/PMFV2DU26T6YLVUTL2P7OYUKDN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PMFV2DU26T6YLVUTL2P7OYUKDN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PMFV2DU26T6YLVUTL2P7OYUKDN/action/storage_attestation","attest_author":"https://pith.science/pith/PMFV2DU26T6YLVUTL2P7OYUKDN/action/author_attestation","sign_citation":"https://pith.science/pith/PMFV2DU26T6YLVUTL2P7OYUKDN/action/citation_signature","submit_replication":"https://pith.science/pith/PMFV2DU26T6YLVUTL2P7OYUKDN/action/replication_record"}},"created_at":"2026-07-05T10:30:44.106266+00:00","updated_at":"2026-07-05T10:30:44.106266+00:00"}