{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:E5Y3OYWAZD7IYU6DHU2INONHQO","short_pith_number":"pith:E5Y3OYWA","schema_version":"1.0","canonical_sha256":"2771b762c0c8fe8c53c33d3486b9a78385fc019dffa460ceb9df535bdf8e9134","source":{"kind":"arxiv","id":"2509.23722","version":2},"attestation_state":"computed","paper":{"title":"OctoPipe: Reducing Pipeline Bubbles for Heterogeneous Models via Co-Optimizing Partitioning, Placement, and Scheduling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.DC","authors_text":"Dahua Lin, Jiaxing Li, Jihu Guo, Peng Sun, Tenghui Ma, Wei Gao, Xun Chen, Yuyang Jin, Zhisheng Ye","submitted_at":"2025-09-28T08:05:13Z","abstract_excerpt":"Pipeline parallelism is widely used to train large language models (LLMs). However, increasing heterogeneity in model architectures exacerbates pipeline bubbles, thereby reducing training efficiency. Prior approaches typically optimize a single phase of the pipeline schedule (i.e., partitioning, placement, or scheduling), leaving substantial pipeline bubbles. While promising, co-optimization poses three key challenges: (1) complex performance modeling, (2) a combinatorial search space, and (3) irregular execution orders. To address these challenges, we propose OctoPipe, a pipeline parallelism "},"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":"2509.23722","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2025-09-28T08:05:13Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"c6f96a193eee6230cba2246db56f40e6adc064557625573dbdb27866fe4f5377","abstract_canon_sha256":"6d5a116ad87cc82edd6290773e69c2a3dc2d3262e96c5fc120e8c562a13bd4c7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-07T01:15:58.658526Z","signature_b64":"Wll6fE0/dntrZDVMRjcX8/U5jaOntqN8liPPJa2idUdgP+VCuhxjUoFwMk1dgEv50mWBNx00zACItvQLM6gpCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2771b762c0c8fe8c53c33d3486b9a78385fc019dffa460ceb9df535bdf8e9134","last_reissued_at":"2026-07-07T01:15:58.657650Z","signature_status":"signed_v1","first_computed_at":"2026-07-07T01:15:58.657650Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"OctoPipe: Reducing Pipeline Bubbles for Heterogeneous Models via Co-Optimizing Partitioning, Placement, and Scheduling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.DC","authors_text":"Dahua Lin, Jiaxing Li, Jihu Guo, Peng Sun, Tenghui Ma, Wei Gao, Xun Chen, Yuyang Jin, Zhisheng Ye","submitted_at":"2025-09-28T08:05:13Z","abstract_excerpt":"Pipeline parallelism is widely used to train large language models (LLMs). However, increasing heterogeneity in model architectures exacerbates pipeline bubbles, thereby reducing training efficiency. Prior approaches typically optimize a single phase of the pipeline schedule (i.e., partitioning, placement, or scheduling), leaving substantial pipeline bubbles. While promising, co-optimization poses three key challenges: (1) complex performance modeling, (2) a combinatorial search space, and (3) irregular execution orders. To address these challenges, we propose OctoPipe, a pipeline parallelism "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.23722","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/2509.23722/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":"2509.23722","created_at":"2026-07-07T01:15:58.657739+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.23722v2","created_at":"2026-07-07T01:15:58.657739+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.23722","created_at":"2026-07-07T01:15:58.657739+00:00"},{"alias_kind":"pith_short_12","alias_value":"E5Y3OYWAZD7I","created_at":"2026-07-07T01:15:58.657739+00:00"},{"alias_kind":"pith_short_16","alias_value":"E5Y3OYWAZD7IYU6D","created_at":"2026-07-07T01:15:58.657739+00:00"},{"alias_kind":"pith_short_8","alias_value":"E5Y3OYWA","created_at":"2026-07-07T01:15:58.657739+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":4,"sample":[{"citing_arxiv_id":"2605.18750","citing_title":"A Readiness-Driven Runtime for Pipeline-Parallel Training under Runtime Variability","ref_index":17,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06374","citing_title":"ResiHP: Taming LLM Training Failures with Dynamic Hybrid Parallelism","ref_index":17,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06374","citing_title":"ResiHP: Taming LLM Training Failures with Dynamic Hybrid Parallelism","ref_index":17,"is_internal_anchor":true},{"citing_arxiv_id":"2604.06956","citing_title":"NestPipe: Large-Scale Recommendation Training on 1,500+ Accelerators via Nested Pipelining","ref_index":44,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/E5Y3OYWAZD7IYU6DHU2INONHQO","json":"https://pith.science/pith/E5Y3OYWAZD7IYU6DHU2INONHQO.json","graph_json":"https://pith.science/api/pith-number/E5Y3OYWAZD7IYU6DHU2INONHQO/graph.json","events_json":"https://pith.science/api/pith-number/E5Y3OYWAZD7IYU6DHU2INONHQO/events.json","paper":"https://pith.science/paper/E5Y3OYWA"},"agent_actions":{"view_html":"https://pith.science/pith/E5Y3OYWAZD7IYU6DHU2INONHQO","download_json":"https://pith.science/pith/E5Y3OYWAZD7IYU6DHU2INONHQO.json","view_paper":"https://pith.science/paper/E5Y3OYWA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.23722&json=true","fetch_graph":"https://pith.science/api/pith-number/E5Y3OYWAZD7IYU6DHU2INONHQO/graph.json","fetch_events":"https://pith.science/api/pith-number/E5Y3OYWAZD7IYU6DHU2INONHQO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/E5Y3OYWAZD7IYU6DHU2INONHQO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/E5Y3OYWAZD7IYU6DHU2INONHQO/action/storage_attestation","attest_author":"https://pith.science/pith/E5Y3OYWAZD7IYU6DHU2INONHQO/action/author_attestation","sign_citation":"https://pith.science/pith/E5Y3OYWAZD7IYU6DHU2INONHQO/action/citation_signature","submit_replication":"https://pith.science/pith/E5Y3OYWAZD7IYU6DHU2INONHQO/action/replication_record"}},"created_at":"2026-07-07T01:15:58.657739+00:00","updated_at":"2026-07-07T01:15:58.657739+00:00"}