{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:VEJRWXPT5P73ZRLVROB2X2ZDUD","short_pith_number":"pith:VEJRWXPT","schema_version":"1.0","canonical_sha256":"a9131b5df3ebffbcc5758b83abeb23a0df85832f7ff3fcf9cdc2db9276729489","source":{"kind":"arxiv","id":"2409.03365","version":3},"attestation_state":"computed","paper":{"title":"Spindle: Efficient Distributed Training of Multi-Task Large Models via Wavefront Scheduling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.DC","authors_text":"Bin Cui, Fangcheng Fu, Fan Hong, Jie Zhang, Juan Zhu, Shenhan Zhu, Xupeng Miao, Yong Li, Yujie Wang","submitted_at":"2024-09-05T09:10:40Z","abstract_excerpt":"Recent foundation models are capable of handling multiple tasks and multiple data modalities with the unified base model structure and several specialized model components. However, efficient training of such multi-task (MT) multi-modal (MM) models poses significant system challenges due to the sophisticated model architecture and the heterogeneous workloads of different tasks and modalities.\n  In this paper, we propose Spindle, a brand new training system tailored for resource-efficient and high-performance training of MT MM models via wavefront scheduling. The key idea of Spindle is to decom"},"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":"2409.03365","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2024-09-05T09:10:40Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"160281d863edae3ea7d86e81de0e96b77a0057e93edd1771edcf9bfdbf2cc423","abstract_canon_sha256":"d8b2957c6e717a22ef53d7a4a440cdd7afc53e146fee09b1870dfa1adddfb015"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:12:19.540604Z","signature_b64":"dBLzvIMfJH/VzIqlV94rgVZ0JwgtqMPkE5Gqw2ePjMgpyzHPp2aVGWJJ20gA76bYoHqXJAH2Q7TdoU5M3K5lAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a9131b5df3ebffbcc5758b83abeb23a0df85832f7ff3fcf9cdc2db9276729489","last_reissued_at":"2026-07-05T10:12:19.540186Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:12:19.540186Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Spindle: Efficient Distributed Training of Multi-Task Large Models via Wavefront Scheduling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.DC","authors_text":"Bin Cui, Fangcheng Fu, Fan Hong, Jie Zhang, Juan Zhu, Shenhan Zhu, Xupeng Miao, Yong Li, Yujie Wang","submitted_at":"2024-09-05T09:10:40Z","abstract_excerpt":"Recent foundation models are capable of handling multiple tasks and multiple data modalities with the unified base model structure and several specialized model components. However, efficient training of such multi-task (MT) multi-modal (MM) models poses significant system challenges due to the sophisticated model architecture and the heterogeneous workloads of different tasks and modalities.\n  In this paper, we propose Spindle, a brand new training system tailored for resource-efficient and high-performance training of MT MM models via wavefront scheduling. The key idea of Spindle is to decom"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.03365","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/2409.03365/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":"2409.03365","created_at":"2026-07-05T10:12:19.540242+00:00"},{"alias_kind":"arxiv_version","alias_value":"2409.03365v3","created_at":"2026-07-05T10:12:19.540242+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.03365","created_at":"2026-07-05T10:12:19.540242+00:00"},{"alias_kind":"pith_short_12","alias_value":"VEJRWXPT5P73","created_at":"2026-07-05T10:12:19.540242+00:00"},{"alias_kind":"pith_short_16","alias_value":"VEJRWXPT5P73ZRLV","created_at":"2026-07-05T10:12:19.540242+00:00"},{"alias_kind":"pith_short_8","alias_value":"VEJRWXPT","created_at":"2026-07-05T10:12:19.540242+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.19163","citing_title":"Pulse: Training Acceleration for Large Diffusion Models with Automatic Pipeline Parallelism","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18710","citing_title":"Mosaic: Towards Efficient Training of Multimodal Models with Spatial Resource Multiplexing","ref_index":63,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18750","citing_title":"A Readiness-Driven Runtime for Pipeline-Parallel Training under Runtime Variability","ref_index":53,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08962","citing_title":"MegaScale-Omni: A Hyper-Scale, Workload-Resilient System for MultiModal LLM Training in Production","ref_index":54,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VEJRWXPT5P73ZRLVROB2X2ZDUD","json":"https://pith.science/pith/VEJRWXPT5P73ZRLVROB2X2ZDUD.json","graph_json":"https://pith.science/api/pith-number/VEJRWXPT5P73ZRLVROB2X2ZDUD/graph.json","events_json":"https://pith.science/api/pith-number/VEJRWXPT5P73ZRLVROB2X2ZDUD/events.json","paper":"https://pith.science/paper/VEJRWXPT"},"agent_actions":{"view_html":"https://pith.science/pith/VEJRWXPT5P73ZRLVROB2X2ZDUD","download_json":"https://pith.science/pith/VEJRWXPT5P73ZRLVROB2X2ZDUD.json","view_paper":"https://pith.science/paper/VEJRWXPT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2409.03365&json=true","fetch_graph":"https://pith.science/api/pith-number/VEJRWXPT5P73ZRLVROB2X2ZDUD/graph.json","fetch_events":"https://pith.science/api/pith-number/VEJRWXPT5P73ZRLVROB2X2ZDUD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VEJRWXPT5P73ZRLVROB2X2ZDUD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VEJRWXPT5P73ZRLVROB2X2ZDUD/action/storage_attestation","attest_author":"https://pith.science/pith/VEJRWXPT5P73ZRLVROB2X2ZDUD/action/author_attestation","sign_citation":"https://pith.science/pith/VEJRWXPT5P73ZRLVROB2X2ZDUD/action/citation_signature","submit_replication":"https://pith.science/pith/VEJRWXPT5P73ZRLVROB2X2ZDUD/action/replication_record"}},"created_at":"2026-07-05T10:12:19.540242+00:00","updated_at":"2026-07-05T10:12:19.540242+00:00"}