{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:VPJ4AAWWTBCVH7Q2F3ETHTFPYI","short_pith_number":"pith:VPJ4AAWW","schema_version":"1.0","canonical_sha256":"abd3c002d6984553fe1a2ec933ccafc21f53d930c65f4f65d4f8cf97c884555f","source":{"kind":"arxiv","id":"2211.05953","version":2},"attestation_state":"computed","paper":{"title":"Breadth-First Pipeline Parallelism","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.LG"],"primary_cat":"cs.DC","authors_text":"Joel Lamy-Poirier","submitted_at":"2022-11-11T02:00:32Z","abstract_excerpt":"We introduce Breadth-First Pipeline Parallelism, a novel training schedule which optimizes the combination of pipeline and data parallelism. Breadth-First Pipeline Parallelism lowers training time, cost and memory usage by combining a high GPU utilization with a small batch size per GPU, and by making use of fully sharded data parallelism. Experimentally, we observed an increase of up to 43% in training throughput for a 52 billion-parameter model using a small batch size per GPU compared to Megatron-LM, which would reduce the training time and cost by the same amount on a large GPU cluster."},"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":"2211.05953","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.DC","submitted_at":"2022-11-11T02:00:32Z","cross_cats_sorted":["cs.AI","cs.CL","cs.LG"],"title_canon_sha256":"142a546f88497f752e5a7a9fc9cae09f8b39e2899dc10967eae120f42b8bd242","abstract_canon_sha256":"261dbe7d7774077dfa91e57ee3246b589d57ec400c46187c62133269381bbdda"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:28:32.630382Z","signature_b64":"3gdb9u59HkQpjbN2hGuZNR0RkMIG2BQ5tsSonpNmCNohe0FvK9EfMTVjIneBhTvWhnXzXzOlF6Fmg5nxVqW6Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"abd3c002d6984553fe1a2ec933ccafc21f53d930c65f4f65d4f8cf97c884555f","last_reissued_at":"2026-07-05T06:28:32.629949Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:28:32.629949Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Breadth-First Pipeline Parallelism","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.LG"],"primary_cat":"cs.DC","authors_text":"Joel Lamy-Poirier","submitted_at":"2022-11-11T02:00:32Z","abstract_excerpt":"We introduce Breadth-First Pipeline Parallelism, a novel training schedule which optimizes the combination of pipeline and data parallelism. Breadth-First Pipeline Parallelism lowers training time, cost and memory usage by combining a high GPU utilization with a small batch size per GPU, and by making use of fully sharded data parallelism. Experimentally, we observed an increase of up to 43% in training throughput for a 52 billion-parameter model using a small batch size per GPU compared to Megatron-LM, which would reduce the training time and cost by the same amount on a large GPU cluster."},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2211.05953","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/2211.05953/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":"2211.05953","created_at":"2026-07-05T06:28:32.630007+00:00"},{"alias_kind":"arxiv_version","alias_value":"2211.05953v2","created_at":"2026-07-05T06:28:32.630007+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2211.05953","created_at":"2026-07-05T06:28:32.630007+00:00"},{"alias_kind":"pith_short_12","alias_value":"VPJ4AAWWTBCV","created_at":"2026-07-05T06:28:32.630007+00:00"},{"alias_kind":"pith_short_16","alias_value":"VPJ4AAWWTBCVH7Q2","created_at":"2026-07-05T06:28:32.630007+00:00"},{"alias_kind":"pith_short_8","alias_value":"VPJ4AAWW","created_at":"2026-07-05T06:28:32.630007+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.18750","citing_title":"A Readiness-Driven Runtime for Pipeline-Parallel Training under Runtime Variability","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2604.27085","citing_title":"Efficient Training on Multiple Consumer GPUs with RoundPipe","ref_index":25,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VPJ4AAWWTBCVH7Q2F3ETHTFPYI","json":"https://pith.science/pith/VPJ4AAWWTBCVH7Q2F3ETHTFPYI.json","graph_json":"https://pith.science/api/pith-number/VPJ4AAWWTBCVH7Q2F3ETHTFPYI/graph.json","events_json":"https://pith.science/api/pith-number/VPJ4AAWWTBCVH7Q2F3ETHTFPYI/events.json","paper":"https://pith.science/paper/VPJ4AAWW"},"agent_actions":{"view_html":"https://pith.science/pith/VPJ4AAWWTBCVH7Q2F3ETHTFPYI","download_json":"https://pith.science/pith/VPJ4AAWWTBCVH7Q2F3ETHTFPYI.json","view_paper":"https://pith.science/paper/VPJ4AAWW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2211.05953&json=true","fetch_graph":"https://pith.science/api/pith-number/VPJ4AAWWTBCVH7Q2F3ETHTFPYI/graph.json","fetch_events":"https://pith.science/api/pith-number/VPJ4AAWWTBCVH7Q2F3ETHTFPYI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VPJ4AAWWTBCVH7Q2F3ETHTFPYI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VPJ4AAWWTBCVH7Q2F3ETHTFPYI/action/storage_attestation","attest_author":"https://pith.science/pith/VPJ4AAWWTBCVH7Q2F3ETHTFPYI/action/author_attestation","sign_citation":"https://pith.science/pith/VPJ4AAWWTBCVH7Q2F3ETHTFPYI/action/citation_signature","submit_replication":"https://pith.science/pith/VPJ4AAWWTBCVH7Q2F3ETHTFPYI/action/replication_record"}},"created_at":"2026-07-05T06:28:32.630007+00:00","updated_at":"2026-07-05T06:28:32.630007+00:00"}