{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:USC5KRHWSV7JODEGQFC4FQLJ6B","short_pith_number":"pith:USC5KRHW","schema_version":"1.0","canonical_sha256":"a485d544f6957e970c868145c2c169f06f0b52dfc8a5552c06e400cc51b60daa","source":{"kind":"arxiv","id":"2602.03515","version":2},"attestation_state":"computed","paper":{"title":"Mitigating Staleness in Asynchronous Pipeline Parallelism via Basis Rotation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.DC"],"primary_cat":"cs.LG","authors_text":"Hyunji Jung, Namhoon Lee, Sungbin Shin","submitted_at":"2026-02-03T13:31:51Z","abstract_excerpt":"Asynchronous pipeline parallelism maximizes hardware utilization by eliminating the pipeline bubbles inherent in synchronous execution, offering a path toward efficient large-scale distributed training. However, this efficiency gain can be compromised by gradient staleness, where the immediate model updates with delayed gradients introduce noise into the optimization process. Crucially, we identify a critical, yet often overlooked, pathology: this delay scales linearly with pipeline depth, fundamentally undermining the very scalability that the method originally intends to provide. We trace th"},"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":"2602.03515","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-03T13:31:51Z","cross_cats_sorted":["cs.AI","cs.DC"],"title_canon_sha256":"d3ccd4f64e41154cf9a4a37fc682ae8ac3e8df2cdbc146dcd5b1ee7dc2a35cf2","abstract_canon_sha256":"eb320a042c49836c2252b557d9c9e5ad09c0b36c2e9a751eb725e048f123381c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:05:11.737529Z","signature_b64":"HUIRwLzILml1HEEA4zKaJ/xQFulZ1zf2uwj05DEWD7P8BlibVo3PdNZyo60NSTGVy1hP20J7nz/fkroBAirICA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a485d544f6957e970c868145c2c169f06f0b52dfc8a5552c06e400cc51b60daa","last_reissued_at":"2026-05-28T01:05:11.737045Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:05:11.737045Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mitigating Staleness in Asynchronous Pipeline Parallelism via Basis Rotation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.DC"],"primary_cat":"cs.LG","authors_text":"Hyunji Jung, Namhoon Lee, Sungbin Shin","submitted_at":"2026-02-03T13:31:51Z","abstract_excerpt":"Asynchronous pipeline parallelism maximizes hardware utilization by eliminating the pipeline bubbles inherent in synchronous execution, offering a path toward efficient large-scale distributed training. However, this efficiency gain can be compromised by gradient staleness, where the immediate model updates with delayed gradients introduce noise into the optimization process. Crucially, we identify a critical, yet often overlooked, pathology: this delay scales linearly with pipeline depth, fundamentally undermining the very scalability that the method originally intends to provide. We trace th"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.03515","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/2602.03515/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":"2602.03515","created_at":"2026-05-28T01:05:11.737107+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.03515v2","created_at":"2026-05-28T01:05:11.737107+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.03515","created_at":"2026-05-28T01:05:11.737107+00:00"},{"alias_kind":"pith_short_12","alias_value":"USC5KRHWSV7J","created_at":"2026-05-28T01:05:11.737107+00:00"},{"alias_kind":"pith_short_16","alias_value":"USC5KRHWSV7JODEG","created_at":"2026-05-28T01:05:11.737107+00:00"},{"alias_kind":"pith_short_8","alias_value":"USC5KRHW","created_at":"2026-05-28T01:05:11.737107+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2605.13434","citing_title":"Rescaled Asynchronous SGD: Optimal Distributed Optimization under Data and System Heterogeneity","ref_index":110,"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":46,"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":47,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/USC5KRHWSV7JODEGQFC4FQLJ6B","json":"https://pith.science/pith/USC5KRHWSV7JODEGQFC4FQLJ6B.json","graph_json":"https://pith.science/api/pith-number/USC5KRHWSV7JODEGQFC4FQLJ6B/graph.json","events_json":"https://pith.science/api/pith-number/USC5KRHWSV7JODEGQFC4FQLJ6B/events.json","paper":"https://pith.science/paper/USC5KRHW"},"agent_actions":{"view_html":"https://pith.science/pith/USC5KRHWSV7JODEGQFC4FQLJ6B","download_json":"https://pith.science/pith/USC5KRHWSV7JODEGQFC4FQLJ6B.json","view_paper":"https://pith.science/paper/USC5KRHW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.03515&json=true","fetch_graph":"https://pith.science/api/pith-number/USC5KRHWSV7JODEGQFC4FQLJ6B/graph.json","fetch_events":"https://pith.science/api/pith-number/USC5KRHWSV7JODEGQFC4FQLJ6B/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/USC5KRHWSV7JODEGQFC4FQLJ6B/action/timestamp_anchor","attest_storage":"https://pith.science/pith/USC5KRHWSV7JODEGQFC4FQLJ6B/action/storage_attestation","attest_author":"https://pith.science/pith/USC5KRHWSV7JODEGQFC4FQLJ6B/action/author_attestation","sign_citation":"https://pith.science/pith/USC5KRHWSV7JODEGQFC4FQLJ6B/action/citation_signature","submit_replication":"https://pith.science/pith/USC5KRHWSV7JODEGQFC4FQLJ6B/action/replication_record"}},"created_at":"2026-05-28T01:05:11.737107+00:00","updated_at":"2026-05-28T01:05:11.737107+00:00"}