{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:LFOP2BZD33MIH57NR33SZI4B7T","short_pith_number":"pith:LFOP2BZD","schema_version":"1.0","canonical_sha256":"595cfd0723ded883f7ed8ef72ca381fccc7f1f1b95f6ed8b39e1c9c96e2f7b00","source":{"kind":"arxiv","id":"2203.12533","version":1},"attestation_state":"computed","paper":{"title":"Pathways: Asynchronous Distributed Dataflow for ML","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.DC","authors_text":"Aakanksha Chowdhery, Brennan Saeta, Chandramohan A. Thekkath, Dan Hurt, Hyeontaek Lim, Jeff Dean, Laurent El Shafey, Michael Isard, Parker Schuh, Paul Barham, Ruoming Pang, Ryan Sepassi, Sanjay Ghemawat, Steven Hand, Sudip Roy, Yonghui Wu","submitted_at":"2022-03-23T16:50:53Z","abstract_excerpt":"We present the design of a new large scale orchestration layer for accelerators. Our system, Pathways, is explicitly designed to enable exploration of new systems and ML research ideas, while retaining state of the art performance for current models. Pathways uses a sharded dataflow graph of asynchronous operators that consume and produce futures, and efficiently gang-schedules heterogeneous parallel computations on thousands of accelerators while coordinating data transfers over their dedicated interconnects. Pathways makes use of a novel asynchronous distributed dataflow design that lets the"},"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":"2203.12533","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2022-03-23T16:50:53Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"85e5acbb128ff007431a9c2c0f21aefe93faec9bb3ae8e43825c58f48aed85ae","abstract_canon_sha256":"ec2906d00f91ccf48a455bd36a541d0fce5bbeb01505f1ad29fd59b8e4a51dec"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:08:00.561067Z","signature_b64":"oQIv2Hn5ekai7ZpV7ABzYfWSMZDi0MsKt76+8/ZuMrqqW7ZYMuwsNokDLPx+a8aJftXvpN5sF1Z7IQfv/toyAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"595cfd0723ded883f7ed8ef72ca381fccc7f1f1b95f6ed8b39e1c9c96e2f7b00","last_reissued_at":"2026-07-05T04:08:00.560654Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:08:00.560654Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Pathways: Asynchronous Distributed Dataflow for ML","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.DC","authors_text":"Aakanksha Chowdhery, Brennan Saeta, Chandramohan A. Thekkath, Dan Hurt, Hyeontaek Lim, Jeff Dean, Laurent El Shafey, Michael Isard, Parker Schuh, Paul Barham, Ruoming Pang, Ryan Sepassi, Sanjay Ghemawat, Steven Hand, Sudip Roy, Yonghui Wu","submitted_at":"2022-03-23T16:50:53Z","abstract_excerpt":"We present the design of a new large scale orchestration layer for accelerators. Our system, Pathways, is explicitly designed to enable exploration of new systems and ML research ideas, while retaining state of the art performance for current models. Pathways uses a sharded dataflow graph of asynchronous operators that consume and produce futures, and efficiently gang-schedules heterogeneous parallel computations on thousands of accelerators while coordinating data transfers over their dedicated interconnects. Pathways makes use of a novel asynchronous distributed dataflow design that lets the"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2203.12533","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/2203.12533/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":"2203.12533","created_at":"2026-07-05T04:08:00.560709+00:00"},{"alias_kind":"arxiv_version","alias_value":"2203.12533v1","created_at":"2026-07-05T04:08:00.560709+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2203.12533","created_at":"2026-07-05T04:08:00.560709+00:00"},{"alias_kind":"pith_short_12","alias_value":"LFOP2BZD33MI","created_at":"2026-07-05T04:08:00.560709+00:00"},{"alias_kind":"pith_short_16","alias_value":"LFOP2BZD33MIH57N","created_at":"2026-07-05T04:08:00.560709+00:00"},{"alias_kind":"pith_short_8","alias_value":"LFOP2BZD","created_at":"2026-07-05T04:08:00.560709+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.11169","citing_title":"Piper: A Programmable Distributed Training System","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2205.01917","citing_title":"CoCa: Contrastive Captioners are Image-Text Foundation Models","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03598","citing_title":"Unifying Dynamical Systems and Graph Theory to Mechanistically Understand Computation in Neural Networks","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2204.02311","citing_title":"PaLM: Scaling Language Modeling with Pathways","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03598","citing_title":"Unifying Dynamical Systems and Graph Theory to Mechanistically Understand Computation in Neural Networks","ref_index":36,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LFOP2BZD33MIH57NR33SZI4B7T","json":"https://pith.science/pith/LFOP2BZD33MIH57NR33SZI4B7T.json","graph_json":"https://pith.science/api/pith-number/LFOP2BZD33MIH57NR33SZI4B7T/graph.json","events_json":"https://pith.science/api/pith-number/LFOP2BZD33MIH57NR33SZI4B7T/events.json","paper":"https://pith.science/paper/LFOP2BZD"},"agent_actions":{"view_html":"https://pith.science/pith/LFOP2BZD33MIH57NR33SZI4B7T","download_json":"https://pith.science/pith/LFOP2BZD33MIH57NR33SZI4B7T.json","view_paper":"https://pith.science/paper/LFOP2BZD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2203.12533&json=true","fetch_graph":"https://pith.science/api/pith-number/LFOP2BZD33MIH57NR33SZI4B7T/graph.json","fetch_events":"https://pith.science/api/pith-number/LFOP2BZD33MIH57NR33SZI4B7T/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LFOP2BZD33MIH57NR33SZI4B7T/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LFOP2BZD33MIH57NR33SZI4B7T/action/storage_attestation","attest_author":"https://pith.science/pith/LFOP2BZD33MIH57NR33SZI4B7T/action/author_attestation","sign_citation":"https://pith.science/pith/LFOP2BZD33MIH57NR33SZI4B7T/action/citation_signature","submit_replication":"https://pith.science/pith/LFOP2BZD33MIH57NR33SZI4B7T/action/replication_record"}},"created_at":"2026-07-05T04:08:00.560709+00:00","updated_at":"2026-07-05T04:08:00.560709+00:00"}