{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:UU4R3VYXPSBW25IVHEN5QCPQDZ","short_pith_number":"pith:UU4R3VYX","schema_version":"1.0","canonical_sha256":"a5391dd7177c836d7515391bd809f01e5a34d778f0bbe8d7aacd848d347fc71b","source":{"kind":"arxiv","id":"2307.12169","version":5},"attestation_state":"computed","paper":{"title":"Rail-only: A Low-Cost High-Performance Network for Training LLMs with Trillion Parameters","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.NI","authors_text":"Kayvon Shakeri, Manya Ghobadi, Naader Hasani, Weiyang Wang, Ying Zhang","submitted_at":"2023-07-22T21:18:41Z","abstract_excerpt":"This paper presents a low-cost network architecture for training large language models (LLMs) at hyperscale. We study the optimal parallelization strategy of LLMs and propose a novel datacenter network design tailored to LLM's unique communication pattern. We show that LLM training generates sparse communication patterns in the network and, therefore, does not require any-to-any full-bisection network to complete efficiently. As a result, our design eliminates the spine layer in traditional GPU clusters. We name this design a Rail-only network and demonstrate that it achieves the same training"},"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":"2307.12169","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.NI","submitted_at":"2023-07-22T21:18:41Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"7f9c843fa71b4fc13447df3711fe5d41cc99a14d4edf1cff369dd977ab39db92","abstract_canon_sha256":"584bba403c17106905a580cd83e6262bec449b8ad4ef3b60d99da6a05ceceb76"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:06:54.392465Z","signature_b64":"fx/GMO3q+QAgnVuZ2KOCXTEoQPmO92wGrwImks9rV5QTQt4imm7UlDGptKqqY0lNMIZUPUHiF7GFp16geyq+BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a5391dd7177c836d7515391bd809f01e5a34d778f0bbe8d7aacd848d347fc71b","last_reissued_at":"2026-07-05T09:06:54.392016Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:06:54.392016Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Rail-only: A Low-Cost High-Performance Network for Training LLMs with Trillion Parameters","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.NI","authors_text":"Kayvon Shakeri, Manya Ghobadi, Naader Hasani, Weiyang Wang, Ying Zhang","submitted_at":"2023-07-22T21:18:41Z","abstract_excerpt":"This paper presents a low-cost network architecture for training large language models (LLMs) at hyperscale. We study the optimal parallelization strategy of LLMs and propose a novel datacenter network design tailored to LLM's unique communication pattern. We show that LLM training generates sparse communication patterns in the network and, therefore, does not require any-to-any full-bisection network to complete efficiently. As a result, our design eliminates the spine layer in traditional GPU clusters. We name this design a Rail-only network and demonstrate that it achieves the same training"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2307.12169","kind":"arxiv","version":5},"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/2307.12169/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":"2307.12169","created_at":"2026-07-05T09:06:54.392070+00:00"},{"alias_kind":"arxiv_version","alias_value":"2307.12169v5","created_at":"2026-07-05T09:06:54.392070+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2307.12169","created_at":"2026-07-05T09:06:54.392070+00:00"},{"alias_kind":"pith_short_12","alias_value":"UU4R3VYXPSBW","created_at":"2026-07-05T09:06:54.392070+00:00"},{"alias_kind":"pith_short_16","alias_value":"UU4R3VYXPSBW25IV","created_at":"2026-07-05T09:06:54.392070+00:00"},{"alias_kind":"pith_short_8","alias_value":"UU4R3VYX","created_at":"2026-07-05T09:06:54.392070+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.20982","citing_title":"Diagnosing Overhead in Dispatch Operations: Cross-architecture Observatory","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2604.14690","citing_title":"Switching Efficiency: A Novel Framework for Dissecting AI Data Center Network Efficiency","ref_index":25,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UU4R3VYXPSBW25IVHEN5QCPQDZ","json":"https://pith.science/pith/UU4R3VYXPSBW25IVHEN5QCPQDZ.json","graph_json":"https://pith.science/api/pith-number/UU4R3VYXPSBW25IVHEN5QCPQDZ/graph.json","events_json":"https://pith.science/api/pith-number/UU4R3VYXPSBW25IVHEN5QCPQDZ/events.json","paper":"https://pith.science/paper/UU4R3VYX"},"agent_actions":{"view_html":"https://pith.science/pith/UU4R3VYXPSBW25IVHEN5QCPQDZ","download_json":"https://pith.science/pith/UU4R3VYXPSBW25IVHEN5QCPQDZ.json","view_paper":"https://pith.science/paper/UU4R3VYX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2307.12169&json=true","fetch_graph":"https://pith.science/api/pith-number/UU4R3VYXPSBW25IVHEN5QCPQDZ/graph.json","fetch_events":"https://pith.science/api/pith-number/UU4R3VYXPSBW25IVHEN5QCPQDZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UU4R3VYXPSBW25IVHEN5QCPQDZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UU4R3VYXPSBW25IVHEN5QCPQDZ/action/storage_attestation","attest_author":"https://pith.science/pith/UU4R3VYXPSBW25IVHEN5QCPQDZ/action/author_attestation","sign_citation":"https://pith.science/pith/UU4R3VYXPSBW25IVHEN5QCPQDZ/action/citation_signature","submit_replication":"https://pith.science/pith/UU4R3VYXPSBW25IVHEN5QCPQDZ/action/replication_record"}},"created_at":"2026-07-05T09:06:54.392070+00:00","updated_at":"2026-07-05T09:06:54.392070+00:00"}