{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:R7RTDTYVTZOYCLBCTHRO3UHPUN","short_pith_number":"pith:R7RTDTYV","schema_version":"1.0","canonical_sha256":"8fe331cf159e5d812c2299e2edd0efa36a83973e370b18081c060e717d3eb57a","source":{"kind":"arxiv","id":"2504.06095","version":1},"attestation_state":"computed","paper":{"title":"Nonuniform-Tensor-Parallelism: Mitigating GPU failure impact for Scaled-up LLM Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.DC","authors_text":"Ahmet Inci, Ankit More, Bhargava Gopireddy, Daiyaan Arfeen, Dheevatsa Mudigere, Gregory R. Ganger","submitted_at":"2025-04-08T14:35:40Z","abstract_excerpt":"LLM training is scaled up to 10Ks of GPUs by a mix of data-(DP) and model-parallel (MP) execution. Critical to achieving efficiency is tensor-parallel (TP; a form of MP) execution within tightly-coupled subsets of GPUs, referred to as a scale-up domain, and the larger the scale-up domain the better the performance. New datacenter architectures are emerging with more GPUs able to be tightly-coupled in a scale-up domain, such as moving from 8 GPUs to 72 GPUs connected via NVLink. Unfortunately, larger scale-up domains increase the blast-radius of failures, with a failure of single GPU potentiall"},"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":"2504.06095","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2025-04-08T14:35:40Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"e295cedbd614f7b9dd799eedc6414e5a9a1a68b6c48f14aa8d0ccad571708441","abstract_canon_sha256":"219aaf5c8bbf3e116b2e942d92f9a8ee0b0c2e77c4561b5b4d543a4db2289afc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:46:15.363084Z","signature_b64":"mcIP+4IZO6Xm52YOnHVYfDSP34OToMoCxrT3GeOCgNqFIDTZmToMGCgqcG2EJcC8W+FcEnCUouawTnU2P2bfDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8fe331cf159e5d812c2299e2edd0efa36a83973e370b18081c060e717d3eb57a","last_reissued_at":"2026-07-05T10:46:15.362498Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:46:15.362498Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Nonuniform-Tensor-Parallelism: Mitigating GPU failure impact for Scaled-up LLM Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.DC","authors_text":"Ahmet Inci, Ankit More, Bhargava Gopireddy, Daiyaan Arfeen, Dheevatsa Mudigere, Gregory R. Ganger","submitted_at":"2025-04-08T14:35:40Z","abstract_excerpt":"LLM training is scaled up to 10Ks of GPUs by a mix of data-(DP) and model-parallel (MP) execution. Critical to achieving efficiency is tensor-parallel (TP; a form of MP) execution within tightly-coupled subsets of GPUs, referred to as a scale-up domain, and the larger the scale-up domain the better the performance. New datacenter architectures are emerging with more GPUs able to be tightly-coupled in a scale-up domain, such as moving from 8 GPUs to 72 GPUs connected via NVLink. Unfortunately, larger scale-up domains increase the blast-radius of failures, with a failure of single GPU potentiall"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.06095","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/2504.06095/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":"2504.06095","created_at":"2026-07-05T10:46:15.362582+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.06095v1","created_at":"2026-07-05T10:46:15.362582+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.06095","created_at":"2026-07-05T10:46:15.362582+00:00"},{"alias_kind":"pith_short_12","alias_value":"R7RTDTYVTZOY","created_at":"2026-07-05T10:46:15.362582+00:00"},{"alias_kind":"pith_short_16","alias_value":"R7RTDTYVTZOYCLBC","created_at":"2026-07-05T10:46:15.362582+00:00"},{"alias_kind":"pith_short_8","alias_value":"R7RTDTYV","created_at":"2026-07-05T10:46:15.362582+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26633","citing_title":"Simulating Unified Tensor Resharding in heterogeneous AI systems","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/R7RTDTYVTZOYCLBCTHRO3UHPUN","json":"https://pith.science/pith/R7RTDTYVTZOYCLBCTHRO3UHPUN.json","graph_json":"https://pith.science/api/pith-number/R7RTDTYVTZOYCLBCTHRO3UHPUN/graph.json","events_json":"https://pith.science/api/pith-number/R7RTDTYVTZOYCLBCTHRO3UHPUN/events.json","paper":"https://pith.science/paper/R7RTDTYV"},"agent_actions":{"view_html":"https://pith.science/pith/R7RTDTYVTZOYCLBCTHRO3UHPUN","download_json":"https://pith.science/pith/R7RTDTYVTZOYCLBCTHRO3UHPUN.json","view_paper":"https://pith.science/paper/R7RTDTYV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.06095&json=true","fetch_graph":"https://pith.science/api/pith-number/R7RTDTYVTZOYCLBCTHRO3UHPUN/graph.json","fetch_events":"https://pith.science/api/pith-number/R7RTDTYVTZOYCLBCTHRO3UHPUN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/R7RTDTYVTZOYCLBCTHRO3UHPUN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/R7RTDTYVTZOYCLBCTHRO3UHPUN/action/storage_attestation","attest_author":"https://pith.science/pith/R7RTDTYVTZOYCLBCTHRO3UHPUN/action/author_attestation","sign_citation":"https://pith.science/pith/R7RTDTYVTZOYCLBCTHRO3UHPUN/action/citation_signature","submit_replication":"https://pith.science/pith/R7RTDTYVTZOYCLBCTHRO3UHPUN/action/replication_record"}},"created_at":"2026-07-05T10:46:15.362582+00:00","updated_at":"2026-07-05T10:46:15.362582+00:00"}