{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:S4JG3NXFAOHNKFN5DJT4SC5AZP","short_pith_number":"pith:S4JG3NXF","schema_version":"1.0","canonical_sha256":"97126db6e5038ed515bd1a67c90ba0cbd225b4206e9a715d649b73ca7ddc74d0","source":{"kind":"arxiv","id":"2504.20490","version":2},"attestation_state":"computed","paper":{"title":"Hetu v2: A General and Scalable Deep Learning System with Hierarchical and Heterogeneous Single Program Multiple Data Annotations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Bin Cui, Fangcheng Fu, Hao Ge, Haoyang Li, Jiawen Niu, Sheng Lin, Xuanyu Wang, Xupeng Miao, Yuming Zhou","submitted_at":"2025-04-29T07:27:54Z","abstract_excerpt":"The Single-Program Multiple-Data (SPMD) paradigm provides a unified abstraction to annotate various parallel dimensions in distributed deep learning (DL) training. With SPMD, users can write training programs from the viewpoint of a single device, and the system will automatically deduce the tensor sharding and communication patterns. However, with the recent development in large-scale DL models, distributed training exhibits spatial and temporal workload heterogeneity, arising from both device disparities (e.g., mixed hardware, failures) and data variations (e.g., uneven sequence lengths). Su"},"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":"2504.20490","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2025-04-29T07:27:54Z","cross_cats_sorted":[],"title_canon_sha256":"3a118d13ff641bc1ef74ba5d0426175e6545dbd4d903f68bcf12adf3fcc12b6d","abstract_canon_sha256":"dac55a321eb75d9cc9f4746a327ae8bc565c6eb6c802e6bf93bcb93244aea2c1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-02T01:18:02.760352Z","signature_b64":"vt44zBAVv4XqwQvhzjHreDpz+e/GyCmvIHj2eR21j4RUhqual+c4RN+dMSHXYXl5Ak8gmxDidHXfLc+YhEuxAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"97126db6e5038ed515bd1a67c90ba0cbd225b4206e9a715d649b73ca7ddc74d0","last_reissued_at":"2026-07-02T01:18:02.759853Z","signature_status":"signed_v1","first_computed_at":"2026-07-02T01:18:02.759853Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Hetu v2: A General and Scalable Deep Learning System with Hierarchical and Heterogeneous Single Program Multiple Data Annotations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Bin Cui, Fangcheng Fu, Hao Ge, Haoyang Li, Jiawen Niu, Sheng Lin, Xuanyu Wang, Xupeng Miao, Yuming Zhou","submitted_at":"2025-04-29T07:27:54Z","abstract_excerpt":"The Single-Program Multiple-Data (SPMD) paradigm provides a unified abstraction to annotate various parallel dimensions in distributed deep learning (DL) training. With SPMD, users can write training programs from the viewpoint of a single device, and the system will automatically deduce the tensor sharding and communication patterns. However, with the recent development in large-scale DL models, distributed training exhibits spatial and temporal workload heterogeneity, arising from both device disparities (e.g., mixed hardware, failures) and data variations (e.g., uneven sequence lengths). Su"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.20490","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/2504.20490/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":"2504.20490","created_at":"2026-07-02T01:18:02.759915+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.20490v2","created_at":"2026-07-02T01:18:02.759915+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.20490","created_at":"2026-07-02T01:18:02.759915+00:00"},{"alias_kind":"pith_short_12","alias_value":"S4JG3NXFAOHN","created_at":"2026-07-02T01:18:02.759915+00:00"},{"alias_kind":"pith_short_16","alias_value":"S4JG3NXFAOHNKFN5","created_at":"2026-07-02T01:18:02.759915+00:00"},{"alias_kind":"pith_short_8","alias_value":"S4JG3NXF","created_at":"2026-07-02T01:18:02.759915+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2509.24859","citing_title":"HARP: Orchestrating Automated Parallel Training on Heterogeneous GPU Clusters","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07569","citing_title":"HexiSeq: Accommodating Long Context Training of LLMs over Heterogeneous Hardware","ref_index":29,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/S4JG3NXFAOHNKFN5DJT4SC5AZP","json":"https://pith.science/pith/S4JG3NXFAOHNKFN5DJT4SC5AZP.json","graph_json":"https://pith.science/api/pith-number/S4JG3NXFAOHNKFN5DJT4SC5AZP/graph.json","events_json":"https://pith.science/api/pith-number/S4JG3NXFAOHNKFN5DJT4SC5AZP/events.json","paper":"https://pith.science/paper/S4JG3NXF"},"agent_actions":{"view_html":"https://pith.science/pith/S4JG3NXFAOHNKFN5DJT4SC5AZP","download_json":"https://pith.science/pith/S4JG3NXFAOHNKFN5DJT4SC5AZP.json","view_paper":"https://pith.science/paper/S4JG3NXF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.20490&json=true","fetch_graph":"https://pith.science/api/pith-number/S4JG3NXFAOHNKFN5DJT4SC5AZP/graph.json","fetch_events":"https://pith.science/api/pith-number/S4JG3NXFAOHNKFN5DJT4SC5AZP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/S4JG3NXFAOHNKFN5DJT4SC5AZP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/S4JG3NXFAOHNKFN5DJT4SC5AZP/action/storage_attestation","attest_author":"https://pith.science/pith/S4JG3NXFAOHNKFN5DJT4SC5AZP/action/author_attestation","sign_citation":"https://pith.science/pith/S4JG3NXFAOHNKFN5DJT4SC5AZP/action/citation_signature","submit_replication":"https://pith.science/pith/S4JG3NXFAOHNKFN5DJT4SC5AZP/action/replication_record"}},"created_at":"2026-07-02T01:18:02.759915+00:00","updated_at":"2026-07-02T01:18:02.759915+00:00"}