{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:L2BSKLNG4HZJNAPWVWJMNLME5Y","short_pith_number":"pith:L2BSKLNG","schema_version":"1.0","canonical_sha256":"5e83252da6e1f29681f6ad92c6ad84ee34e8d6952e68db33be4f3fa2931ff5cb","source":{"kind":"arxiv","id":"2406.01566","version":2},"attestation_state":"computed","paper":{"title":"Helix: Serving Large Language Models over Heterogeneous GPUs and Network via Max-Flow","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.DC","authors_text":"Juncheng Yang, Rashmi Vinayak, Xupeng Miao, Yixuan Mei, Yonghao Zhuang, Zhihao Jia","submitted_at":"2024-06-03T17:47:53Z","abstract_excerpt":"This paper introduces Helix, a distributed system for high-throughput, low-latency large language model (LLM) serving in heterogeneous GPU clusters. The key idea behind Helix is to formulate inference computation of LLMs over heterogeneous GPUs and network connections as a max-flow problem on directed, weighted graphs, whose nodes represent GPU instances and edges capture both GPU and network heterogeneity through their capacities. Helix then uses a mixed integer linear programming (MILP) algorithm to discover highly optimized strategies to serve LLMs on heterogeneous GPUs. This approach allow"},"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":"2406.01566","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2024-06-03T17:47:53Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"2c452c7ecf832806ab4fdc4b1e06936c75349ffe07116a88955b0fe2f30b12fa","abstract_canon_sha256":"e8bdab36ca5eca12b9b1ecdcf783750b2b285331f14c6cc59a8dc9647f5f9bcf"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:25:07.030872Z","signature_b64":"kgU7G2kU/89sNG7rMkbTBYV+g0iW9ZEMUPNpcW69szr397RLAStzOOaBKAtJjQB7kkn+iW5IZRO1dQLF4nPDBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5e83252da6e1f29681f6ad92c6ad84ee34e8d6952e68db33be4f3fa2931ff5cb","last_reissued_at":"2026-07-05T10:25:07.030177Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:25:07.030177Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Helix: Serving Large Language Models over Heterogeneous GPUs and Network via Max-Flow","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.DC","authors_text":"Juncheng Yang, Rashmi Vinayak, Xupeng Miao, Yixuan Mei, Yonghao Zhuang, Zhihao Jia","submitted_at":"2024-06-03T17:47:53Z","abstract_excerpt":"This paper introduces Helix, a distributed system for high-throughput, low-latency large language model (LLM) serving in heterogeneous GPU clusters. The key idea behind Helix is to formulate inference computation of LLMs over heterogeneous GPUs and network connections as a max-flow problem on directed, weighted graphs, whose nodes represent GPU instances and edges capture both GPU and network heterogeneity through their capacities. Helix then uses a mixed integer linear programming (MILP) algorithm to discover highly optimized strategies to serve LLMs on heterogeneous GPUs. This approach allow"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.01566","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/2406.01566/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":"2406.01566","created_at":"2026-07-05T10:25:07.030259+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.01566v2","created_at":"2026-07-05T10:25:07.030259+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.01566","created_at":"2026-07-05T10:25:07.030259+00:00"},{"alias_kind":"pith_short_12","alias_value":"L2BSKLNG4HZJ","created_at":"2026-07-05T10:25:07.030259+00:00"},{"alias_kind":"pith_short_16","alias_value":"L2BSKLNG4HZJNAPW","created_at":"2026-07-05T10:25:07.030259+00:00"},{"alias_kind":"pith_short_8","alias_value":"L2BSKLNG","created_at":"2026-07-05T10:25:07.030259+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.06453","citing_title":"Vortex: Efficient and Programmable Sparse Attention Serving for AI Agents","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2409.01143","citing_title":"HexiScale: Facilitating Large Language Model Training over Heterogeneous Hardware","ref_index":27,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/L2BSKLNG4HZJNAPWVWJMNLME5Y","json":"https://pith.science/pith/L2BSKLNG4HZJNAPWVWJMNLME5Y.json","graph_json":"https://pith.science/api/pith-number/L2BSKLNG4HZJNAPWVWJMNLME5Y/graph.json","events_json":"https://pith.science/api/pith-number/L2BSKLNG4HZJNAPWVWJMNLME5Y/events.json","paper":"https://pith.science/paper/L2BSKLNG"},"agent_actions":{"view_html":"https://pith.science/pith/L2BSKLNG4HZJNAPWVWJMNLME5Y","download_json":"https://pith.science/pith/L2BSKLNG4HZJNAPWVWJMNLME5Y.json","view_paper":"https://pith.science/paper/L2BSKLNG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.01566&json=true","fetch_graph":"https://pith.science/api/pith-number/L2BSKLNG4HZJNAPWVWJMNLME5Y/graph.json","fetch_events":"https://pith.science/api/pith-number/L2BSKLNG4HZJNAPWVWJMNLME5Y/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/L2BSKLNG4HZJNAPWVWJMNLME5Y/action/timestamp_anchor","attest_storage":"https://pith.science/pith/L2BSKLNG4HZJNAPWVWJMNLME5Y/action/storage_attestation","attest_author":"https://pith.science/pith/L2BSKLNG4HZJNAPWVWJMNLME5Y/action/author_attestation","sign_citation":"https://pith.science/pith/L2BSKLNG4HZJNAPWVWJMNLME5Y/action/citation_signature","submit_replication":"https://pith.science/pith/L2BSKLNG4HZJNAPWVWJMNLME5Y/action/replication_record"}},"created_at":"2026-07-05T10:25:07.030259+00:00","updated_at":"2026-07-05T10:25:07.030259+00:00"}