{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:TKHUEXQ2IO4NVTUOSRKM47WBJC","short_pith_number":"pith:TKHUEXQ2","schema_version":"1.0","canonical_sha256":"9a8f425e1a43b8dace8e9454ce7ec1488e19f16966663c41785487381563248c","source":{"kind":"arxiv","id":"2608.13057","version":1},"attestation_state":"computed","paper":{"title":"TEMPO: Makespan-Aware Expert-Parallel Load Balancing Across Memory- and Compute-Bound Regimes","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.GT"],"primary_cat":"cs.DC","authors_text":"Chengru Song, Chenxin Jia, Cunzhuang Liu, Jianwen Xian, Jie Li, Jinliang Shen, Kang He, Ruiyi Ding","submitted_at":"2026-08-13T10:21:11Z","abstract_excerpt":"In expert-parallel (EP) MoE serving, every layer synchronizes at the slowest GPU. Dispatchers balance token counts (EPLB, LPLB, UltraEP) or activated-expert counts (METRO), assuming expert time is linear in one. Measurements on two datacenter GPU generations show it is neither: below $\\nstar\\!\\approx\\!156$--$168$ tokens, HBM weight streaming dominates---cost attaches to \\emph{activated replicas}, not tokens; above it, grouped GEMM rounds tokens to 128-tile $M$-tiles, so \\emph{splitting} an expert adds padded compute. A max-affine profile $t=\\max(a+bG,\\,c+\\beta N)$ captures both regimes. Realis"},"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":"2608.13057","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2026-08-13T10:21:11Z","cross_cats_sorted":["cs.AI","cs.CL","cs.GT"],"title_canon_sha256":"ee85d4171f2ae106ebd17743f908eef7abad2ac4238cc2cf6abcd7c0232db584","abstract_canon_sha256":"29237db8b23dac52dc9bf3de54b9b82c4dc49555e807cfd0b897eb3b9ea17ac2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-14T01:00:14.429653Z","signature_b64":"vhP5xJZVUXS4oURA79QoLGSdMeCbzJyz2nUgbLkL5tFcvFHn3RzgVzTIIge0O4y7nqk4PG0XKfAX0SEKaIlnBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9a8f425e1a43b8dace8e9454ce7ec1488e19f16966663c41785487381563248c","last_reissued_at":"2026-08-14T01:00:14.427619Z","signature_status":"signed_v1","first_computed_at":"2026-08-14T01:00:14.427619Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TEMPO: Makespan-Aware Expert-Parallel Load Balancing Across Memory- and Compute-Bound Regimes","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.GT"],"primary_cat":"cs.DC","authors_text":"Chengru Song, Chenxin Jia, Cunzhuang Liu, Jianwen Xian, Jie Li, Jinliang Shen, Kang He, Ruiyi Ding","submitted_at":"2026-08-13T10:21:11Z","abstract_excerpt":"In expert-parallel (EP) MoE serving, every layer synchronizes at the slowest GPU. Dispatchers balance token counts (EPLB, LPLB, UltraEP) or activated-expert counts (METRO), assuming expert time is linear in one. Measurements on two datacenter GPU generations show it is neither: below $\\nstar\\!\\approx\\!156$--$168$ tokens, HBM weight streaming dominates---cost attaches to \\emph{activated replicas}, not tokens; above it, grouped GEMM rounds tokens to 128-tile $M$-tiles, so \\emph{splitting} an expert adds padded compute. A max-affine profile $t=\\max(a+bG,\\,c+\\beta N)$ captures both regimes. Realis"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.13057","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/2608.13057/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":"2608.13057","created_at":"2026-08-14T01:00:14.428633+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.13057v1","created_at":"2026-08-14T01:00:14.428633+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.13057","created_at":"2026-08-14T01:00:14.428633+00:00"},{"alias_kind":"pith_short_12","alias_value":"TKHUEXQ2IO4N","created_at":"2026-08-14T01:00:14.428633+00:00"},{"alias_kind":"pith_short_16","alias_value":"TKHUEXQ2IO4NVTUO","created_at":"2026-08-14T01:00:14.428633+00:00"},{"alias_kind":"pith_short_8","alias_value":"TKHUEXQ2","created_at":"2026-08-14T01:00:14.428633+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TKHUEXQ2IO4NVTUOSRKM47WBJC","json":"https://pith.science/pith/TKHUEXQ2IO4NVTUOSRKM47WBJC.json","graph_json":"https://pith.science/api/pith-number/TKHUEXQ2IO4NVTUOSRKM47WBJC/graph.json","events_json":"https://pith.science/api/pith-number/TKHUEXQ2IO4NVTUOSRKM47WBJC/events.json","paper":"https://pith.science/paper/TKHUEXQ2"},"agent_actions":{"view_html":"https://pith.science/pith/TKHUEXQ2IO4NVTUOSRKM47WBJC","download_json":"https://pith.science/pith/TKHUEXQ2IO4NVTUOSRKM47WBJC.json","view_paper":"https://pith.science/paper/TKHUEXQ2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.13057&json=true","fetch_graph":"https://pith.science/api/pith-number/TKHUEXQ2IO4NVTUOSRKM47WBJC/graph.json","fetch_events":"https://pith.science/api/pith-number/TKHUEXQ2IO4NVTUOSRKM47WBJC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TKHUEXQ2IO4NVTUOSRKM47WBJC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TKHUEXQ2IO4NVTUOSRKM47WBJC/action/storage_attestation","attest_author":"https://pith.science/pith/TKHUEXQ2IO4NVTUOSRKM47WBJC/action/author_attestation","sign_citation":"https://pith.science/pith/TKHUEXQ2IO4NVTUOSRKM47WBJC/action/citation_signature","submit_replication":"https://pith.science/pith/TKHUEXQ2IO4NVTUOSRKM47WBJC/action/replication_record"}},"created_at":"2026-08-14T01:00:14.428633+00:00","updated_at":"2026-08-14T01:00:14.428633+00:00"}