{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:EDMAOBFXTUAKYLPYOKUKEZMXRI","short_pith_number":"pith:EDMAOBFX","schema_version":"1.0","canonical_sha256":"20d80704b79d00ac2df872a8a265978a3ff7e37b43df577d9f3865c4eff91ae5","source":{"kind":"arxiv","id":"2511.17826","version":2},"attestation_state":"computed","paper":{"title":"Deterministic Inference across Tensor Parallel Sizes That Eliminates Training-Inference Mismatch","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Huizi Mao, Jiarong Xing, Jiayi Yuan, Rixin Liu, Xinheng Ding, Zirui Liu, Ziyang Zhang","submitted_at":"2025-11-21T22:40:00Z","abstract_excerpt":"Deterministic inference is increasingly critical for large language model (LLM) applications such as LLM-as-a-judge evaluation, multi-agent systems, and Reinforcement Learning (RL). However, existing LLM serving frameworks exhibit non-deterministic behavior: identical inputs can yield different outputs when system configurations (e.g., tensor parallel (TP) size, batch size) vary, even under greedy decoding. This arises from the non-associativity of floating-point arithmetic and inconsistent reduction orders across GPUs. While prior work has addressed batch-size-related nondeterminism through b"},"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":"2511.17826","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-11-21T22:40:00Z","cross_cats_sorted":["cs.CL","stat.ML"],"title_canon_sha256":"b112aad11ebef86ae8c82ab616b71895806eff38f5522cb163e5f549cd0644b2","abstract_canon_sha256":"37935a4b2a52edbd131291edd0f1a6dc2470a66e77e2d7ed56b1202bd783557e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-01T01:02:25.936797Z","signature_b64":"CaFvwLi2/l5OBHBvbNokOT9U4UETkdn7Xue4fkRM9hvZSoTDAUneAN72z5J3OTeQJJSX0K/gXy/cY9NC0u+SBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"20d80704b79d00ac2df872a8a265978a3ff7e37b43df577d9f3865c4eff91ae5","last_reissued_at":"2026-06-01T01:02:25.935743Z","signature_status":"signed_v1","first_computed_at":"2026-06-01T01:02:25.935743Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Deterministic Inference across Tensor Parallel Sizes That Eliminates Training-Inference Mismatch","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Huizi Mao, Jiarong Xing, Jiayi Yuan, Rixin Liu, Xinheng Ding, Zirui Liu, Ziyang Zhang","submitted_at":"2025-11-21T22:40:00Z","abstract_excerpt":"Deterministic inference is increasingly critical for large language model (LLM) applications such as LLM-as-a-judge evaluation, multi-agent systems, and Reinforcement Learning (RL). However, existing LLM serving frameworks exhibit non-deterministic behavior: identical inputs can yield different outputs when system configurations (e.g., tensor parallel (TP) size, batch size) vary, even under greedy decoding. This arises from the non-associativity of floating-point arithmetic and inconsistent reduction orders across GPUs. While prior work has addressed batch-size-related nondeterminism through b"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2511.17826","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/2511.17826/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":"2511.17826","created_at":"2026-06-01T01:02:25.935888+00:00"},{"alias_kind":"arxiv_version","alias_value":"2511.17826v2","created_at":"2026-06-01T01:02:25.935888+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2511.17826","created_at":"2026-06-01T01:02:25.935888+00:00"},{"alias_kind":"pith_short_12","alias_value":"EDMAOBFXTUAK","created_at":"2026-06-01T01:02:25.935888+00:00"},{"alias_kind":"pith_short_16","alias_value":"EDMAOBFXTUAKYLPY","created_at":"2026-06-01T01:02:25.935888+00:00"},{"alias_kind":"pith_short_8","alias_value":"EDMAOBFX","created_at":"2026-06-01T01:02:25.935888+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/EDMAOBFXTUAKYLPYOKUKEZMXRI","json":"https://pith.science/pith/EDMAOBFXTUAKYLPYOKUKEZMXRI.json","graph_json":"https://pith.science/api/pith-number/EDMAOBFXTUAKYLPYOKUKEZMXRI/graph.json","events_json":"https://pith.science/api/pith-number/EDMAOBFXTUAKYLPYOKUKEZMXRI/events.json","paper":"https://pith.science/paper/EDMAOBFX"},"agent_actions":{"view_html":"https://pith.science/pith/EDMAOBFXTUAKYLPYOKUKEZMXRI","download_json":"https://pith.science/pith/EDMAOBFXTUAKYLPYOKUKEZMXRI.json","view_paper":"https://pith.science/paper/EDMAOBFX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2511.17826&json=true","fetch_graph":"https://pith.science/api/pith-number/EDMAOBFXTUAKYLPYOKUKEZMXRI/graph.json","fetch_events":"https://pith.science/api/pith-number/EDMAOBFXTUAKYLPYOKUKEZMXRI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EDMAOBFXTUAKYLPYOKUKEZMXRI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EDMAOBFXTUAKYLPYOKUKEZMXRI/action/storage_attestation","attest_author":"https://pith.science/pith/EDMAOBFXTUAKYLPYOKUKEZMXRI/action/author_attestation","sign_citation":"https://pith.science/pith/EDMAOBFXTUAKYLPYOKUKEZMXRI/action/citation_signature","submit_replication":"https://pith.science/pith/EDMAOBFXTUAKYLPYOKUKEZMXRI/action/replication_record"}},"created_at":"2026-06-01T01:02:25.935888+00:00","updated_at":"2026-06-01T01:02:25.935888+00:00"}