{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:IWHBLG4AMZEWO6E2E7PMZ7R7QW","short_pith_number":"pith:IWHBLG4A","schema_version":"1.0","canonical_sha256":"458e159b80664967789a27deccfe3f85a5375c0b5d7ac13a872e225375086f9b","source":{"kind":"arxiv","id":"2508.06133","version":4},"attestation_state":"computed","paper":{"title":"LLM Serving Optimization with Variable Prefill and Decode Lengths","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"math.OC","authors_text":"Meixuan Wang, Yinyu Ye, Zijie Zhou","submitted_at":"2025-08-08T08:54:21Z","abstract_excerpt":"We study offline scheduling for large language model (LLM) serving under a fixed KV-cache memory budget, where requests have heterogeneous prompt (prefill) and response (decode) lengths. Prompt tokens determine initial KV-cache usage, while each generated token further increases memory consumption, creating dynamic memory constraints during autoregressive decoding. Given a backlog of n requests arriving together, the goal is to form mixed prefill and decode batches over time to minimize total end-to-end latency. We show that heterogeneous prompt lengths fundamentally change the scheduling prob"},"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":"2508.06133","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"math.OC","submitted_at":"2025-08-08T08:54:21Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"c7cc9c445cb243aa5993295280b171c5af5d2a44e64707dc3433e93788cc3649","abstract_canon_sha256":"1d901e92d8071462f17d175265ee2092d010d8b8f9e7367fdeb87cf07611d8c1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-30T01:17:25.191963Z","signature_b64":"ynlEknoN9cyjIkqnazChWrrH+7kpQeIgAF8HKvz+1VbgvNrK8X5UElluNoI2/lhWqAmOY9eiV8d6pVltRXyfBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"458e159b80664967789a27deccfe3f85a5375c0b5d7ac13a872e225375086f9b","last_reissued_at":"2026-06-30T01:17:25.191276Z","signature_status":"signed_v1","first_computed_at":"2026-06-30T01:17:25.191276Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LLM Serving Optimization with Variable Prefill and Decode Lengths","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"math.OC","authors_text":"Meixuan Wang, Yinyu Ye, Zijie Zhou","submitted_at":"2025-08-08T08:54:21Z","abstract_excerpt":"We study offline scheduling for large language model (LLM) serving under a fixed KV-cache memory budget, where requests have heterogeneous prompt (prefill) and response (decode) lengths. Prompt tokens determine initial KV-cache usage, while each generated token further increases memory consumption, creating dynamic memory constraints during autoregressive decoding. Given a backlog of n requests arriving together, the goal is to form mixed prefill and decode batches over time to minimize total end-to-end latency. We show that heterogeneous prompt lengths fundamentally change the scheduling prob"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.06133","kind":"arxiv","version":4},"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/2508.06133/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":"2508.06133","created_at":"2026-06-30T01:17:25.191364+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.06133v4","created_at":"2026-06-30T01:17:25.191364+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.06133","created_at":"2026-06-30T01:17:25.191364+00:00"},{"alias_kind":"pith_short_12","alias_value":"IWHBLG4AMZEW","created_at":"2026-06-30T01:17:25.191364+00:00"},{"alias_kind":"pith_short_16","alias_value":"IWHBLG4AMZEWO6E2","created_at":"2026-06-30T01:17:25.191364+00:00"},{"alias_kind":"pith_short_8","alias_value":"IWHBLG4A","created_at":"2026-06-30T01:17:25.191364+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":7,"sample":[{"citing_arxiv_id":"2606.22327","citing_title":"Geometry-Aware Online Scheduling for LLM Serving: From Theoretical Bound to System Practice","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2504.11320","citing_title":"Optimizing LLM Inference: Fluid-Guided Online Scheduling with Memory Constraints","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06113","citing_title":"Tackling the Data-Parallel Load Balancing Bottleneck in LLM Serving: Practical Online Routing at Scale","ref_index":49,"is_internal_anchor":true},{"citing_arxiv_id":"2605.04595","citing_title":"A Queueing-Theoretic Framework for Stability Analysis of LLM Inference with KV Cache Memory Constraints","ref_index":17,"is_internal_anchor":true},{"citing_arxiv_id":"2605.01280","citing_title":"Position: LLM Serving Needs Mathematical Optimization and Algorithmic Foundations, Not Just Heuristics","ref_index":53,"is_internal_anchor":true},{"citing_arxiv_id":"2604.11001","citing_title":"Flow-Controlled Scheduling for LLM Inference with Provable Stability Guarantees","ref_index":28,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06113","citing_title":"Tackling the Data-Parallel Load Balancing Bottleneck in LLM Serving: Practical Online Routing at Scale","ref_index":49,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IWHBLG4AMZEWO6E2E7PMZ7R7QW","json":"https://pith.science/pith/IWHBLG4AMZEWO6E2E7PMZ7R7QW.json","graph_json":"https://pith.science/api/pith-number/IWHBLG4AMZEWO6E2E7PMZ7R7QW/graph.json","events_json":"https://pith.science/api/pith-number/IWHBLG4AMZEWO6E2E7PMZ7R7QW/events.json","paper":"https://pith.science/paper/IWHBLG4A"},"agent_actions":{"view_html":"https://pith.science/pith/IWHBLG4AMZEWO6E2E7PMZ7R7QW","download_json":"https://pith.science/pith/IWHBLG4AMZEWO6E2E7PMZ7R7QW.json","view_paper":"https://pith.science/paper/IWHBLG4A","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.06133&json=true","fetch_graph":"https://pith.science/api/pith-number/IWHBLG4AMZEWO6E2E7PMZ7R7QW/graph.json","fetch_events":"https://pith.science/api/pith-number/IWHBLG4AMZEWO6E2E7PMZ7R7QW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IWHBLG4AMZEWO6E2E7PMZ7R7QW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IWHBLG4AMZEWO6E2E7PMZ7R7QW/action/storage_attestation","attest_author":"https://pith.science/pith/IWHBLG4AMZEWO6E2E7PMZ7R7QW/action/author_attestation","sign_citation":"https://pith.science/pith/IWHBLG4AMZEWO6E2E7PMZ7R7QW/action/citation_signature","submit_replication":"https://pith.science/pith/IWHBLG4AMZEWO6E2E7PMZ7R7QW/action/replication_record"}},"created_at":"2026-06-30T01:17:25.191364+00:00","updated_at":"2026-06-30T01:17:25.191364+00:00"}