{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:CGXQZXFSNW7MP5FJFM5PLOFTDB","short_pith_number":"pith:CGXQZXFS","schema_version":"1.0","canonical_sha256":"11af0cdcb26dbec7f4a92b3af5b8b3186a632b6a823a70f78ebb110aad6d6d2f","source":{"kind":"arxiv","id":"2310.18813","version":1},"attestation_state":"computed","paper":{"title":"The Synergy of Speculative Decoding and Batching in Serving Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.DC"],"primary_cat":"cs.LG","authors_text":"Christina Giannoula, Gennady Pekhimenko, Qidong Su","submitted_at":"2023-10-28T20:36:36Z","abstract_excerpt":"Large Language Models (LLMs) like GPT are state-of-the-art text generation models that provide significant assistance in daily routines. However, LLM execution is inherently sequential, since they only produce one token at a time, thus incurring low hardware utilization on modern GPUs. Batching and speculative decoding are two techniques to improve GPU hardware utilization in LLM inference. To study their synergy, we implement a prototype implementation and perform an extensive characterization analysis on various LLM models and GPU architectures. We observe that the optimal speculation length"},"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":"2310.18813","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-10-28T20:36:36Z","cross_cats_sorted":["cs.DC"],"title_canon_sha256":"266d29daee87f96c6985eda4ff98114b615494b6dcb02803f485994e024f5c24","abstract_canon_sha256":"38cf249ee3c18e3a9c605f1fee46d48a328edcf9131481f3422231bf3671fab1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:06:41.911445Z","signature_b64":"fMizhZiJAU3j1Yx+ur14fAOxDpt2cRkn09+yndho/Kvrs/lg+SKli3abVTL5TiJdNIkXENWIYgzSolq3CQmbCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"11af0cdcb26dbec7f4a92b3af5b8b3186a632b6a823a70f78ebb110aad6d6d2f","last_reissued_at":"2026-07-05T07:06:41.910969Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:06:41.910969Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Synergy of Speculative Decoding and Batching in Serving Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.DC"],"primary_cat":"cs.LG","authors_text":"Christina Giannoula, Gennady Pekhimenko, Qidong Su","submitted_at":"2023-10-28T20:36:36Z","abstract_excerpt":"Large Language Models (LLMs) like GPT are state-of-the-art text generation models that provide significant assistance in daily routines. However, LLM execution is inherently sequential, since they only produce one token at a time, thus incurring low hardware utilization on modern GPUs. Batching and speculative decoding are two techniques to improve GPU hardware utilization in LLM inference. To study their synergy, we implement a prototype implementation and perform an extensive characterization analysis on various LLM models and GPU architectures. We observe that the optimal speculation length"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.18813","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/2310.18813/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":"2310.18813","created_at":"2026-07-05T07:06:41.911025+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.18813v1","created_at":"2026-07-05T07:06:41.911025+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.18813","created_at":"2026-07-05T07:06:41.911025+00:00"},{"alias_kind":"pith_short_12","alias_value":"CGXQZXFSNW7M","created_at":"2026-07-05T07:06:41.911025+00:00"},{"alias_kind":"pith_short_16","alias_value":"CGXQZXFSNW7MP5FJ","created_at":"2026-07-05T07:06:41.911025+00:00"},{"alias_kind":"pith_short_8","alias_value":"CGXQZXFS","created_at":"2026-07-05T07:06:41.911025+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.08446","citing_title":"Sparrow: Sparse Rollout for Stable and Efficient Long-context RL of Large Language Models","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04581","citing_title":"Multi-SPIN: Multi-Access Speculative Inference for Cooperative Token Generation at the Edge","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30265","citing_title":"When Is a Draft Accepted? A Theory of Acceptance in Speculative Decoding","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2401.10774","citing_title":"Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads","ref_index":63,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CGXQZXFSNW7MP5FJFM5PLOFTDB","json":"https://pith.science/pith/CGXQZXFSNW7MP5FJFM5PLOFTDB.json","graph_json":"https://pith.science/api/pith-number/CGXQZXFSNW7MP5FJFM5PLOFTDB/graph.json","events_json":"https://pith.science/api/pith-number/CGXQZXFSNW7MP5FJFM5PLOFTDB/events.json","paper":"https://pith.science/paper/CGXQZXFS"},"agent_actions":{"view_html":"https://pith.science/pith/CGXQZXFSNW7MP5FJFM5PLOFTDB","download_json":"https://pith.science/pith/CGXQZXFSNW7MP5FJFM5PLOFTDB.json","view_paper":"https://pith.science/paper/CGXQZXFS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.18813&json=true","fetch_graph":"https://pith.science/api/pith-number/CGXQZXFSNW7MP5FJFM5PLOFTDB/graph.json","fetch_events":"https://pith.science/api/pith-number/CGXQZXFSNW7MP5FJFM5PLOFTDB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CGXQZXFSNW7MP5FJFM5PLOFTDB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CGXQZXFSNW7MP5FJFM5PLOFTDB/action/storage_attestation","attest_author":"https://pith.science/pith/CGXQZXFSNW7MP5FJFM5PLOFTDB/action/author_attestation","sign_citation":"https://pith.science/pith/CGXQZXFSNW7MP5FJFM5PLOFTDB/action/citation_signature","submit_replication":"https://pith.science/pith/CGXQZXFSNW7MP5FJFM5PLOFTDB/action/replication_record"}},"created_at":"2026-07-05T07:06:41.911025+00:00","updated_at":"2026-07-05T07:06:41.911025+00:00"}