{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:DKP6BOW5J7RA77XFVZKUFO5TTA","short_pith_number":"pith:DKP6BOW5","schema_version":"1.0","canonical_sha256":"1a9fe0badd4fe20ffee5ae5542bbb3982b04c5acb998c89a1ae8084469451ca2","source":{"kind":"arxiv","id":"2308.04623","version":1},"attestation_state":"computed","paper":{"title":"Accelerating LLM Inference with Staged Speculative Decoding","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.AI","authors_text":"Benjamin Spector, Chris Re","submitted_at":"2023-08-08T23:29:55Z","abstract_excerpt":"Recent advances with large language models (LLM) illustrate their diverse capabilities. We propose a novel algorithm, staged speculative decoding, to accelerate LLM inference in small-batch, on-device scenarios. We address the low arithmetic intensity of small-batch inference by improving upon previous work in speculative decoding. First, we restructure the speculative batch as a tree, which reduces generation costs and increases the expected tokens per batch. Second, we add a second stage of speculative decoding. Taken together, we reduce single-batch decoding latency by 3.16x with a 762M par"},"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":"2308.04623","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2023-08-08T23:29:55Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"582d2e1aa1c37a836e83b03edafac6b309aae63f535dcaf622932531b43578e8","abstract_canon_sha256":"762f259fe2a27d9975911ed67a8e37de52632e0a6e9519986c6abb0de4e588f2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:39:40.442002Z","signature_b64":"+P3PQfi29NKLA/pkv7SRmIBKYdqsrsPHQZWLKfmxFqP+zdfpXHzJ7TKo6bbf7fF6my3b1T6U2E8ak/SUHR5LCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1a9fe0badd4fe20ffee5ae5542bbb3982b04c5acb998c89a1ae8084469451ca2","last_reissued_at":"2026-07-05T06:39:40.441554Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:39:40.441554Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Accelerating LLM Inference with Staged Speculative Decoding","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.AI","authors_text":"Benjamin Spector, Chris Re","submitted_at":"2023-08-08T23:29:55Z","abstract_excerpt":"Recent advances with large language models (LLM) illustrate their diverse capabilities. We propose a novel algorithm, staged speculative decoding, to accelerate LLM inference in small-batch, on-device scenarios. We address the low arithmetic intensity of small-batch inference by improving upon previous work in speculative decoding. First, we restructure the speculative batch as a tree, which reduces generation costs and increases the expected tokens per batch. Second, we add a second stage of speculative decoding. Taken together, we reduce single-batch decoding latency by 3.16x with a 762M par"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2308.04623","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/2308.04623/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":"2308.04623","created_at":"2026-07-05T06:39:40.441605+00:00"},{"alias_kind":"arxiv_version","alias_value":"2308.04623v1","created_at":"2026-07-05T06:39:40.441605+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2308.04623","created_at":"2026-07-05T06:39:40.441605+00:00"},{"alias_kind":"pith_short_12","alias_value":"DKP6BOW5J7RA","created_at":"2026-07-05T06:39:40.441605+00:00"},{"alias_kind":"pith_short_16","alias_value":"DKP6BOW5J7RA77XF","created_at":"2026-07-05T06:39:40.441605+00:00"},{"alias_kind":"pith_short_8","alias_value":"DKP6BOW5","created_at":"2026-07-05T06:39:40.441605+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":12,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.06763","citing_title":"Trees from Marginals: Autoregressive drafting with factorized priors","ref_index":53,"is_internal_anchor":true},{"citing_arxiv_id":"2606.22175","citing_title":"StickyInvoc: Rethinking Task Models for High-throughput Workflows in the LLM Era","ref_index":72,"is_internal_anchor":false},{"citing_arxiv_id":"2606.10935","citing_title":"CLP: Collocation-Length Prediction for Zero-Loss Adaptive Multi-Token Inference","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01813","citing_title":"Cost-Aware Diffusion Draft Trees for Speculative Decoding","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29639","citing_title":"RTP-LLM: High-Performance Alibaba LLM Inference Engine","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2501.05465","citing_title":"Small Language Models (SLMs) Can Still Pack a Punch: A survey (updated 2026)","ref_index":119,"is_internal_anchor":false},{"citing_arxiv_id":"2404.14294","citing_title":"A Survey on Efficient Inference for Large Language Models","ref_index":244,"is_internal_anchor":false},{"citing_arxiv_id":"2401.15077","citing_title":"EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty","ref_index":71,"is_internal_anchor":false},{"citing_arxiv_id":"2604.02965","citing_title":"Open-Loop Planning, Closed-Loop Verification: Speculative Verification for VLA","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2401.10774","citing_title":"Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads","ref_index":84,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20021","citing_title":"Continuous Semantic Caching for Low-Cost LLM Serving","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17701","citing_title":"WISV: Wireless-Informed Semantic Verification for Distributed Speculative Decoding in Device-Edge LLM Inference","ref_index":24,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DKP6BOW5J7RA77XFVZKUFO5TTA","json":"https://pith.science/pith/DKP6BOW5J7RA77XFVZKUFO5TTA.json","graph_json":"https://pith.science/api/pith-number/DKP6BOW5J7RA77XFVZKUFO5TTA/graph.json","events_json":"https://pith.science/api/pith-number/DKP6BOW5J7RA77XFVZKUFO5TTA/events.json","paper":"https://pith.science/paper/DKP6BOW5"},"agent_actions":{"view_html":"https://pith.science/pith/DKP6BOW5J7RA77XFVZKUFO5TTA","download_json":"https://pith.science/pith/DKP6BOW5J7RA77XFVZKUFO5TTA.json","view_paper":"https://pith.science/paper/DKP6BOW5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2308.04623&json=true","fetch_graph":"https://pith.science/api/pith-number/DKP6BOW5J7RA77XFVZKUFO5TTA/graph.json","fetch_events":"https://pith.science/api/pith-number/DKP6BOW5J7RA77XFVZKUFO5TTA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DKP6BOW5J7RA77XFVZKUFO5TTA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DKP6BOW5J7RA77XFVZKUFO5TTA/action/storage_attestation","attest_author":"https://pith.science/pith/DKP6BOW5J7RA77XFVZKUFO5TTA/action/author_attestation","sign_citation":"https://pith.science/pith/DKP6BOW5J7RA77XFVZKUFO5TTA/action/citation_signature","submit_replication":"https://pith.science/pith/DKP6BOW5J7RA77XFVZKUFO5TTA/action/replication_record"}},"created_at":"2026-07-05T06:39:40.441605+00:00","updated_at":"2026-07-05T06:39:40.441605+00:00"}