{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:CTTX4GMKXSMMCGHLLQUZEH3UHM","short_pith_number":"pith:CTTX4GMK","schema_version":"1.0","canonical_sha256":"14e77e198abc98c118eb5c29921f743b2b8e7b5131a98f3ce2c433361e852e58","source":{"kind":"arxiv","id":"2510.01336","version":2},"attestation_state":"computed","paper":{"title":"HiSpec: Hierarchical Speculative Decoding for LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Avinash Kumar, Poulami Das, Sujay Sanghavi","submitted_at":"2025-10-01T18:04:14Z","abstract_excerpt":"Speculative decoding accelerates LLM inference by using a smaller draft model to speculate tokens that a larger target model verifies. Verification is often the bottleneck (e.g. verification is $4\\times$ slower than token generation when a 3B model speculates for a 70B target model), but most prior works focus only on accelerating drafting. $\\textit{``Intermediate\"}$ verification reduces verification time by discarding inaccurate draft tokens early, but existing methods incur substantial training overheads in incorporating the intermediate verifier, increase the memory footprint to orchestrate"},"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":"2510.01336","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-10-01T18:04:14Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"751629ccda389f32d8f5fd653a8d139149ecc085aac724d0a12a4dbb51529c28","abstract_canon_sha256":"fb5914bf822069576d63ee3b9924a741bbf893f67b9a8180420ae276e0860c7b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-27T01:04:51.632806Z","signature_b64":"m5TBnH/eMPYFBYwYWgWHSzz3IMQGdbVUHHGbnZK9Uk0eyjBSHkg23bnWixxqbSAo5ZE/3xuTzi1NnHp2qApuAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"14e77e198abc98c118eb5c29921f743b2b8e7b5131a98f3ce2c433361e852e58","last_reissued_at":"2026-05-27T01:04:51.631825Z","signature_status":"signed_v1","first_computed_at":"2026-05-27T01:04:51.631825Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"HiSpec: Hierarchical Speculative Decoding for LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Avinash Kumar, Poulami Das, Sujay Sanghavi","submitted_at":"2025-10-01T18:04:14Z","abstract_excerpt":"Speculative decoding accelerates LLM inference by using a smaller draft model to speculate tokens that a larger target model verifies. Verification is often the bottleneck (e.g. verification is $4\\times$ slower than token generation when a 3B model speculates for a 70B target model), but most prior works focus only on accelerating drafting. $\\textit{``Intermediate\"}$ verification reduces verification time by discarding inaccurate draft tokens early, but existing methods incur substantial training overheads in incorporating the intermediate verifier, increase the memory footprint to orchestrate"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2510.01336","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/2510.01336/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":"2510.01336","created_at":"2026-05-27T01:04:51.631944+00:00"},{"alias_kind":"arxiv_version","alias_value":"2510.01336v2","created_at":"2026-05-27T01:04:51.631944+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2510.01336","created_at":"2026-05-27T01:04:51.631944+00:00"},{"alias_kind":"pith_short_12","alias_value":"CTTX4GMKXSMM","created_at":"2026-05-27T01:04:51.631944+00:00"},{"alias_kind":"pith_short_16","alias_value":"CTTX4GMKXSMMCGHL","created_at":"2026-05-27T01:04:51.631944+00:00"},{"alias_kind":"pith_short_8","alias_value":"CTTX4GMK","created_at":"2026-05-27T01:04:51.631944+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2604.20503","citing_title":"FASER: Fine-Grained Phase Management for Speculative Decoding in Dynamic LLM Serving","ref_index":18,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CTTX4GMKXSMMCGHLLQUZEH3UHM","json":"https://pith.science/pith/CTTX4GMKXSMMCGHLLQUZEH3UHM.json","graph_json":"https://pith.science/api/pith-number/CTTX4GMKXSMMCGHLLQUZEH3UHM/graph.json","events_json":"https://pith.science/api/pith-number/CTTX4GMKXSMMCGHLLQUZEH3UHM/events.json","paper":"https://pith.science/paper/CTTX4GMK"},"agent_actions":{"view_html":"https://pith.science/pith/CTTX4GMKXSMMCGHLLQUZEH3UHM","download_json":"https://pith.science/pith/CTTX4GMKXSMMCGHLLQUZEH3UHM.json","view_paper":"https://pith.science/paper/CTTX4GMK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2510.01336&json=true","fetch_graph":"https://pith.science/api/pith-number/CTTX4GMKXSMMCGHLLQUZEH3UHM/graph.json","fetch_events":"https://pith.science/api/pith-number/CTTX4GMKXSMMCGHLLQUZEH3UHM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CTTX4GMKXSMMCGHLLQUZEH3UHM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CTTX4GMKXSMMCGHLLQUZEH3UHM/action/storage_attestation","attest_author":"https://pith.science/pith/CTTX4GMKXSMMCGHLLQUZEH3UHM/action/author_attestation","sign_citation":"https://pith.science/pith/CTTX4GMKXSMMCGHLLQUZEH3UHM/action/citation_signature","submit_replication":"https://pith.science/pith/CTTX4GMKXSMMCGHLLQUZEH3UHM/action/replication_record"}},"created_at":"2026-05-27T01:04:51.631944+00:00","updated_at":"2026-05-27T01:04:51.631944+00:00"}