{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:GUSABTS5F363ZWVN35IRM2M7AU","short_pith_number":"pith:GUSABTS5","schema_version":"1.0","canonical_sha256":"352400ce5d2efdbcdaaddf5116699f053232880cafd76e45823260a42c1951c1","source":{"kind":"arxiv","id":"2506.03566","version":1},"attestation_state":"computed","paper":{"title":"POSS: Position Specialist Generates Better Draft for Speculative Decoding","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chengsong Huang, Di Huang, Jiaxin Huang, Jixuan Leng, Langlin Huang","submitted_at":"2025-06-04T04:30:30Z","abstract_excerpt":"Speculative decoding accelerates Large Language Model (LLM) inference by using a small draft model to predict multiple tokens, and a large target model to verify these tokens in parallel. Recent studies leverage the hidden state of the target model to enhance draft model prediction accuracy. However, existing methods suffer from the degrading quality of draft token predictions at later positions, due to error accumulation in draft model generated features. In this paper, we propose Position Specialists (PosS), which consist of multiple position-specialized draft layers to generate tokens at as"},"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":"2506.03566","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-04T04:30:30Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"c7bed8c82ee84cfbefd1239ffe40ab81461ef4504d10c3311e2ca18a59f52d60","abstract_canon_sha256":"61170e20324c7408d86a8d9b200844c6faa179009afecdbf4cc9be38799847c8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:15:30.028770Z","signature_b64":"aopz7x84TRlLW3yiqihMzBy0y1bWMy/l/Wad2+u/Od3JO5LBgN0QGEnP8ApgtXMFaqgefPglj9zcyIH1IGyvCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"352400ce5d2efdbcdaaddf5116699f053232880cafd76e45823260a42c1951c1","last_reissued_at":"2026-07-05T11:15:30.028334Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:15:30.028334Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"POSS: Position Specialist Generates Better Draft for Speculative Decoding","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chengsong Huang, Di Huang, Jiaxin Huang, Jixuan Leng, Langlin Huang","submitted_at":"2025-06-04T04:30:30Z","abstract_excerpt":"Speculative decoding accelerates Large Language Model (LLM) inference by using a small draft model to predict multiple tokens, and a large target model to verify these tokens in parallel. Recent studies leverage the hidden state of the target model to enhance draft model prediction accuracy. However, existing methods suffer from the degrading quality of draft token predictions at later positions, due to error accumulation in draft model generated features. In this paper, we propose Position Specialists (PosS), which consist of multiple position-specialized draft layers to generate tokens at as"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.03566","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/2506.03566/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":"2506.03566","created_at":"2026-07-05T11:15:30.028392+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.03566v1","created_at":"2026-07-05T11:15:30.028392+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.03566","created_at":"2026-07-05T11:15:30.028392+00:00"},{"alias_kind":"pith_short_12","alias_value":"GUSABTS5F363","created_at":"2026-07-05T11:15:30.028392+00:00"},{"alias_kind":"pith_short_16","alias_value":"GUSABTS5F363ZWVN","created_at":"2026-07-05T11:15:30.028392+00:00"},{"alias_kind":"pith_short_8","alias_value":"GUSABTS5","created_at":"2026-07-05T11:15:30.028392+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/GUSABTS5F363ZWVN35IRM2M7AU","json":"https://pith.science/pith/GUSABTS5F363ZWVN35IRM2M7AU.json","graph_json":"https://pith.science/api/pith-number/GUSABTS5F363ZWVN35IRM2M7AU/graph.json","events_json":"https://pith.science/api/pith-number/GUSABTS5F363ZWVN35IRM2M7AU/events.json","paper":"https://pith.science/paper/GUSABTS5"},"agent_actions":{"view_html":"https://pith.science/pith/GUSABTS5F363ZWVN35IRM2M7AU","download_json":"https://pith.science/pith/GUSABTS5F363ZWVN35IRM2M7AU.json","view_paper":"https://pith.science/paper/GUSABTS5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.03566&json=true","fetch_graph":"https://pith.science/api/pith-number/GUSABTS5F363ZWVN35IRM2M7AU/graph.json","fetch_events":"https://pith.science/api/pith-number/GUSABTS5F363ZWVN35IRM2M7AU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GUSABTS5F363ZWVN35IRM2M7AU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GUSABTS5F363ZWVN35IRM2M7AU/action/storage_attestation","attest_author":"https://pith.science/pith/GUSABTS5F363ZWVN35IRM2M7AU/action/author_attestation","sign_citation":"https://pith.science/pith/GUSABTS5F363ZWVN35IRM2M7AU/action/citation_signature","submit_replication":"https://pith.science/pith/GUSABTS5F363ZWVN35IRM2M7AU/action/replication_record"}},"created_at":"2026-07-05T11:15:30.028392+00:00","updated_at":"2026-07-05T11:15:30.028392+00:00"}