{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:NEYVBA627KANMD7AR7GTVDZY6D","short_pith_number":"pith:NEYVBA62","schema_version":"1.0","canonical_sha256":"69315083dafa80d60fe08fcd3a8f38f0e9f94e2aefbcaba22c70112c5310f998","source":{"kind":"arxiv","id":"2503.13565","version":1},"attestation_state":"computed","paper":{"title":"ML-SpecQD: Multi-Level Speculative Decoding with Quantized Drafts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Alexander Heinecke, Alexander Kozlov, Dhiraj Kalamkar, Evangelos Georganas","submitted_at":"2025-03-17T08:38:45Z","abstract_excerpt":"Speculative decoding (SD) has emerged as a method to accelerate LLM inference without sacrificing any accuracy over the 16-bit model inference. In a typical SD setup, the idea is to use a full-precision, small, fast model as \"draft\" to generate the next few tokens and use the \"target\" large model to verify the draft-generated tokens. The efficacy of this method heavily relies on the acceptance ratio of the draft-generated tokens and the relative token throughput of the draft versus the target model. Nevertheless, an efficient SD pipeline requires pre-training and aligning the draft model to th"},"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":"2503.13565","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-03-17T08:38:45Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"f614c579c62684eba31d55c3c110bf984f013a045d65bc8ae7b805a584db4dca","abstract_canon_sha256":"21be91904b91b77e5bcedc8257e0cc0c7cd733b661d2b06326a113577eb7c264"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:33:12.969814Z","signature_b64":"guWA9B9bW1MKkHG1nf8v6U0D1wTmTZmaj+D1jWIT7ktu0LaHDDCe1z8HUMK+r6YtXpKTV2i54njemyNh8GgZBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"69315083dafa80d60fe08fcd3a8f38f0e9f94e2aefbcaba22c70112c5310f998","last_reissued_at":"2026-07-05T10:33:12.969306Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:33:12.969306Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ML-SpecQD: Multi-Level Speculative Decoding with Quantized Drafts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Alexander Heinecke, Alexander Kozlov, Dhiraj Kalamkar, Evangelos Georganas","submitted_at":"2025-03-17T08:38:45Z","abstract_excerpt":"Speculative decoding (SD) has emerged as a method to accelerate LLM inference without sacrificing any accuracy over the 16-bit model inference. In a typical SD setup, the idea is to use a full-precision, small, fast model as \"draft\" to generate the next few tokens and use the \"target\" large model to verify the draft-generated tokens. The efficacy of this method heavily relies on the acceptance ratio of the draft-generated tokens and the relative token throughput of the draft versus the target model. Nevertheless, an efficient SD pipeline requires pre-training and aligning the draft model to th"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.13565","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/2503.13565/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":"2503.13565","created_at":"2026-07-05T10:33:12.969378+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.13565v1","created_at":"2026-07-05T10:33:12.969378+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.13565","created_at":"2026-07-05T10:33:12.969378+00:00"},{"alias_kind":"pith_short_12","alias_value":"NEYVBA627KAN","created_at":"2026-07-05T10:33:12.969378+00:00"},{"alias_kind":"pith_short_16","alias_value":"NEYVBA627KANMD7A","created_at":"2026-07-05T10:33:12.969378+00:00"},{"alias_kind":"pith_short_8","alias_value":"NEYVBA62","created_at":"2026-07-05T10:33:12.969378+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.02011","citing_title":"Extreme Low-Bit Inference in Reasoning Models: Failure Modes and Targeted Recovery","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2511.22972","citing_title":"Training-Free Loosely Speculative Decoding: Accepting Semantically Correct Drafts Beyond Exact Match","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2604.14626","citing_title":"ELMoE-3D: Leveraging Intrinsic Elasticity of MoE for Hybrid-Bonding-Enabled Self-Speculative Decoding in On-Premises Serving","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NEYVBA627KANMD7AR7GTVDZY6D","json":"https://pith.science/pith/NEYVBA627KANMD7AR7GTVDZY6D.json","graph_json":"https://pith.science/api/pith-number/NEYVBA627KANMD7AR7GTVDZY6D/graph.json","events_json":"https://pith.science/api/pith-number/NEYVBA627KANMD7AR7GTVDZY6D/events.json","paper":"https://pith.science/paper/NEYVBA62"},"agent_actions":{"view_html":"https://pith.science/pith/NEYVBA627KANMD7AR7GTVDZY6D","download_json":"https://pith.science/pith/NEYVBA627KANMD7AR7GTVDZY6D.json","view_paper":"https://pith.science/paper/NEYVBA62","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.13565&json=true","fetch_graph":"https://pith.science/api/pith-number/NEYVBA627KANMD7AR7GTVDZY6D/graph.json","fetch_events":"https://pith.science/api/pith-number/NEYVBA627KANMD7AR7GTVDZY6D/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NEYVBA627KANMD7AR7GTVDZY6D/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NEYVBA627KANMD7AR7GTVDZY6D/action/storage_attestation","attest_author":"https://pith.science/pith/NEYVBA627KANMD7AR7GTVDZY6D/action/author_attestation","sign_citation":"https://pith.science/pith/NEYVBA627KANMD7AR7GTVDZY6D/action/citation_signature","submit_replication":"https://pith.science/pith/NEYVBA627KANMD7AR7GTVDZY6D/action/replication_record"}},"created_at":"2026-07-05T10:33:12.969378+00:00","updated_at":"2026-07-05T10:33:12.969378+00:00"}