{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:IAGFAQZV46ROOYNYHWSQ5F3NAQ","short_pith_number":"pith:IAGFAQZV","schema_version":"1.0","canonical_sha256":"400c504335e7a2e761b83da50e976d043e5ae22da44368d0614e095c8130aea0","source":{"kind":"arxiv","id":"2403.00858","version":4},"attestation_state":"computed","paper":{"title":"Direct Alignment of Draft Model for Speculative Decoding with Chat-Fine-Tuned LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Christopher Lott, Junyoung Park, Mingu Lee, Mukul Gagrani, Raghavv Goel, Wonseok Jeon","submitted_at":"2024-02-29T19:55:06Z","abstract_excerpt":"Text generation with Large Language Models (LLMs) is known to be memory bound due to the combination of their auto-regressive nature, huge parameter counts, and limited memory bandwidths, often resulting in low token rates. Speculative decoding has been proposed as a solution for LLM inference acceleration. However, since draft models are often unavailable in the modern open-source LLM families, e.g., for Llama 2 7B, training a high-quality draft model is required to enable inference acceleration via speculative decoding. In this paper, we propose a simple draft model training framework for di"},"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":"2403.00858","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-29T19:55:06Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"e8e8287f166040ac5f6e090408eb7d7f5cd166095701aeab6d8298e28a5f914a","abstract_canon_sha256":"14bff0ccecb966c1a0ad76f144b944b5e2518f5e0152d672ab61e4667cc97e56"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:18:50.268171Z","signature_b64":"jjmUGQEcBE6Ubsgx1chZSyBtenpK3VwbsGnFRK9St2fVmXrqzPjNyy4rHZ+dRfsblTt65WY9k8DXEiEad8KsAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"400c504335e7a2e761b83da50e976d043e5ae22da44368d0614e095c8130aea0","last_reissued_at":"2026-07-05T08:18:50.267608Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:18:50.267608Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Direct Alignment of Draft Model for Speculative Decoding with Chat-Fine-Tuned LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Christopher Lott, Junyoung Park, Mingu Lee, Mukul Gagrani, Raghavv Goel, Wonseok Jeon","submitted_at":"2024-02-29T19:55:06Z","abstract_excerpt":"Text generation with Large Language Models (LLMs) is known to be memory bound due to the combination of their auto-regressive nature, huge parameter counts, and limited memory bandwidths, often resulting in low token rates. Speculative decoding has been proposed as a solution for LLM inference acceleration. However, since draft models are often unavailable in the modern open-source LLM families, e.g., for Llama 2 7B, training a high-quality draft model is required to enable inference acceleration via speculative decoding. In this paper, we propose a simple draft model training framework for di"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.00858","kind":"arxiv","version":4},"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/2403.00858/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":"2403.00858","created_at":"2026-07-05T08:18:50.267673+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.00858v4","created_at":"2026-07-05T08:18:50.267673+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.00858","created_at":"2026-07-05T08:18:50.267673+00:00"},{"alias_kind":"pith_short_12","alias_value":"IAGFAQZV46RO","created_at":"2026-07-05T08:18:50.267673+00:00"},{"alias_kind":"pith_short_16","alias_value":"IAGFAQZV46ROOYNY","created_at":"2026-07-05T08:18:50.267673+00:00"},{"alias_kind":"pith_short_8","alias_value":"IAGFAQZV","created_at":"2026-07-05T08:18:50.267673+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25097","citing_title":"Speculative Decoding at Temperature Zero: A Scoped Safety-Invariance Screen with a 48,072-Sample Expansion","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2606.18394","citing_title":"JetSpec: Breaking the Scaling Ceiling of Speculative Decoding with Parallel Tree Drafting","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2603.08899","citing_title":"ConFu: Contemplate the Future for Better Speculative Sampling","ref_index":4,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IAGFAQZV46ROOYNYHWSQ5F3NAQ","json":"https://pith.science/pith/IAGFAQZV46ROOYNYHWSQ5F3NAQ.json","graph_json":"https://pith.science/api/pith-number/IAGFAQZV46ROOYNYHWSQ5F3NAQ/graph.json","events_json":"https://pith.science/api/pith-number/IAGFAQZV46ROOYNYHWSQ5F3NAQ/events.json","paper":"https://pith.science/paper/IAGFAQZV"},"agent_actions":{"view_html":"https://pith.science/pith/IAGFAQZV46ROOYNYHWSQ5F3NAQ","download_json":"https://pith.science/pith/IAGFAQZV46ROOYNYHWSQ5F3NAQ.json","view_paper":"https://pith.science/paper/IAGFAQZV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.00858&json=true","fetch_graph":"https://pith.science/api/pith-number/IAGFAQZV46ROOYNYHWSQ5F3NAQ/graph.json","fetch_events":"https://pith.science/api/pith-number/IAGFAQZV46ROOYNYHWSQ5F3NAQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IAGFAQZV46ROOYNYHWSQ5F3NAQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IAGFAQZV46ROOYNYHWSQ5F3NAQ/action/storage_attestation","attest_author":"https://pith.science/pith/IAGFAQZV46ROOYNYHWSQ5F3NAQ/action/author_attestation","sign_citation":"https://pith.science/pith/IAGFAQZV46ROOYNYHWSQ5F3NAQ/action/citation_signature","submit_replication":"https://pith.science/pith/IAGFAQZV46ROOYNYHWSQ5F3NAQ/action/replication_record"}},"created_at":"2026-07-05T08:18:50.267673+00:00","updated_at":"2026-07-05T08:18:50.267673+00:00"}