{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:DYHHTR7PCPDCBJUUJTHGMRQFYN","short_pith_number":"pith:DYHHTR7P","schema_version":"1.0","canonical_sha256":"1e0e79c7ef13c620a6944cce664605c379e2131a0a08c358dd2bc936e3af19bd","source":{"kind":"arxiv","id":"2408.08146","version":1},"attestation_state":"computed","paper":{"title":"KOALA: Enhancing Speculative Decoding for LLM via Multi-Layer Draft Heads with Adversarial Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Jing Zhao, Kaiqi Zhang, Rui Chen","submitted_at":"2024-08-15T13:29:48Z","abstract_excerpt":"Large Language Models (LLMs) exhibit high inference latency due to their autoregressive decoding nature. While the draft head in speculative decoding mitigates this issue, its full potential remains unexplored. In this paper, we introduce KOALA (K-layer Optimized Adversarial Learning Architecture), an orthogonal approach to the draft head. By transforming the conventional single-layer draft head into a multi-layer architecture and incorporating adversarial learning into the traditional supervised training, KOALA significantly improves the accuracy of the draft head in predicting subsequent tok"},"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":"2408.08146","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-08-15T13:29:48Z","cross_cats_sorted":[],"title_canon_sha256":"cc8306f8d9df71c47889c7026ae4b7e188332cfdbc0562187f7d2caa0b429381","abstract_canon_sha256":"7efd15619dc8eae65140963f5363a1e0630303af710853cb7f6b3f07b78cf48b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:55:45.058597Z","signature_b64":"uIu2grv8njvFUgEL6bl5ZO3hEVNy/PoQqJlUHCoHntZ3KHs5LqAHADrFlNWYXcsiDShHihOAG92FnNbkH3PqDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1e0e79c7ef13c620a6944cce664605c379e2131a0a08c358dd2bc936e3af19bd","last_reissued_at":"2026-07-05T08:55:45.058219Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:55:45.058219Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"KOALA: Enhancing Speculative Decoding for LLM via Multi-Layer Draft Heads with Adversarial Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Jing Zhao, Kaiqi Zhang, Rui Chen","submitted_at":"2024-08-15T13:29:48Z","abstract_excerpt":"Large Language Models (LLMs) exhibit high inference latency due to their autoregressive decoding nature. While the draft head in speculative decoding mitigates this issue, its full potential remains unexplored. In this paper, we introduce KOALA (K-layer Optimized Adversarial Learning Architecture), an orthogonal approach to the draft head. By transforming the conventional single-layer draft head into a multi-layer architecture and incorporating adversarial learning into the traditional supervised training, KOALA significantly improves the accuracy of the draft head in predicting subsequent tok"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.08146","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/2408.08146/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":"2408.08146","created_at":"2026-07-05T08:55:45.058276+00:00"},{"alias_kind":"arxiv_version","alias_value":"2408.08146v1","created_at":"2026-07-05T08:55:45.058276+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.08146","created_at":"2026-07-05T08:55:45.058276+00:00"},{"alias_kind":"pith_short_12","alias_value":"DYHHTR7PCPDC","created_at":"2026-07-05T08:55:45.058276+00:00"},{"alias_kind":"pith_short_16","alias_value":"DYHHTR7PCPDCBJUU","created_at":"2026-07-05T08:55:45.058276+00:00"},{"alias_kind":"pith_short_8","alias_value":"DYHHTR7P","created_at":"2026-07-05T08:55:45.058276+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.20022","citing_title":"FlexDraft: Flexible Speculative Decoding via Attention Tuning and Bonus-Guided Calibration","ref_index":36,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DYHHTR7PCPDCBJUUJTHGMRQFYN","json":"https://pith.science/pith/DYHHTR7PCPDCBJUUJTHGMRQFYN.json","graph_json":"https://pith.science/api/pith-number/DYHHTR7PCPDCBJUUJTHGMRQFYN/graph.json","events_json":"https://pith.science/api/pith-number/DYHHTR7PCPDCBJUUJTHGMRQFYN/events.json","paper":"https://pith.science/paper/DYHHTR7P"},"agent_actions":{"view_html":"https://pith.science/pith/DYHHTR7PCPDCBJUUJTHGMRQFYN","download_json":"https://pith.science/pith/DYHHTR7PCPDCBJUUJTHGMRQFYN.json","view_paper":"https://pith.science/paper/DYHHTR7P","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2408.08146&json=true","fetch_graph":"https://pith.science/api/pith-number/DYHHTR7PCPDCBJUUJTHGMRQFYN/graph.json","fetch_events":"https://pith.science/api/pith-number/DYHHTR7PCPDCBJUUJTHGMRQFYN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DYHHTR7PCPDCBJUUJTHGMRQFYN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DYHHTR7PCPDCBJUUJTHGMRQFYN/action/storage_attestation","attest_author":"https://pith.science/pith/DYHHTR7PCPDCBJUUJTHGMRQFYN/action/author_attestation","sign_citation":"https://pith.science/pith/DYHHTR7PCPDCBJUUJTHGMRQFYN/action/citation_signature","submit_replication":"https://pith.science/pith/DYHHTR7PCPDCBJUUJTHGMRQFYN/action/replication_record"}},"created_at":"2026-07-05T08:55:45.058276+00:00","updated_at":"2026-07-05T08:55:45.058276+00:00"}