{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:XZJWWCMBWS463UBGUV763ZPBDY","short_pith_number":"pith:XZJWWCMB","schema_version":"1.0","canonical_sha256":"be536b0981b4b9edd026a57fede5e11e03e215cbc61639b928a290fcea30eb94","source":{"kind":"arxiv","id":"2404.18911","version":1},"attestation_state":"computed","paper":{"title":"Kangaroo: Lossless Self-Speculative Decoding via Double Early Exiting","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Fangcheng Liu, Kai Han, Yehui Tang, Yunhe Wang, Yunsheng Ni, Zhenhua Liu","submitted_at":"2024-04-29T17:53:54Z","abstract_excerpt":"Speculative decoding has demonstrated its effectiveness in accelerating the inference of large language models while maintaining a consistent sampling distribution. However, the conventional approach of training a separate draft model to achieve a satisfactory token acceptance rate can be costly. Drawing inspiration from early exiting, we propose a novel self-speculative decoding framework \\emph{Kangaroo}, which uses a fixed shallow sub-network as a self-draft model, with the remaining layers serving as the larger target model. We train a lightweight and efficient adapter module on top of the "},"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":"2404.18911","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-04-29T17:53:54Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"5caf63f3e7723c3504eadfcb53c23a729a5464544b576b4cc2e2aae74da2ffe0","abstract_canon_sha256":"b5c53b263d2f3a65856e220b934f62f8d52191c2abc6a035d19f47a75d65bd1b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:13:22.275069Z","signature_b64":"pXbf9K1dH38U3Koh+GEqh3e7EtSp2VpYyHtn3lpp9dQxSQ9EBIHZNqlf0ipOBAS21i1gLZrkL5zU30rmu2DwAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"be536b0981b4b9edd026a57fede5e11e03e215cbc61639b928a290fcea30eb94","last_reissued_at":"2026-07-05T08:13:22.274625Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:13:22.274625Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Kangaroo: Lossless Self-Speculative Decoding via Double Early Exiting","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Fangcheng Liu, Kai Han, Yehui Tang, Yunhe Wang, Yunsheng Ni, Zhenhua Liu","submitted_at":"2024-04-29T17:53:54Z","abstract_excerpt":"Speculative decoding has demonstrated its effectiveness in accelerating the inference of large language models while maintaining a consistent sampling distribution. However, the conventional approach of training a separate draft model to achieve a satisfactory token acceptance rate can be costly. Drawing inspiration from early exiting, we propose a novel self-speculative decoding framework \\emph{Kangaroo}, which uses a fixed shallow sub-network as a self-draft model, with the remaining layers serving as the larger target model. We train a lightweight and efficient adapter module on top of the "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.18911","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/2404.18911/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":"2404.18911","created_at":"2026-07-05T08:13:22.274683+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.18911v1","created_at":"2026-07-05T08:13:22.274683+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.18911","created_at":"2026-07-05T08:13:22.274683+00:00"},{"alias_kind":"pith_short_12","alias_value":"XZJWWCMBWS46","created_at":"2026-07-05T08:13:22.274683+00:00"},{"alias_kind":"pith_short_16","alias_value":"XZJWWCMBWS463UBG","created_at":"2026-07-05T08:13:22.274683+00:00"},{"alias_kind":"pith_short_8","alias_value":"XZJWWCMB","created_at":"2026-07-05T08:13:22.274683+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.00535","citing_title":"DREAM-S: Speculative Decoding with Searchable Drafting and Target-Aware Refinement for Multimodal Generation","ref_index":103,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07243","citing_title":"SpecBlock: Block-Iterative Speculative Decoding with Dynamic Tree Drafting","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2404.14294","citing_title":"A Survey on Efficient Inference for Large Language Models","ref_index":249,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18592","citing_title":"Two-dimensional early exit optimisation of LLM inference","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07243","citing_title":"SpecBlock: Block-Iterative Speculative Decoding with Dynamic Tree Drafting","ref_index":13,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XZJWWCMBWS463UBGUV763ZPBDY","json":"https://pith.science/pith/XZJWWCMBWS463UBGUV763ZPBDY.json","graph_json":"https://pith.science/api/pith-number/XZJWWCMBWS463UBGUV763ZPBDY/graph.json","events_json":"https://pith.science/api/pith-number/XZJWWCMBWS463UBGUV763ZPBDY/events.json","paper":"https://pith.science/paper/XZJWWCMB"},"agent_actions":{"view_html":"https://pith.science/pith/XZJWWCMBWS463UBGUV763ZPBDY","download_json":"https://pith.science/pith/XZJWWCMBWS463UBGUV763ZPBDY.json","view_paper":"https://pith.science/paper/XZJWWCMB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.18911&json=true","fetch_graph":"https://pith.science/api/pith-number/XZJWWCMBWS463UBGUV763ZPBDY/graph.json","fetch_events":"https://pith.science/api/pith-number/XZJWWCMBWS463UBGUV763ZPBDY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XZJWWCMBWS463UBGUV763ZPBDY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XZJWWCMBWS463UBGUV763ZPBDY/action/storage_attestation","attest_author":"https://pith.science/pith/XZJWWCMBWS463UBGUV763ZPBDY/action/author_attestation","sign_citation":"https://pith.science/pith/XZJWWCMBWS463UBGUV763ZPBDY/action/citation_signature","submit_replication":"https://pith.science/pith/XZJWWCMBWS463UBGUV763ZPBDY/action/replication_record"}},"created_at":"2026-07-05T08:13:22.274683+00:00","updated_at":"2026-07-05T08:13:22.274683+00:00"}