{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:HFWIRGU25HV6VQ54VLJD37DQ44","short_pith_number":"pith:HFWIRGU2","schema_version":"1.0","canonical_sha256":"396c889a9ae9ebeac3bcaad23dfc70e73551a3a0e6e56c3cf614ece36fd6f758","source":{"kind":"arxiv","id":"2407.09722","version":4},"attestation_state":"computed","paper":{"title":"Optimized Multi-Token Joint Decoding with Auxiliary Model for LLM Inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Jason Cong, Neha Prakriya, Yizhou Sun, Zifan He, Ziniu Hu, Zongyue Qin","submitted_at":"2024-07-12T23:29:54Z","abstract_excerpt":"Large language models (LLMs) have achieved remarkable success across diverse tasks, yet their inference processes are hindered by substantial time and energy demands due to single-token generation at each decoding step. While previous methods such as speculative decoding mitigate these inefficiencies by producing multiple tokens per step, each token is still generated by its single-token distribution, thereby enhancing speed without improving effectiveness. In contrast, our work simultaneously enhances inference speed and improves the output effectiveness. We consider multi-token joint decodin"},"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":"2407.09722","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-07-12T23:29:54Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"4f30f4f5b2caed8e0e7f93c48f7cac6f5645ef4e28b6064a4cfcfef182ae8d1c","abstract_canon_sha256":"6c42310b5341f59a4134af0d3aabecfe82e9aa6828185e6d3780f0142fe50df7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:46:55.841418Z","signature_b64":"SYl+TyuAVISvWnKnl5P8zxf9Du6bTR+cTKSDQBg17IWHRAUYOLR3AUa/7JGrP1c3aDDEmF9BVEur8nhg7rJgDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"396c889a9ae9ebeac3bcaad23dfc70e73551a3a0e6e56c3cf614ece36fd6f758","last_reissued_at":"2026-07-05T10:46:55.840915Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:46:55.840915Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Optimized Multi-Token Joint Decoding with Auxiliary Model for LLM Inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Jason Cong, Neha Prakriya, Yizhou Sun, Zifan He, Ziniu Hu, Zongyue Qin","submitted_at":"2024-07-12T23:29:54Z","abstract_excerpt":"Large language models (LLMs) have achieved remarkable success across diverse tasks, yet their inference processes are hindered by substantial time and energy demands due to single-token generation at each decoding step. While previous methods such as speculative decoding mitigate these inefficiencies by producing multiple tokens per step, each token is still generated by its single-token distribution, thereby enhancing speed without improving effectiveness. In contrast, our work simultaneously enhances inference speed and improves the output effectiveness. We consider multi-token joint decodin"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.09722","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/2407.09722/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":"2407.09722","created_at":"2026-07-05T10:46:55.840973+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.09722v4","created_at":"2026-07-05T10:46:55.840973+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.09722","created_at":"2026-07-05T10:46:55.840973+00:00"},{"alias_kind":"pith_short_12","alias_value":"HFWIRGU25HV6","created_at":"2026-07-05T10:46:55.840973+00:00"},{"alias_kind":"pith_short_16","alias_value":"HFWIRGU25HV6VQ54","created_at":"2026-07-05T10:46:55.840973+00:00"},{"alias_kind":"pith_short_8","alias_value":"HFWIRGU2","created_at":"2026-07-05T10:46:55.840973+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2603.08899","citing_title":"ConFu: Contemplate the Future for Better Speculative Sampling","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11186","citing_title":"CATS: Cascaded Adaptive Tree Speculation for Memory-Limited LLM Inference Acceleration","ref_index":43,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HFWIRGU25HV6VQ54VLJD37DQ44","json":"https://pith.science/pith/HFWIRGU25HV6VQ54VLJD37DQ44.json","graph_json":"https://pith.science/api/pith-number/HFWIRGU25HV6VQ54VLJD37DQ44/graph.json","events_json":"https://pith.science/api/pith-number/HFWIRGU25HV6VQ54VLJD37DQ44/events.json","paper":"https://pith.science/paper/HFWIRGU2"},"agent_actions":{"view_html":"https://pith.science/pith/HFWIRGU25HV6VQ54VLJD37DQ44","download_json":"https://pith.science/pith/HFWIRGU25HV6VQ54VLJD37DQ44.json","view_paper":"https://pith.science/paper/HFWIRGU2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.09722&json=true","fetch_graph":"https://pith.science/api/pith-number/HFWIRGU25HV6VQ54VLJD37DQ44/graph.json","fetch_events":"https://pith.science/api/pith-number/HFWIRGU25HV6VQ54VLJD37DQ44/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HFWIRGU25HV6VQ54VLJD37DQ44/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HFWIRGU25HV6VQ54VLJD37DQ44/action/storage_attestation","attest_author":"https://pith.science/pith/HFWIRGU25HV6VQ54VLJD37DQ44/action/author_attestation","sign_citation":"https://pith.science/pith/HFWIRGU25HV6VQ54VLJD37DQ44/action/citation_signature","submit_replication":"https://pith.science/pith/HFWIRGU25HV6VQ54VLJD37DQ44/action/replication_record"}},"created_at":"2026-07-05T10:46:55.840973+00:00","updated_at":"2026-07-05T10:46:55.840973+00:00"}