{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:U6XM5DADF5WW2F6MKTUYAQHGCZ","short_pith_number":"pith:U6XM5DAD","schema_version":"1.0","canonical_sha256":"a7aece8c032f6d6d17cc54e98040e6164b494fe1aaa1a1d25f9aa13c18981c0f","source":{"kind":"arxiv","id":"2408.11049","version":5},"attestation_state":"computed","paper":{"title":"MagicDec: Breaking the Latency-Throughput Tradeoff for Long Context Generation with Speculative Decoding","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Avner May, Beidi Chen, Ian En-Hsu Yen, Jian Chen, Jinyuan Shi, Ranajoy Sadhukhan, Ruihang Lai, Tianqi Chen, Vashisth Tiwari, Zhuoming Chen","submitted_at":"2024-08-20T17:57:31Z","abstract_excerpt":"Large Language Models (LLMs) have become more prevalent in long-context applications such as interactive chatbots, document analysis, and agent workflows, but it is challenging to serve long-context requests with low latency and high throughput. Speculative decoding (SD) is a widely used technique to reduce latency losslessly, but the conventional wisdom suggests that its efficacy is limited to small batch sizes. In MagicDec, we show that surprisingly SD can achieve speedup even for a high throughput inference regime for moderate to long sequences. More interestingly, an intelligent drafting s"},"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":"2408.11049","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-08-20T17:57:31Z","cross_cats_sorted":[],"title_canon_sha256":"5f8ac418e5ccdbef5d13d5dd3cf972f76fcb8b6a1f9d16edde2a4f6ac9f29b84","abstract_canon_sha256":"904771b8cba403a7631fbef56cbf966d35507a91877d088b34bd9263117b68ff"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:43:00.854749Z","signature_b64":"1DmQoFuQxYJDwk7ncWaTY3tSQOa6SW7Vu+BAGM47O/zDGTQES6vo00OyNNw1x0lnUKZ8AxaPi3bv6BTpu9wQCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a7aece8c032f6d6d17cc54e98040e6164b494fe1aaa1a1d25f9aa13c18981c0f","last_reissued_at":"2026-07-05T10:43:00.854137Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:43:00.854137Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MagicDec: Breaking the Latency-Throughput Tradeoff for Long Context Generation with Speculative Decoding","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Avner May, Beidi Chen, Ian En-Hsu Yen, Jian Chen, Jinyuan Shi, Ranajoy Sadhukhan, Ruihang Lai, Tianqi Chen, Vashisth Tiwari, Zhuoming Chen","submitted_at":"2024-08-20T17:57:31Z","abstract_excerpt":"Large Language Models (LLMs) have become more prevalent in long-context applications such as interactive chatbots, document analysis, and agent workflows, but it is challenging to serve long-context requests with low latency and high throughput. Speculative decoding (SD) is a widely used technique to reduce latency losslessly, but the conventional wisdom suggests that its efficacy is limited to small batch sizes. In MagicDec, we show that surprisingly SD can achieve speedup even for a high throughput inference regime for moderate to long sequences. More interestingly, an intelligent drafting s"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.11049","kind":"arxiv","version":5},"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/2408.11049/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":"2408.11049","created_at":"2026-07-05T10:43:00.854211+00:00"},{"alias_kind":"arxiv_version","alias_value":"2408.11049v5","created_at":"2026-07-05T10:43:00.854211+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.11049","created_at":"2026-07-05T10:43:00.854211+00:00"},{"alias_kind":"pith_short_12","alias_value":"U6XM5DADF5WW","created_at":"2026-07-05T10:43:00.854211+00:00"},{"alias_kind":"pith_short_16","alias_value":"U6XM5DADF5WW2F6M","created_at":"2026-07-05T10:43:00.854211+00:00"},{"alias_kind":"pith_short_8","alias_value":"U6XM5DAD","created_at":"2026-07-05T10:43:00.854211+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.24957","citing_title":"Dustin: Draft-Augmented Sparse Verification for Efficient Long-Context Generation with Speculative Decoding","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26558","citing_title":"Cassandra: Enabling Reasoning LLMs at Edge via Self-Speculative Decoding","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29727","citing_title":"Bastion: Budget-Aware Speculative Decoding with Tree-structured Block Diffusion Drafting","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20104","citing_title":"Draft Less, Retrieve More: Hybrid Tree Construction for Speculative Decoding","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17613","citing_title":"VeriCache: Turning Lossy KV Cache into Lossless LLM Inference","ref_index":57,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/U6XM5DADF5WW2F6MKTUYAQHGCZ","json":"https://pith.science/pith/U6XM5DADF5WW2F6MKTUYAQHGCZ.json","graph_json":"https://pith.science/api/pith-number/U6XM5DADF5WW2F6MKTUYAQHGCZ/graph.json","events_json":"https://pith.science/api/pith-number/U6XM5DADF5WW2F6MKTUYAQHGCZ/events.json","paper":"https://pith.science/paper/U6XM5DAD"},"agent_actions":{"view_html":"https://pith.science/pith/U6XM5DADF5WW2F6MKTUYAQHGCZ","download_json":"https://pith.science/pith/U6XM5DADF5WW2F6MKTUYAQHGCZ.json","view_paper":"https://pith.science/paper/U6XM5DAD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2408.11049&json=true","fetch_graph":"https://pith.science/api/pith-number/U6XM5DADF5WW2F6MKTUYAQHGCZ/graph.json","fetch_events":"https://pith.science/api/pith-number/U6XM5DADF5WW2F6MKTUYAQHGCZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/U6XM5DADF5WW2F6MKTUYAQHGCZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/U6XM5DADF5WW2F6MKTUYAQHGCZ/action/storage_attestation","attest_author":"https://pith.science/pith/U6XM5DADF5WW2F6MKTUYAQHGCZ/action/author_attestation","sign_citation":"https://pith.science/pith/U6XM5DADF5WW2F6MKTUYAQHGCZ/action/citation_signature","submit_replication":"https://pith.science/pith/U6XM5DADF5WW2F6MKTUYAQHGCZ/action/replication_record"}},"created_at":"2026-07-05T10:43:00.854211+00:00","updated_at":"2026-07-05T10:43:00.854211+00:00"}