{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:YFLR7FACMBKQHTJ3IFF5JIYKAH","short_pith_number":"pith:YFLR7FAC","canonical_record":{"source":{"id":"2401.15077","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-01-26T18:59:01Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"1abf05061a67129137aebd8e455aefe112aec094cec7643e2ee0c38c3085bd2a","abstract_canon_sha256":"7b4d4b7b129dde05beaacea123d7de51ce50bb91329608596258a28144d06f98"},"schema_version":"1.0"},"canonical_sha256":"c1571f9402605503cd3b414bd4a30a01fec6e10fe7eceaccb0df6b3038f776a5","source":{"kind":"arxiv","id":"2401.15077","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2401.15077","created_at":"2026-05-17T23:39:19Z"},{"alias_kind":"arxiv_version","alias_value":"2401.15077v3","created_at":"2026-05-17T23:39:19Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.15077","created_at":"2026-05-17T23:39:19Z"},{"alias_kind":"pith_short_12","alias_value":"YFLR7FACMBKQ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"YFLR7FACMBKQHTJ3","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"YFLR7FAC","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:YFLR7FACMBKQHTJ3IFF5JIYKAH","target":"record","payload":{"canonical_record":{"source":{"id":"2401.15077","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-01-26T18:59:01Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"1abf05061a67129137aebd8e455aefe112aec094cec7643e2ee0c38c3085bd2a","abstract_canon_sha256":"7b4d4b7b129dde05beaacea123d7de51ce50bb91329608596258a28144d06f98"},"schema_version":"1.0"},"canonical_sha256":"c1571f9402605503cd3b414bd4a30a01fec6e10fe7eceaccb0df6b3038f776a5","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:19.746943Z","signature_b64":"b04fUVmxJViBaieXwSTFfNXJG2TdUdP3gyphgN51LQuAsGZOPHK6X6X7LVnQqjIO7onE/q2vRyKLMkn5TtxZBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c1571f9402605503cd3b414bd4a30a01fec6e10fe7eceaccb0df6b3038f776a5","last_reissued_at":"2026-05-17T23:39:19.746254Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:19.746254Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2401.15077","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:19Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"k69c7wRtERc+3OiSRubo+t22PSzK+KUlXf1goYKHrTOL0HfnNoY6XJS09xihKvgLYc7opVY0Q08u3pw4+tQABA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-30T17:46:50.374678Z"},"content_sha256":"31a0624dc37aef0d713747d15080fd7b95c67121e1c34debac689fc95ca33551","schema_version":"1.0","event_id":"sha256:31a0624dc37aef0d713747d15080fd7b95c67121e1c34debac689fc95ca33551"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:YFLR7FACMBKQHTJ3IFF5JIYKAH","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Advancing the token sequence by one step resolves uncertainty in second-to-top-layer features, enabling precise and low-overhead speculative sampling for LLMs.","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Chao Zhang, Fangyun Wei, Hongyang Zhang, Yuhui Li","submitted_at":"2024-01-26T18:59:01Z","abstract_excerpt":"Autoregressive decoding makes the inference of Large Language Models (LLMs) time-consuming. In this paper, we reconsider speculative sampling and derive two key observations. Firstly, autoregression at the feature (second-to-top-layer) level is more straightforward than at the token level. Secondly, the inherent uncertainty in feature (second-to-top-layer) level autoregression constrains its performance. Based on these insights, we introduce EAGLE (Extrapolation Algorithm for Greater Language-model Efficiency), a simple yet highly efficient speculative sampling framework. By incorporating a to"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"By incorporating a token sequence advanced by one time step, EAGLE effectively resolves the uncertainty, enabling precise second-to-top-layer feature prediction with minimal overhead.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That advancing the token sequence by exactly one step is sufficient to remove the inherent uncertainty in second-to-top-layer autoregression without introducing new distribution shift or verification errors.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"EAGLE resolves feature-level uncertainty in speculative sampling via one-step token advancement, delivering 2.7x-3.5x speedup on LLaMA2-Chat 70B and doubled throughput across multiple model families and tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Advancing the token sequence by one step resolves uncertainty in second-to-top-layer features, enabling precise and low-overhead speculative sampling for LLMs.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"465342844f719adde0e271b90e183d09b8d36cc76363488f074d7c08b901581c"},"source":{"id":"2401.15077","kind":"arxiv","version":3},"verdict":{"id":"02c4d278-fb24-4c5c-9b88-4bfc7beb7a52","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T00:11:25.391789Z","strongest_claim":"By incorporating a token sequence advanced by one time step, EAGLE effectively resolves the uncertainty, enabling precise second-to-top-layer feature prediction with minimal overhead.","one_line_summary":"EAGLE resolves feature-level uncertainty in speculative sampling via one-step token advancement, delivering 2.7x-3.5x speedup on LLaMA2-Chat 70B and doubled throughput across multiple model families and tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That advancing the token sequence by exactly one step is sufficient to remove the inherent uncertainty in second-to-top-layer autoregression without introducing new distribution shift or verification errors.","pith_extraction_headline":"Advancing the token sequence by one step resolves uncertainty in second-to-top-layer features, enabling precise and low-overhead speculative sampling for LLMs."},"references":{"count":66,"sample":[{"doi":"","year":null,"title":"journal of machine learning research , volume=","work_id":"3105c574-6648-471f-a395-fa9f95ce318c","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"International Conference on Machine Learning , pages=","work_id":"1a650b5b-a768-4080-8629-3f3f3fe0d908","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Miao, Xupeng and Oliaro, Gabriele and Zhang, Zhihao and Cheng, Xinhao and Wang, Zeyu and Wong, Rae Ying Yee and Chen, Zhuoming and Arfeen, Daiyaan and Abhyankar, Reyna and Jia, Zhihao , journal=","work_id":"3fd842e7-4cb4-4931-95ce-e32771cbffa5","ref_index":6,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Accelerating","work_id":"755e0f06-7758-4131-9f03-369a594b552d","ref_index":7,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Cascade Speculative Drafting for Even Faster","work_id":"823e0e4e-681d-4a9a-bef7-bc83a421eee4","ref_index":8,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":66,"snapshot_sha256":"d5b270af4ce41455f0375fdc05c7ade37a190435712428b1e1ac4dac6b50d241","internal_anchors":12},"formal_canon":{"evidence_count":2,"snapshot_sha256":"5d05effd675914170bb8ab242748277c5892e91aea88c6d9554f6bd37a38534b"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"02c4d278-fb24-4c5c-9b88-4bfc7beb7a52"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:19Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"um/AYSbTeNhRvHKj0uoGzfgruq1bT7NsFHcd53N5pCSvNUmcqWWnP/jAwJkQY4T60u0cpJTstQhaDxZrVP89BA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-30T17:46:50.375658Z"},"content_sha256":"fd6e415fe76cab1734370fded048d12dfb3bbed75eb482e5b9ff8417b0669c00","schema_version":"1.0","event_id":"sha256:fd6e415fe76cab1734370fded048d12dfb3bbed75eb482e5b9ff8417b0669c00"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/YFLR7FACMBKQHTJ3IFF5JIYKAH/bundle.json","state_url":"https://pith.science/pith/YFLR7FACMBKQHTJ3IFF5JIYKAH/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/YFLR7FACMBKQHTJ3IFF5JIYKAH/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-30T17:46:50Z","links":{"resolver":"https://pith.science/pith/YFLR7FACMBKQHTJ3IFF5JIYKAH","bundle":"https://pith.science/pith/YFLR7FACMBKQHTJ3IFF5JIYKAH/bundle.json","state":"https://pith.science/pith/YFLR7FACMBKQHTJ3IFF5JIYKAH/state.json","well_known_bundle":"https://pith.science/.well-known/pith/YFLR7FACMBKQHTJ3IFF5JIYKAH/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:YFLR7FACMBKQHTJ3IFF5JIYKAH","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"7b4d4b7b129dde05beaacea123d7de51ce50bb91329608596258a28144d06f98","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-01-26T18:59:01Z","title_canon_sha256":"1abf05061a67129137aebd8e455aefe112aec094cec7643e2ee0c38c3085bd2a"},"schema_version":"1.0","source":{"id":"2401.15077","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2401.15077","created_at":"2026-05-17T23:39:19Z"},{"alias_kind":"arxiv_version","alias_value":"2401.15077v3","created_at":"2026-05-17T23:39:19Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.15077","created_at":"2026-05-17T23:39:19Z"},{"alias_kind":"pith_short_12","alias_value":"YFLR7FACMBKQ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"YFLR7FACMBKQHTJ3","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"YFLR7FAC","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:fd6e415fe76cab1734370fded048d12dfb3bbed75eb482e5b9ff8417b0669c00","target":"graph","created_at":"2026-05-17T23:39:19Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"By incorporating a token sequence advanced by one time step, EAGLE effectively resolves the uncertainty, enabling precise second-to-top-layer feature prediction with minimal overhead."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That advancing the token sequence by exactly one step is sufficient to remove the inherent uncertainty in second-to-top-layer autoregression without introducing new distribution shift or verification errors."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"EAGLE resolves feature-level uncertainty in speculative sampling via one-step token advancement, delivering 2.7x-3.5x speedup on LLaMA2-Chat 70B and doubled throughput across multiple model families and tasks."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Advancing the token sequence by one step resolves uncertainty in second-to-top-layer features, enabling precise and low-overhead speculative sampling for LLMs."}],"snapshot_sha256":"465342844f719adde0e271b90e183d09b8d36cc76363488f074d7c08b901581c"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"5d05effd675914170bb8ab242748277c5892e91aea88c6d9554f6bd37a38534b"},"paper":{"abstract_excerpt":"Autoregressive decoding makes the inference of Large Language Models (LLMs) time-consuming. In this paper, we reconsider speculative sampling and derive two key observations. Firstly, autoregression at the feature (second-to-top-layer) level is more straightforward than at the token level. Secondly, the inherent uncertainty in feature (second-to-top-layer) level autoregression constrains its performance. Based on these insights, we introduce EAGLE (Extrapolation Algorithm for Greater Language-model Efficiency), a simple yet highly efficient speculative sampling framework. By incorporating a to","authors_text":"Chao Zhang, Fangyun Wei, Hongyang Zhang, Yuhui Li","cross_cats":["cs.CL"],"headline":"Advancing the token sequence by one step resolves uncertainty in second-to-top-layer features, enabling precise and low-overhead speculative sampling for LLMs.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-01-26T18:59:01Z","title":"EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty"},"references":{"count":66,"internal_anchors":12,"resolved_work":66,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"journal of machine learning research , volume=","work_id":"3105c574-6648-471f-a395-fa9f95ce318c","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"International Conference on Machine Learning , pages=","work_id":"1a650b5b-a768-4080-8629-3f3f3fe0d908","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":6,"title":"Miao, Xupeng and Oliaro, Gabriele and Zhang, Zhihao and Cheng, Xinhao and Wang, Zeyu and Wong, Rae Ying Yee and Chen, Zhuoming and Arfeen, Daiyaan and Abhyankar, Reyna and Jia, Zhihao , journal=","work_id":"3fd842e7-4cb4-4931-95ce-e32771cbffa5","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":7,"title":"Accelerating","work_id":"755e0f06-7758-4131-9f03-369a594b552d","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":8,"title":"Cascade Speculative Drafting for Even Faster","work_id":"823e0e4e-681d-4a9a-bef7-bc83a421eee4","year":null}],"snapshot_sha256":"d5b270af4ce41455f0375fdc05c7ade37a190435712428b1e1ac4dac6b50d241"},"source":{"id":"2401.15077","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-15T00:11:25.391789Z","id":"02c4d278-fb24-4c5c-9b88-4bfc7beb7a52","model_set":{"reader":"grok-4.3"},"one_line_summary":"EAGLE resolves feature-level uncertainty in speculative sampling via one-step token advancement, delivering 2.7x-3.5x speedup on LLaMA2-Chat 70B and doubled throughput across multiple model families and tasks.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Advancing the token sequence by one step resolves uncertainty in second-to-top-layer features, enabling precise and low-overhead speculative sampling for LLMs.","strongest_claim":"By incorporating a token sequence advanced by one time step, EAGLE effectively resolves the uncertainty, enabling precise second-to-top-layer feature prediction with minimal overhead.","weakest_assumption":"That advancing the token sequence by exactly one step is sufficient to remove the inherent uncertainty in second-to-top-layer autoregression without introducing new distribution shift or verification errors."}},"verdict_id":"02c4d278-fb24-4c5c-9b88-4bfc7beb7a52"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:31a0624dc37aef0d713747d15080fd7b95c67121e1c34debac689fc95ca33551","target":"record","created_at":"2026-05-17T23:39:19Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"7b4d4b7b129dde05beaacea123d7de51ce50bb91329608596258a28144d06f98","cross_cats_sorted":["cs.CL"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-01-26T18:59:01Z","title_canon_sha256":"1abf05061a67129137aebd8e455aefe112aec094cec7643e2ee0c38c3085bd2a"},"schema_version":"1.0","source":{"id":"2401.15077","kind":"arxiv","version":3}},"canonical_sha256":"c1571f9402605503cd3b414bd4a30a01fec6e10fe7eceaccb0df6b3038f776a5","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"c1571f9402605503cd3b414bd4a30a01fec6e10fe7eceaccb0df6b3038f776a5","first_computed_at":"2026-05-17T23:39:19.746254Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:19.746254Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"b04fUVmxJViBaieXwSTFfNXJG2TdUdP3gyphgN51LQuAsGZOPHK6X6X7LVnQqjIO7onE/q2vRyKLMkn5TtxZBw==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:19.746943Z","signed_message":"canonical_sha256_bytes"},"source_id":"2401.15077","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:31a0624dc37aef0d713747d15080fd7b95c67121e1c34debac689fc95ca33551","sha256:fd6e415fe76cab1734370fded048d12dfb3bbed75eb482e5b9ff8417b0669c00"],"state_sha256":"b19f3cf95374fd8f3023d9d7f33b77284f4ccef3b11a1a6940ce829a58c87f05"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Ba6D9ojvA9Z3TiyuQiwtdJnZRzDt+XWO+wkoV/qJQRLnw9+3opzIwO0l0CBHaViir9MGFKL0+esWWx+VeVVQDQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-30T17:46:50.379603Z","bundle_sha256":"1e06cb991ef4fbfdb6e8e5d762b26718c02f2acb66d03128f514ddeffe80387e"}}