{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:JZP5NTJIZSQZXIF5MUJFFJAAUQ","short_pith_number":"pith:JZP5NTJI","schema_version":"1.0","canonical_sha256":"4e5fd6cd28cca19ba0bd651252a400a414f9906ec70ba2848bb8cbd8fc8c10a0","source":{"kind":"arxiv","id":"2104.08803","version":2},"attestation_state":"computed","paper":{"title":"Consistent Accelerated Inference via Confident Adaptive Transformers","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Adam Fisch, Regina Barzilay, Tal Schuster, Tommi Jaakkola","submitted_at":"2021-04-18T10:22:28Z","abstract_excerpt":"We develop a novel approach for confidently accelerating inference in the large and expensive multilayer Transformers that are now ubiquitous in natural language processing (NLP). Amortized or approximate computational methods increase efficiency, but can come with unpredictable performance costs. In this work, we present CATs -- Confident Adaptive Transformers -- in which we simultaneously increase computational efficiency, while guaranteeing a specifiable degree of consistency with the original model with high confidence. Our method trains additional prediction heads on top of intermediate l"},"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":"2104.08803","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-04-18T10:22:28Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"44b08660b0962e0859bad74efd93bfbc36c8f30e42ed6be28f03d4dd972b53d5","abstract_canon_sha256":"86067fec639301356880fea47103d15141cf7708e444c916ffd127c4f09f3875"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:12:48.367026Z","signature_b64":"7ejpQW5n0QBXuQv9g/0KLr1SJPUgaFt03ZIhnFHk2QL6dl7wTnbUwICaAvxoyia/JVDdZlLcimlwAJ0XpVwjBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4e5fd6cd28cca19ba0bd651252a400a414f9906ec70ba2848bb8cbd8fc8c10a0","last_reissued_at":"2026-07-05T03:12:48.366274Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:12:48.366274Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Consistent Accelerated Inference via Confident Adaptive Transformers","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Adam Fisch, Regina Barzilay, Tal Schuster, Tommi Jaakkola","submitted_at":"2021-04-18T10:22:28Z","abstract_excerpt":"We develop a novel approach for confidently accelerating inference in the large and expensive multilayer Transformers that are now ubiquitous in natural language processing (NLP). Amortized or approximate computational methods increase efficiency, but can come with unpredictable performance costs. In this work, we present CATs -- Confident Adaptive Transformers -- in which we simultaneously increase computational efficiency, while guaranteeing a specifiable degree of consistency with the original model with high confidence. Our method trains additional prediction heads on top of intermediate l"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2104.08803","kind":"arxiv","version":2},"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/2104.08803/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":"2104.08803","created_at":"2026-07-05T03:12:48.366577+00:00"},{"alias_kind":"arxiv_version","alias_value":"2104.08803v2","created_at":"2026-07-05T03:12:48.366577+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2104.08803","created_at":"2026-07-05T03:12:48.366577+00:00"},{"alias_kind":"pith_short_12","alias_value":"JZP5NTJIZSQZ","created_at":"2026-07-05T03:12:48.366577+00:00"},{"alias_kind":"pith_short_16","alias_value":"JZP5NTJIZSQZXIF5","created_at":"2026-07-05T03:12:48.366577+00:00"},{"alias_kind":"pith_short_8","alias_value":"JZP5NTJI","created_at":"2026-07-05T03:12:48.366577+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.18592","citing_title":"Two-dimensional early exit optimisation of LLM inference","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2604.01502","citing_title":"Conformal Risk Control under Non-Monotone Losses: Theory and Finite-Sample Guarantees","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17286","citing_title":"Depth Adaptive Efficient Visual Autoregressive Modeling","ref_index":46,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JZP5NTJIZSQZXIF5MUJFFJAAUQ","json":"https://pith.science/pith/JZP5NTJIZSQZXIF5MUJFFJAAUQ.json","graph_json":"https://pith.science/api/pith-number/JZP5NTJIZSQZXIF5MUJFFJAAUQ/graph.json","events_json":"https://pith.science/api/pith-number/JZP5NTJIZSQZXIF5MUJFFJAAUQ/events.json","paper":"https://pith.science/paper/JZP5NTJI"},"agent_actions":{"view_html":"https://pith.science/pith/JZP5NTJIZSQZXIF5MUJFFJAAUQ","download_json":"https://pith.science/pith/JZP5NTJIZSQZXIF5MUJFFJAAUQ.json","view_paper":"https://pith.science/paper/JZP5NTJI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2104.08803&json=true","fetch_graph":"https://pith.science/api/pith-number/JZP5NTJIZSQZXIF5MUJFFJAAUQ/graph.json","fetch_events":"https://pith.science/api/pith-number/JZP5NTJIZSQZXIF5MUJFFJAAUQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JZP5NTJIZSQZXIF5MUJFFJAAUQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JZP5NTJIZSQZXIF5MUJFFJAAUQ/action/storage_attestation","attest_author":"https://pith.science/pith/JZP5NTJIZSQZXIF5MUJFFJAAUQ/action/author_attestation","sign_citation":"https://pith.science/pith/JZP5NTJIZSQZXIF5MUJFFJAAUQ/action/citation_signature","submit_replication":"https://pith.science/pith/JZP5NTJIZSQZXIF5MUJFFJAAUQ/action/replication_record"}},"created_at":"2026-07-05T03:12:48.366577+00:00","updated_at":"2026-07-05T03:12:48.366577+00:00"}