{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:K62GKY6K4IYKIFQNPBUZA3VLGZ","short_pith_number":"pith:K62GKY6K","schema_version":"1.0","canonical_sha256":"57b46563cae230a4160d7869906eab364432bd1d2a16af6e3da869b5a0bfbfd6","source":{"kind":"arxiv","id":"2309.16354","version":2},"attestation_state":"computed","paper":{"title":"Transformer-VQ: Linear-Time Transformers via Vector Quantization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CV"],"primary_cat":"cs.LG","authors_text":"Lucas D. Lingle","submitted_at":"2023-09-28T11:26:52Z","abstract_excerpt":"We introduce Transformer-VQ, a decoder-only transformer computing softmax-based dense self-attention in linear time. Transformer-VQ's efficient attention is enabled by vector-quantized keys and a novel caching mechanism. In our large-scale experiments, Transformer-VQ is shown highly competitive in quality, obtaining 0.99 bpb on Enwik8, 26.6 ppl on PG-19, and 3.16 bpb on ImageNet64. In addition, the optimized implementation of Transformer-VQ is over 3x faster than a comparable quadratic-time transformer at sequence length 8k, is over 12x faster at 32k, and can scale to 131k with similar through"},"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":"2309.16354","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-09-28T11:26:52Z","cross_cats_sorted":["cs.CL","cs.CV"],"title_canon_sha256":"a8d70092b78c29ca6b0ed3cf62d981fe486a468abe3fdc262bacab2483b9de9b","abstract_canon_sha256":"e290c8228dce4a2eb84ec9e18c15140eac7a45079cbf2665bb66759d4055be93"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:48:48.403449Z","signature_b64":"t9bvmGWQjr0Zk2VtJAhFEO6VAS0vI9C7IuZ7iRkyvH4nSi5AWlOuDo3JvLWPBc+SgkNF2OTNHJcsYV7dgzF3Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"57b46563cae230a4160d7869906eab364432bd1d2a16af6e3da869b5a0bfbfd6","last_reissued_at":"2026-07-05T07:48:48.403001Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:48:48.403001Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Transformer-VQ: Linear-Time Transformers via Vector Quantization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CV"],"primary_cat":"cs.LG","authors_text":"Lucas D. Lingle","submitted_at":"2023-09-28T11:26:52Z","abstract_excerpt":"We introduce Transformer-VQ, a decoder-only transformer computing softmax-based dense self-attention in linear time. Transformer-VQ's efficient attention is enabled by vector-quantized keys and a novel caching mechanism. In our large-scale experiments, Transformer-VQ is shown highly competitive in quality, obtaining 0.99 bpb on Enwik8, 26.6 ppl on PG-19, and 3.16 bpb on ImageNet64. In addition, the optimized implementation of Transformer-VQ is over 3x faster than a comparable quadratic-time transformer at sequence length 8k, is over 12x faster at 32k, and can scale to 131k with similar through"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2309.16354","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/2309.16354/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":"2309.16354","created_at":"2026-07-05T07:48:48.403059+00:00"},{"alias_kind":"arxiv_version","alias_value":"2309.16354v2","created_at":"2026-07-05T07:48:48.403059+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2309.16354","created_at":"2026-07-05T07:48:48.403059+00:00"},{"alias_kind":"pith_short_12","alias_value":"K62GKY6K4IYK","created_at":"2026-07-05T07:48:48.403059+00:00"},{"alias_kind":"pith_short_16","alias_value":"K62GKY6K4IYKIFQN","created_at":"2026-07-05T07:48:48.403059+00:00"},{"alias_kind":"pith_short_8","alias_value":"K62GKY6K","created_at":"2026-07-05T07:48:48.403059+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.28405","citing_title":"Enhancing Layer Interaction Using Key-Correlated Layer Attention","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2312.06635","citing_title":"Gated Linear Attention Transformers with Hardware-Efficient Training","ref_index":49,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/K62GKY6K4IYKIFQNPBUZA3VLGZ","json":"https://pith.science/pith/K62GKY6K4IYKIFQNPBUZA3VLGZ.json","graph_json":"https://pith.science/api/pith-number/K62GKY6K4IYKIFQNPBUZA3VLGZ/graph.json","events_json":"https://pith.science/api/pith-number/K62GKY6K4IYKIFQNPBUZA3VLGZ/events.json","paper":"https://pith.science/paper/K62GKY6K"},"agent_actions":{"view_html":"https://pith.science/pith/K62GKY6K4IYKIFQNPBUZA3VLGZ","download_json":"https://pith.science/pith/K62GKY6K4IYKIFQNPBUZA3VLGZ.json","view_paper":"https://pith.science/paper/K62GKY6K","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2309.16354&json=true","fetch_graph":"https://pith.science/api/pith-number/K62GKY6K4IYKIFQNPBUZA3VLGZ/graph.json","fetch_events":"https://pith.science/api/pith-number/K62GKY6K4IYKIFQNPBUZA3VLGZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/K62GKY6K4IYKIFQNPBUZA3VLGZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/K62GKY6K4IYKIFQNPBUZA3VLGZ/action/storage_attestation","attest_author":"https://pith.science/pith/K62GKY6K4IYKIFQNPBUZA3VLGZ/action/author_attestation","sign_citation":"https://pith.science/pith/K62GKY6K4IYKIFQNPBUZA3VLGZ/action/citation_signature","submit_replication":"https://pith.science/pith/K62GKY6K4IYKIFQNPBUZA3VLGZ/action/replication_record"}},"created_at":"2026-07-05T07:48:48.403059+00:00","updated_at":"2026-07-05T07:48:48.403059+00:00"}