{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:4ILR6XJ5JD7PT3OAPQLTIXPJH5","short_pith_number":"pith:4ILR6XJ5","schema_version":"1.0","canonical_sha256":"e2171f5d3d48fef9edc07c17345de93f4987666cca1eedf674d54ddc64567474","source":{"kind":"arxiv","id":"2106.12566","version":2},"attestation_state":"computed","paper":{"title":"Stable, Fast and Accurate: Kernelized Attention with Relative Positional Encoding","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Di He, Dinglan Peng, Guolin Ke, Liwei Wang, Shanda Li, Shengjie Luo, Shuxin Zheng, Tianle Cai, Tie-Yan Liu","submitted_at":"2021-06-23T17:51:26Z","abstract_excerpt":"The attention module, which is a crucial component in Transformer, cannot scale efficiently to long sequences due to its quadratic complexity. Many works focus on approximating the dot-then-exponentiate softmax function in the original attention, leading to sub-quadratic or even linear-complexity Transformer architectures. However, we show that these methods cannot be applied to more powerful attention modules that go beyond the dot-then-exponentiate style, e.g., Transformers with relative positional encoding (RPE). Since in many state-of-the-art models, relative positional encoding is used as"},"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":"2106.12566","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-06-23T17:51:26Z","cross_cats_sorted":["cs.CL","stat.ML"],"title_canon_sha256":"969f564268769628f1606fb843f835af14abd76c17eea11413df504e69f9f860","abstract_canon_sha256":"a7bbf1c06b465427439d08c930f019cd00927518accdbf972f9f8d8754d584cd"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:28:38.241979Z","signature_b64":"tHOtSfQm/C0WVsXl9Jb80QWXiY0xrlZ3AMxx3E891X8ToNcLhP+x1Gh+jxM5FXZ/clVRYYLRRmQ/dJfx7rgwBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e2171f5d3d48fef9edc07c17345de93f4987666cca1eedf674d54ddc64567474","last_reissued_at":"2026-07-05T03:28:38.241569Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:28:38.241569Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Stable, Fast and Accurate: Kernelized Attention with Relative Positional Encoding","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Di He, Dinglan Peng, Guolin Ke, Liwei Wang, Shanda Li, Shengjie Luo, Shuxin Zheng, Tianle Cai, Tie-Yan Liu","submitted_at":"2021-06-23T17:51:26Z","abstract_excerpt":"The attention module, which is a crucial component in Transformer, cannot scale efficiently to long sequences due to its quadratic complexity. Many works focus on approximating the dot-then-exponentiate softmax function in the original attention, leading to sub-quadratic or even linear-complexity Transformer architectures. However, we show that these methods cannot be applied to more powerful attention modules that go beyond the dot-then-exponentiate style, e.g., Transformers with relative positional encoding (RPE). Since in many state-of-the-art models, relative positional encoding is used as"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2106.12566","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/2106.12566/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":"2106.12566","created_at":"2026-07-05T03:28:38.241623+00:00"},{"alias_kind":"arxiv_version","alias_value":"2106.12566v2","created_at":"2026-07-05T03:28:38.241623+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2106.12566","created_at":"2026-07-05T03:28:38.241623+00:00"},{"alias_kind":"pith_short_12","alias_value":"4ILR6XJ5JD7P","created_at":"2026-07-05T03:28:38.241623+00:00"},{"alias_kind":"pith_short_16","alias_value":"4ILR6XJ5JD7PT3OA","created_at":"2026-07-05T03:28:38.241623+00:00"},{"alias_kind":"pith_short_8","alias_value":"4ILR6XJ5","created_at":"2026-07-05T03:28:38.241623+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.02144","citing_title":"Projection-Free Transformers via Gaussian Kernel Attention","ref_index":21,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4ILR6XJ5JD7PT3OAPQLTIXPJH5","json":"https://pith.science/pith/4ILR6XJ5JD7PT3OAPQLTIXPJH5.json","graph_json":"https://pith.science/api/pith-number/4ILR6XJ5JD7PT3OAPQLTIXPJH5/graph.json","events_json":"https://pith.science/api/pith-number/4ILR6XJ5JD7PT3OAPQLTIXPJH5/events.json","paper":"https://pith.science/paper/4ILR6XJ5"},"agent_actions":{"view_html":"https://pith.science/pith/4ILR6XJ5JD7PT3OAPQLTIXPJH5","download_json":"https://pith.science/pith/4ILR6XJ5JD7PT3OAPQLTIXPJH5.json","view_paper":"https://pith.science/paper/4ILR6XJ5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2106.12566&json=true","fetch_graph":"https://pith.science/api/pith-number/4ILR6XJ5JD7PT3OAPQLTIXPJH5/graph.json","fetch_events":"https://pith.science/api/pith-number/4ILR6XJ5JD7PT3OAPQLTIXPJH5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4ILR6XJ5JD7PT3OAPQLTIXPJH5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4ILR6XJ5JD7PT3OAPQLTIXPJH5/action/storage_attestation","attest_author":"https://pith.science/pith/4ILR6XJ5JD7PT3OAPQLTIXPJH5/action/author_attestation","sign_citation":"https://pith.science/pith/4ILR6XJ5JD7PT3OAPQLTIXPJH5/action/citation_signature","submit_replication":"https://pith.science/pith/4ILR6XJ5JD7PT3OAPQLTIXPJH5/action/replication_record"}},"created_at":"2026-07-05T03:28:38.241623+00:00","updated_at":"2026-07-05T03:28:38.241623+00:00"}