{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:DNAA2BXDVTKB5HONSZN4DZOEAS","short_pith_number":"pith:DNAA2BXD","schema_version":"1.0","canonical_sha256":"1b400d06e3acd41e9dcd965bc1e5c4048e20b9bbe16e76913282b2b82b0aca0e","source":{"kind":"arxiv","id":"2504.08719","version":1},"attestation_state":"computed","paper":{"title":"SWAN-GPT: An Efficient and Scalable Approach for Long-Context Language Modeling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Boris Ginsburg, Cheng-Ping Hsieh, Dima Rekesh, Faisal Ladhak, Fei Jia, Krishna C. Puvvada, Samuel Kriman, Santiago Akle Serrano, Shantanu Acharya, Simeng Sun, Somshubra Majumdar","submitted_at":"2025-04-11T17:33:32Z","abstract_excerpt":"We present a decoder-only Transformer architecture that robustly generalizes to sequence lengths substantially longer than those seen during training. Our model, SWAN-GPT, interleaves layers without positional encodings (NoPE) and sliding-window attention layers equipped with rotary positional encodings (SWA-RoPE). Experiments demonstrate strong performance on sequence lengths significantly longer than the training length without the need for additional long-context training. This robust length extrapolation is achieved through our novel architecture, enhanced by a straightforward dynamic scal"},"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":"2504.08719","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-04-11T17:33:32Z","cross_cats_sorted":[],"title_canon_sha256":"d105e65feb5a8171b4a130889b92e385fb3dd9a2be96dda174abf89a5fbcd87c","abstract_canon_sha256":"7662ca889bf9a2ab241a5e9716cdc78608063606b6b67d9fb42d2b46006824c3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:47:53.111369Z","signature_b64":"iLT6C7XHaflel53hl2SNsNjcHEmwgRBGrqn8wMEZTxNL2WLeaujSd6JNY4EvI/Ck7nK8nxlqEtrv9IEc6cBlBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1b400d06e3acd41e9dcd965bc1e5c4048e20b9bbe16e76913282b2b82b0aca0e","last_reissued_at":"2026-07-05T10:47:53.110888Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:47:53.110888Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SWAN-GPT: An Efficient and Scalable Approach for Long-Context Language Modeling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Boris Ginsburg, Cheng-Ping Hsieh, Dima Rekesh, Faisal Ladhak, Fei Jia, Krishna C. Puvvada, Samuel Kriman, Santiago Akle Serrano, Shantanu Acharya, Simeng Sun, Somshubra Majumdar","submitted_at":"2025-04-11T17:33:32Z","abstract_excerpt":"We present a decoder-only Transformer architecture that robustly generalizes to sequence lengths substantially longer than those seen during training. Our model, SWAN-GPT, interleaves layers without positional encodings (NoPE) and sliding-window attention layers equipped with rotary positional encodings (SWA-RoPE). Experiments demonstrate strong performance on sequence lengths significantly longer than the training length without the need for additional long-context training. This robust length extrapolation is achieved through our novel architecture, enhanced by a straightforward dynamic scal"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.08719","kind":"arxiv","version":1},"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/2504.08719/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":"2504.08719","created_at":"2026-07-05T10:47:53.110943+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.08719v1","created_at":"2026-07-05T10:47:53.110943+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.08719","created_at":"2026-07-05T10:47:53.110943+00:00"},{"alias_kind":"pith_short_12","alias_value":"DNAA2BXDVTKB","created_at":"2026-07-05T10:47:53.110943+00:00"},{"alias_kind":"pith_short_16","alias_value":"DNAA2BXDVTKB5HON","created_at":"2026-07-05T10:47:53.110943+00:00"},{"alias_kind":"pith_short_8","alias_value":"DNAA2BXD","created_at":"2026-07-05T10:47:53.110943+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2510.26692","citing_title":"Kimi Linear: An Expressive, Efficient Attention Architecture","ref_index":77,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DNAA2BXDVTKB5HONSZN4DZOEAS","json":"https://pith.science/pith/DNAA2BXDVTKB5HONSZN4DZOEAS.json","graph_json":"https://pith.science/api/pith-number/DNAA2BXDVTKB5HONSZN4DZOEAS/graph.json","events_json":"https://pith.science/api/pith-number/DNAA2BXDVTKB5HONSZN4DZOEAS/events.json","paper":"https://pith.science/paper/DNAA2BXD"},"agent_actions":{"view_html":"https://pith.science/pith/DNAA2BXDVTKB5HONSZN4DZOEAS","download_json":"https://pith.science/pith/DNAA2BXDVTKB5HONSZN4DZOEAS.json","view_paper":"https://pith.science/paper/DNAA2BXD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.08719&json=true","fetch_graph":"https://pith.science/api/pith-number/DNAA2BXDVTKB5HONSZN4DZOEAS/graph.json","fetch_events":"https://pith.science/api/pith-number/DNAA2BXDVTKB5HONSZN4DZOEAS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DNAA2BXDVTKB5HONSZN4DZOEAS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DNAA2BXDVTKB5HONSZN4DZOEAS/action/storage_attestation","attest_author":"https://pith.science/pith/DNAA2BXDVTKB5HONSZN4DZOEAS/action/author_attestation","sign_citation":"https://pith.science/pith/DNAA2BXDVTKB5HONSZN4DZOEAS/action/citation_signature","submit_replication":"https://pith.science/pith/DNAA2BXDVTKB5HONSZN4DZOEAS/action/replication_record"}},"created_at":"2026-07-05T10:47:53.110943+00:00","updated_at":"2026-07-05T10:47:53.110943+00:00"}