{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:SO62FFJJO5VVK446SZKRSXXI27","short_pith_number":"pith:SO62FFJJ","schema_version":"1.0","canonical_sha256":"93bda29529776b55739e9655195ee8d7fa3eb546dbade718e2ac7aa94619bfce","source":{"kind":"arxiv","id":"2508.18224","version":3},"attestation_state":"computed","paper":{"title":"FSA: An Alternative Efficient Implementation of Native Sparse Attention Kernel","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.DC","authors_text":"Beidi Chen, Binhang Yuan, Haohui Mai, Ran Yan, Youhe Jiang, Zhuoming Chen","submitted_at":"2025-08-25T17:22:15Z","abstract_excerpt":"Recent advances in sparse attention mechanisms have demonstrated strong potential for reducing the computational cost of long-context training and inference in large language models (LLMs). Native Sparse Attention (NSA), one state-of-the-art approach, introduces natively trainable, hardware-aligned sparse attention that delivers substantial system-level performance boosts while maintaining accuracy comparable to full attention. However, the kernel implementation of NSA forces a loop order that is only efficient with a relatively large number of query heads in each Grouped Query Attention (GQA)"},"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":"2508.18224","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2025-08-25T17:22:15Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"7dfe7f24268b0033b10d16c97775e77e133ef0260a934af9cf39f2f191c6b874","abstract_canon_sha256":"d33e866ebfb89fd0fb536b577466e3f5f066201635bba385d27395b7ea4b626b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-02T01:18:04.333880Z","signature_b64":"k5me0LSy2Uo35oSMS8s14qj5gjb5nmhp1XSnVV2DoCgMISBX2Vkc0ENwpHhJgNsh9hckPDta88/FhbBZmSrOBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"93bda29529776b55739e9655195ee8d7fa3eb546dbade718e2ac7aa94619bfce","last_reissued_at":"2026-07-02T01:18:04.333366Z","signature_status":"signed_v1","first_computed_at":"2026-07-02T01:18:04.333366Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"FSA: An Alternative Efficient Implementation of Native Sparse Attention Kernel","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.DC","authors_text":"Beidi Chen, Binhang Yuan, Haohui Mai, Ran Yan, Youhe Jiang, Zhuoming Chen","submitted_at":"2025-08-25T17:22:15Z","abstract_excerpt":"Recent advances in sparse attention mechanisms have demonstrated strong potential for reducing the computational cost of long-context training and inference in large language models (LLMs). Native Sparse Attention (NSA), one state-of-the-art approach, introduces natively trainable, hardware-aligned sparse attention that delivers substantial system-level performance boosts while maintaining accuracy comparable to full attention. However, the kernel implementation of NSA forces a loop order that is only efficient with a relatively large number of query heads in each Grouped Query Attention (GQA)"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.18224","kind":"arxiv","version":3},"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/2508.18224/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":"2508.18224","created_at":"2026-07-02T01:18:04.333430+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.18224v3","created_at":"2026-07-02T01:18:04.333430+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.18224","created_at":"2026-07-02T01:18:04.333430+00:00"},{"alias_kind":"pith_short_12","alias_value":"SO62FFJJO5VV","created_at":"2026-07-02T01:18:04.333430+00:00"},{"alias_kind":"pith_short_16","alias_value":"SO62FFJJO5VVK446","created_at":"2026-07-02T01:18:04.333430+00:00"},{"alias_kind":"pith_short_8","alias_value":"SO62FFJJ","created_at":"2026-07-02T01:18:04.333430+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":4,"sample":[{"citing_arxiv_id":"2606.08446","citing_title":"Sparrow: Sparse Rollout for Stable and Efficient Long-context RL of Large Language Models","ref_index":18,"is_internal_anchor":true},{"citing_arxiv_id":"2605.16637","citing_title":"HexAGenT: Efficient Agentic LLM Serving via Workflow- and Heterogeneity-Aware Scheduling","ref_index":46,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07569","citing_title":"HexiSeq: Accommodating Long Context Training of LLMs over Heterogeneous Hardware","ref_index":57,"is_internal_anchor":true},{"citing_arxiv_id":"2605.05227","citing_title":"Rethinking Data Curation in LLM Training: Online Reweighting Offers Better Generalization than Offline Methods","ref_index":50,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SO62FFJJO5VVK446SZKRSXXI27","json":"https://pith.science/pith/SO62FFJJO5VVK446SZKRSXXI27.json","graph_json":"https://pith.science/api/pith-number/SO62FFJJO5VVK446SZKRSXXI27/graph.json","events_json":"https://pith.science/api/pith-number/SO62FFJJO5VVK446SZKRSXXI27/events.json","paper":"https://pith.science/paper/SO62FFJJ"},"agent_actions":{"view_html":"https://pith.science/pith/SO62FFJJO5VVK446SZKRSXXI27","download_json":"https://pith.science/pith/SO62FFJJO5VVK446SZKRSXXI27.json","view_paper":"https://pith.science/paper/SO62FFJJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.18224&json=true","fetch_graph":"https://pith.science/api/pith-number/SO62FFJJO5VVK446SZKRSXXI27/graph.json","fetch_events":"https://pith.science/api/pith-number/SO62FFJJO5VVK446SZKRSXXI27/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SO62FFJJO5VVK446SZKRSXXI27/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SO62FFJJO5VVK446SZKRSXXI27/action/storage_attestation","attest_author":"https://pith.science/pith/SO62FFJJO5VVK446SZKRSXXI27/action/author_attestation","sign_citation":"https://pith.science/pith/SO62FFJJO5VVK446SZKRSXXI27/action/citation_signature","submit_replication":"https://pith.science/pith/SO62FFJJO5VVK446SZKRSXXI27/action/replication_record"}},"created_at":"2026-07-02T01:18:04.333430+00:00","updated_at":"2026-07-02T01:18:04.333430+00:00"}