{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:X6HMPEMEOTE3UMO2NXL6C4O3CE","short_pith_number":"pith:X6HMPEME","schema_version":"1.0","canonical_sha256":"bf8ec7918474c9ba31da6dd7e171db11293a565173e644dfdc083da612bed735","source":{"kind":"arxiv","id":"2605.15508","version":2},"attestation_state":"computed","paper":{"title":"STS: Efficient Sparse Attention with Speculative Token Sparsity","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Ceyu Xu, Jiangnan Yu, Yongji Wu, Yuan Xie","submitted_at":"2026-05-15T01:05:37Z","abstract_excerpt":"The quadratic complexity of attention imposes severe memory and computational bottlenecks on Large Language Model (LLM) inference. This challenge is particularly acute for emerging agentic applications that require processing multi-million token sequences. We propose STS, a sparse attention mechanism that requires no model retraining. STS leverages the key insight that tokens identified as important by a smaller draft model are highly predictive of important tokens for a larger target model. By integrating into speculative decoding frameworks, STS repurposes the draft model's attention scores "},"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":"2605.15508","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-15T01:05:37Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"1c937bae268b4f53f87d115e6b44e37f994bc44cca024c6d5635921b467f0bfd","abstract_canon_sha256":"2d1583818a0a2ffb0d6222f7e4cf73459ef087721a097e5ed6d5fb4a310a06a1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:04:37.965736Z","signature_b64":"sclABgM1Gd04uwss4TR7sNkhjuX16I0BW+xgvcARV4f65D74YaPh4jGXWVzuLs70m4vozlNpJ2eJwK5dWZn3DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bf8ec7918474c9ba31da6dd7e171db11293a565173e644dfdc083da612bed735","last_reissued_at":"2026-05-20T00:04:37.964682Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:04:37.964682Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"STS: Efficient Sparse Attention with Speculative Token Sparsity","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Ceyu Xu, Jiangnan Yu, Yongji Wu, Yuan Xie","submitted_at":"2026-05-15T01:05:37Z","abstract_excerpt":"The quadratic complexity of attention imposes severe memory and computational bottlenecks on Large Language Model (LLM) inference. This challenge is particularly acute for emerging agentic applications that require processing multi-million token sequences. We propose STS, a sparse attention mechanism that requires no model retraining. STS leverages the key insight that tokens identified as important by a smaller draft model are highly predictive of important tokens for a larger target model. By integrating into speculative decoding frameworks, STS repurposes the draft model's attention scores "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.15508","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/2605.15508/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":"2605.15508","created_at":"2026-05-20T00:04:37.964841+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.15508v2","created_at":"2026-05-20T00:04:37.964841+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.15508","created_at":"2026-05-20T00:04:37.964841+00:00"},{"alias_kind":"pith_short_12","alias_value":"X6HMPEMEOTE3","created_at":"2026-05-20T00:04:37.964841+00:00"},{"alias_kind":"pith_short_16","alias_value":"X6HMPEMEOTE3UMO2","created_at":"2026-05-20T00:04:37.964841+00:00"},{"alias_kind":"pith_short_8","alias_value":"X6HMPEME","created_at":"2026-05-20T00:04:37.964841+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/X6HMPEMEOTE3UMO2NXL6C4O3CE","json":"https://pith.science/pith/X6HMPEMEOTE3UMO2NXL6C4O3CE.json","graph_json":"https://pith.science/api/pith-number/X6HMPEMEOTE3UMO2NXL6C4O3CE/graph.json","events_json":"https://pith.science/api/pith-number/X6HMPEMEOTE3UMO2NXL6C4O3CE/events.json","paper":"https://pith.science/paper/X6HMPEME"},"agent_actions":{"view_html":"https://pith.science/pith/X6HMPEMEOTE3UMO2NXL6C4O3CE","download_json":"https://pith.science/pith/X6HMPEMEOTE3UMO2NXL6C4O3CE.json","view_paper":"https://pith.science/paper/X6HMPEME","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.15508&json=true","fetch_graph":"https://pith.science/api/pith-number/X6HMPEMEOTE3UMO2NXL6C4O3CE/graph.json","fetch_events":"https://pith.science/api/pith-number/X6HMPEMEOTE3UMO2NXL6C4O3CE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/X6HMPEMEOTE3UMO2NXL6C4O3CE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/X6HMPEMEOTE3UMO2NXL6C4O3CE/action/storage_attestation","attest_author":"https://pith.science/pith/X6HMPEMEOTE3UMO2NXL6C4O3CE/action/author_attestation","sign_citation":"https://pith.science/pith/X6HMPEMEOTE3UMO2NXL6C4O3CE/action/citation_signature","submit_replication":"https://pith.science/pith/X6HMPEMEOTE3UMO2NXL6C4O3CE/action/replication_record"}},"created_at":"2026-05-20T00:04:37.964841+00:00","updated_at":"2026-05-20T00:04:37.964841+00:00"}