{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:42ZN5U4UFPKLFI25EIQ4NTILKU","short_pith_number":"pith:42ZN5U4U","schema_version":"1.0","canonical_sha256":"e6b2ded3942bd4b2a35d2221c6cd0b551cf6db73279f1e8ad8ac3c6543680807","source":{"kind":"arxiv","id":"2502.12216","version":1},"attestation_state":"computed","paper":{"title":"Tactic: Adaptive Sparse Attention with Clustering and Distribution Fitting for Long-Context LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Ang Li, Arvind Krishnamurthy, Baris Kasikci, Kan Zhu, Liangyu Zhao, Qinyu Xu, Rohan Kadekodi, Tian Tang, Yile Gu, Zhichen Zeng","submitted_at":"2025-02-17T08:39:43Z","abstract_excerpt":"Long-context models are essential for many applications but face inefficiencies in loading large KV caches during decoding. Prior methods enforce fixed token budgets for sparse attention, assuming a set number of tokens can approximate full attention. However, these methods overlook variations in the importance of attention across heads, layers, and contexts. To address these limitations, we propose Tactic, a sparsity-adaptive and calibration-free sparse attention mechanism that dynamically selects tokens based on their cumulative attention scores rather than a fixed token budget. By setting a"},"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":"2502.12216","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-17T08:39:43Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"28b0cbe1b4a7b539b46af21702dacd6ad4bb53429a0cadc1ed41c9eb4b2c8086","abstract_canon_sha256":"d62172907ec9f9ad7cb4b6d3bb06b2f026a17f8e6b0382a8e92f721806893061"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:15:52.132849Z","signature_b64":"2RZ5wN/i6O6L4weSnDY0Px4T8dxc+XxUPQkymj/XdFjuxnc6hRH6G8sxLBIH1J/4aaDPHaiT/izt+751aoiXDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e6b2ded3942bd4b2a35d2221c6cd0b551cf6db73279f1e8ad8ac3c6543680807","last_reissued_at":"2026-07-05T10:15:52.132241Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:15:52.132241Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Tactic: Adaptive Sparse Attention with Clustering and Distribution Fitting for Long-Context LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Ang Li, Arvind Krishnamurthy, Baris Kasikci, Kan Zhu, Liangyu Zhao, Qinyu Xu, Rohan Kadekodi, Tian Tang, Yile Gu, Zhichen Zeng","submitted_at":"2025-02-17T08:39:43Z","abstract_excerpt":"Long-context models are essential for many applications but face inefficiencies in loading large KV caches during decoding. Prior methods enforce fixed token budgets for sparse attention, assuming a set number of tokens can approximate full attention. However, these methods overlook variations in the importance of attention across heads, layers, and contexts. To address these limitations, we propose Tactic, a sparsity-adaptive and calibration-free sparse attention mechanism that dynamically selects tokens based on their cumulative attention scores rather than a fixed token budget. By setting a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.12216","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/2502.12216/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":"2502.12216","created_at":"2026-07-05T10:15:52.132398+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.12216v1","created_at":"2026-07-05T10:15:52.132398+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.12216","created_at":"2026-07-05T10:15:52.132398+00:00"},{"alias_kind":"pith_short_12","alias_value":"42ZN5U4UFPKL","created_at":"2026-07-05T10:15:52.132398+00:00"},{"alias_kind":"pith_short_16","alias_value":"42ZN5U4UFPKLFI25","created_at":"2026-07-05T10:15:52.132398+00:00"},{"alias_kind":"pith_short_8","alias_value":"42ZN5U4U","created_at":"2026-07-05T10:15:52.132398+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.01483","citing_title":"MURMUR: An Efficient Inference System for Long-Form ASR","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21649","citing_title":"EntmaxKV: Support-Aware Decoding for Entmax Attention","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2602.06283","citing_title":"SOCKET: SOft Collision Kernel EsTimator for Sparse Attention","ref_index":56,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06763","citing_title":"Sparse Attention as a Range Searching Problem: Towards an Inference-Efficient Index for KV Cache","ref_index":56,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/42ZN5U4UFPKLFI25EIQ4NTILKU","json":"https://pith.science/pith/42ZN5U4UFPKLFI25EIQ4NTILKU.json","graph_json":"https://pith.science/api/pith-number/42ZN5U4UFPKLFI25EIQ4NTILKU/graph.json","events_json":"https://pith.science/api/pith-number/42ZN5U4UFPKLFI25EIQ4NTILKU/events.json","paper":"https://pith.science/paper/42ZN5U4U"},"agent_actions":{"view_html":"https://pith.science/pith/42ZN5U4UFPKLFI25EIQ4NTILKU","download_json":"https://pith.science/pith/42ZN5U4UFPKLFI25EIQ4NTILKU.json","view_paper":"https://pith.science/paper/42ZN5U4U","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.12216&json=true","fetch_graph":"https://pith.science/api/pith-number/42ZN5U4UFPKLFI25EIQ4NTILKU/graph.json","fetch_events":"https://pith.science/api/pith-number/42ZN5U4UFPKLFI25EIQ4NTILKU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/42ZN5U4UFPKLFI25EIQ4NTILKU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/42ZN5U4UFPKLFI25EIQ4NTILKU/action/storage_attestation","attest_author":"https://pith.science/pith/42ZN5U4UFPKLFI25EIQ4NTILKU/action/author_attestation","sign_citation":"https://pith.science/pith/42ZN5U4UFPKLFI25EIQ4NTILKU/action/citation_signature","submit_replication":"https://pith.science/pith/42ZN5U4UFPKLFI25EIQ4NTILKU/action/replication_record"}},"created_at":"2026-07-05T10:15:52.132398+00:00","updated_at":"2026-07-05T10:15:52.132398+00:00"}