{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:AMSXYCY7PRK6X5ONIOI3GFXDO7","short_pith_number":"pith:AMSXYCY7","schema_version":"1.0","canonical_sha256":"03257c0b1f7c55ebf5cd4391b316e377e290e122578cd9fef3edd6cb7c7321f0","source":{"kind":"arxiv","id":"2602.08426","version":2},"attestation_state":"computed","paper":{"title":"Prism: Spectral-Aware Block-Sparse Attention","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.CL","authors_text":"Fangxu Liu, Jason Chu, Kai Song, Pengyu Wang, Xiaoran Liu, Xinghao Wang, Xipeng Qiu","submitted_at":"2026-02-09T09:31:06Z","abstract_excerpt":"Block-sparse attention is promising for accelerating long-context LLM pre-filling, yet identifying relevant blocks efficiently remains a bottleneck. Existing methods typically employ coarse-grained attention as a proxy for block importance estimation, but often resort to expensive token-level searching or scoring, resulting in significant selection overhead. In this work, we trace the inaccuracy of standard coarse-grained attention via mean pooling to a theoretical root cause: the interaction between mean pooling and Rotary Positional Embeddings (RoPE). We prove that mean pooling acts as a low"},"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":"2602.08426","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-02-09T09:31:06Z","cross_cats_sorted":["cs.AI","cs.CV"],"title_canon_sha256":"514896451a5188556104925ca8886786d84c16aa75f5d400272a06961e5520c6","abstract_canon_sha256":"eacacd6e4fb7f442f28c10e516f3a28e8a0e68fea1270c77b96ec665002ad255"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T02:04:05.725989Z","signature_b64":"runLJXFsM2pPruVtAO6sE+tqktXBoUV8t2deTkWSwnR8ss1ewhEb3dybTj6vNO0zAnjRb0A+pDNHBJd15VeMAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"03257c0b1f7c55ebf5cd4391b316e377e290e122578cd9fef3edd6cb7c7321f0","last_reissued_at":"2026-05-26T02:04:05.725070Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T02:04:05.725070Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Prism: Spectral-Aware Block-Sparse Attention","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.CL","authors_text":"Fangxu Liu, Jason Chu, Kai Song, Pengyu Wang, Xiaoran Liu, Xinghao Wang, Xipeng Qiu","submitted_at":"2026-02-09T09:31:06Z","abstract_excerpt":"Block-sparse attention is promising for accelerating long-context LLM pre-filling, yet identifying relevant blocks efficiently remains a bottleneck. Existing methods typically employ coarse-grained attention as a proxy for block importance estimation, but often resort to expensive token-level searching or scoring, resulting in significant selection overhead. In this work, we trace the inaccuracy of standard coarse-grained attention via mean pooling to a theoretical root cause: the interaction between mean pooling and Rotary Positional Embeddings (RoPE). We prove that mean pooling acts as a low"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.08426","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/2602.08426/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":"2602.08426","created_at":"2026-05-26T02:04:05.725191+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.08426v2","created_at":"2026-05-26T02:04:05.725191+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.08426","created_at":"2026-05-26T02:04:05.725191+00:00"},{"alias_kind":"pith_short_12","alias_value":"AMSXYCY7PRK6","created_at":"2026-05-26T02:04:05.725191+00:00"},{"alias_kind":"pith_short_16","alias_value":"AMSXYCY7PRK6X5ON","created_at":"2026-05-26T02:04:05.725191+00:00"},{"alias_kind":"pith_short_8","alias_value":"AMSXYCY7","created_at":"2026-05-26T02:04:05.725191+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.16928","citing_title":"Full Attention Strikes Back: Transferring Full Attention into Sparse within Hundred Training Steps","ref_index":24,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AMSXYCY7PRK6X5ONIOI3GFXDO7","json":"https://pith.science/pith/AMSXYCY7PRK6X5ONIOI3GFXDO7.json","graph_json":"https://pith.science/api/pith-number/AMSXYCY7PRK6X5ONIOI3GFXDO7/graph.json","events_json":"https://pith.science/api/pith-number/AMSXYCY7PRK6X5ONIOI3GFXDO7/events.json","paper":"https://pith.science/paper/AMSXYCY7"},"agent_actions":{"view_html":"https://pith.science/pith/AMSXYCY7PRK6X5ONIOI3GFXDO7","download_json":"https://pith.science/pith/AMSXYCY7PRK6X5ONIOI3GFXDO7.json","view_paper":"https://pith.science/paper/AMSXYCY7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.08426&json=true","fetch_graph":"https://pith.science/api/pith-number/AMSXYCY7PRK6X5ONIOI3GFXDO7/graph.json","fetch_events":"https://pith.science/api/pith-number/AMSXYCY7PRK6X5ONIOI3GFXDO7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AMSXYCY7PRK6X5ONIOI3GFXDO7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AMSXYCY7PRK6X5ONIOI3GFXDO7/action/storage_attestation","attest_author":"https://pith.science/pith/AMSXYCY7PRK6X5ONIOI3GFXDO7/action/author_attestation","sign_citation":"https://pith.science/pith/AMSXYCY7PRK6X5ONIOI3GFXDO7/action/citation_signature","submit_replication":"https://pith.science/pith/AMSXYCY7PRK6X5ONIOI3GFXDO7/action/replication_record"}},"created_at":"2026-05-26T02:04:05.725191+00:00","updated_at":"2026-05-26T02:04:05.725191+00:00"}