{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:K6DUBFKM6S6P5PCZUXBTIUAPNM","short_pith_number":"pith:K6DUBFKM","schema_version":"1.0","canonical_sha256":"578740954cf4bcfebc59a5c334500f6b0bafcbc584da6ff133831f43918f4b9f","source":{"kind":"arxiv","id":"2409.15355","version":5},"attestation_state":"computed","paper":{"title":"Block-Attention for Efficient Prefilling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Dongyang Ma, Lan Tian, Yan Wang","submitted_at":"2024-09-14T02:34:26Z","abstract_excerpt":"We introduce Block-attention, an attention mechanism designed to address the increased inference latency and cost in Retrieval-Augmented Generation (RAG) scenarios. Traditional approaches often encode the entire context in an auto-regressive manner. Instead, Block-attention divides retrieved documents into discrete blocks, with each block independently calculating key-value (KV) states except for the final block. In RAG scenarios, by defining each passage as a block, Block-attention enables us to reuse the KV states of passages that have been seen before, thereby significantly reducing the lat"},"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":"2409.15355","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-14T02:34:26Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"8000e8438ee7ede48c56579ba85f563b38c4fa0b63d8895241d6941eae15fd30","abstract_canon_sha256":"e40946a0ef90c1e656706791f2b18855a50ab02321528deec52b85467f1aa2c1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:48:19.530832Z","signature_b64":"OFhZc1mBRpejJ2d9TD7jJXAFhgPoTj0bDdk9M84AvKW3Dg/jA5sjDw9L0dGiGDVnW1gSiF+ROVy03Y9MsGMMBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"578740954cf4bcfebc59a5c334500f6b0bafcbc584da6ff133831f43918f4b9f","last_reissued_at":"2026-07-05T10:48:19.530365Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:48:19.530365Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Block-Attention for Efficient Prefilling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Dongyang Ma, Lan Tian, Yan Wang","submitted_at":"2024-09-14T02:34:26Z","abstract_excerpt":"We introduce Block-attention, an attention mechanism designed to address the increased inference latency and cost in Retrieval-Augmented Generation (RAG) scenarios. Traditional approaches often encode the entire context in an auto-regressive manner. Instead, Block-attention divides retrieved documents into discrete blocks, with each block independently calculating key-value (KV) states except for the final block. In RAG scenarios, by defining each passage as a block, Block-attention enables us to reuse the KV states of passages that have been seen before, thereby significantly reducing the lat"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.15355","kind":"arxiv","version":5},"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/2409.15355/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":"2409.15355","created_at":"2026-07-05T10:48:19.530419+00:00"},{"alias_kind":"arxiv_version","alias_value":"2409.15355v5","created_at":"2026-07-05T10:48:19.530419+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.15355","created_at":"2026-07-05T10:48:19.530419+00:00"},{"alias_kind":"pith_short_12","alias_value":"K6DUBFKM6S6P","created_at":"2026-07-05T10:48:19.530419+00:00"},{"alias_kind":"pith_short_16","alias_value":"K6DUBFKM6S6P5PCZ","created_at":"2026-07-05T10:48:19.530419+00:00"},{"alias_kind":"pith_short_8","alias_value":"K6DUBFKM","created_at":"2026-07-05T10:48:19.530419+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.01538","citing_title":"Can Language Models Actually Retrieve In-Context? Drowning in Documents at Million Token Scale","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2606.13126","citing_title":"MiniPIC: Flexible Position-Independent Caching in <100LOC","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2512.13368","citing_title":"BlossomRec: Block-level Fused Sparse Attention Mechanism for Sequential Recommendations","ref_index":56,"is_internal_anchor":false},{"citing_arxiv_id":"2510.10129","citing_title":"CacheClip: Accelerating RAG with Effective KV Cache Reuse","ref_index":27,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/K6DUBFKM6S6P5PCZUXBTIUAPNM","json":"https://pith.science/pith/K6DUBFKM6S6P5PCZUXBTIUAPNM.json","graph_json":"https://pith.science/api/pith-number/K6DUBFKM6S6P5PCZUXBTIUAPNM/graph.json","events_json":"https://pith.science/api/pith-number/K6DUBFKM6S6P5PCZUXBTIUAPNM/events.json","paper":"https://pith.science/paper/K6DUBFKM"},"agent_actions":{"view_html":"https://pith.science/pith/K6DUBFKM6S6P5PCZUXBTIUAPNM","download_json":"https://pith.science/pith/K6DUBFKM6S6P5PCZUXBTIUAPNM.json","view_paper":"https://pith.science/paper/K6DUBFKM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2409.15355&json=true","fetch_graph":"https://pith.science/api/pith-number/K6DUBFKM6S6P5PCZUXBTIUAPNM/graph.json","fetch_events":"https://pith.science/api/pith-number/K6DUBFKM6S6P5PCZUXBTIUAPNM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/K6DUBFKM6S6P5PCZUXBTIUAPNM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/K6DUBFKM6S6P5PCZUXBTIUAPNM/action/storage_attestation","attest_author":"https://pith.science/pith/K6DUBFKM6S6P5PCZUXBTIUAPNM/action/author_attestation","sign_citation":"https://pith.science/pith/K6DUBFKM6S6P5PCZUXBTIUAPNM/action/citation_signature","submit_replication":"https://pith.science/pith/K6DUBFKM6S6P5PCZUXBTIUAPNM/action/replication_record"}},"created_at":"2026-07-05T10:48:19.530419+00:00","updated_at":"2026-07-05T10:48:19.530419+00:00"}