{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:K3FYOUUUCBIP6GD6EIOWV3V6V3","short_pith_number":"pith:K3FYOUUU","schema_version":"1.0","canonical_sha256":"56cb8752941050ff187e221d6aeebeaed6696c71037bf1dc18291ab88d99cc1b","source":{"kind":"arxiv","id":"2508.09001","version":2},"attestation_state":"computed","paper":{"title":"Retrospective Sparse Attention for Efficient Long-Context Generation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Beomseok Kang, Dongwon Jo, Jae-Joon Kim, Seonghwan Choi","submitted_at":"2025-08-12T15:11:47Z","abstract_excerpt":"Large Language Models (LLMs) are increasingly deployed in long-context tasks such as reasoning, code generation, and multi-turn dialogue. However, inference over extended contexts is bottlenecked by the Key-Value (KV) cache, whose memory footprint grows linearly with sequence length and dominates latency at each decoding step. While recent KV cache compression methods identify and load important few tokens, they focus predominantly on input contexts and fail to address the cumulative attention errors that arise during long decoding. In this paper, we introduce RetroAttention, a novel KV cache "},"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":"2508.09001","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-08-12T15:11:47Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"829adc956500616304910e12c84276edfa2329562cbe976c6f88220d9b74a6ee","abstract_canon_sha256":"c5b801de6a05ad40eb9db1ce54aa8852f14f1f719ba9592c0beda8e885352901"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-21T01:05:07.776941Z","signature_b64":"5OPay1f+boXLUhtF6y+4NiKM97+RuMpXiYV0EcLUEZ7utOzEjP6ZCS9tPC3SKS2eB4j0JjSJhd9y7WA8YYa1Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"56cb8752941050ff187e221d6aeebeaed6696c71037bf1dc18291ab88d99cc1b","last_reissued_at":"2026-05-21T01:05:07.775998Z","signature_status":"signed_v1","first_computed_at":"2026-05-21T01:05:07.775998Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Retrospective Sparse Attention for Efficient Long-Context Generation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Beomseok Kang, Dongwon Jo, Jae-Joon Kim, Seonghwan Choi","submitted_at":"2025-08-12T15:11:47Z","abstract_excerpt":"Large Language Models (LLMs) are increasingly deployed in long-context tasks such as reasoning, code generation, and multi-turn dialogue. However, inference over extended contexts is bottlenecked by the Key-Value (KV) cache, whose memory footprint grows linearly with sequence length and dominates latency at each decoding step. While recent KV cache compression methods identify and load important few tokens, they focus predominantly on input contexts and fail to address the cumulative attention errors that arise during long decoding. In this paper, we introduce RetroAttention, a novel KV cache "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.09001","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/2508.09001/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":"2508.09001","created_at":"2026-05-21T01:05:07.776143+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.09001v2","created_at":"2026-05-21T01:05:07.776143+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.09001","created_at":"2026-05-21T01:05:07.776143+00:00"},{"alias_kind":"pith_short_12","alias_value":"K3FYOUUUCBIP","created_at":"2026-05-21T01:05:07.776143+00:00"},{"alias_kind":"pith_short_16","alias_value":"K3FYOUUUCBIP6GD6","created_at":"2026-05-21T01:05:07.776143+00:00"},{"alias_kind":"pith_short_8","alias_value":"K3FYOUUU","created_at":"2026-05-21T01:05:07.776143+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/K3FYOUUUCBIP6GD6EIOWV3V6V3","json":"https://pith.science/pith/K3FYOUUUCBIP6GD6EIOWV3V6V3.json","graph_json":"https://pith.science/api/pith-number/K3FYOUUUCBIP6GD6EIOWV3V6V3/graph.json","events_json":"https://pith.science/api/pith-number/K3FYOUUUCBIP6GD6EIOWV3V6V3/events.json","paper":"https://pith.science/paper/K3FYOUUU"},"agent_actions":{"view_html":"https://pith.science/pith/K3FYOUUUCBIP6GD6EIOWV3V6V3","download_json":"https://pith.science/pith/K3FYOUUUCBIP6GD6EIOWV3V6V3.json","view_paper":"https://pith.science/paper/K3FYOUUU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.09001&json=true","fetch_graph":"https://pith.science/api/pith-number/K3FYOUUUCBIP6GD6EIOWV3V6V3/graph.json","fetch_events":"https://pith.science/api/pith-number/K3FYOUUUCBIP6GD6EIOWV3V6V3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/K3FYOUUUCBIP6GD6EIOWV3V6V3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/K3FYOUUUCBIP6GD6EIOWV3V6V3/action/storage_attestation","attest_author":"https://pith.science/pith/K3FYOUUUCBIP6GD6EIOWV3V6V3/action/author_attestation","sign_citation":"https://pith.science/pith/K3FYOUUUCBIP6GD6EIOWV3V6V3/action/citation_signature","submit_replication":"https://pith.science/pith/K3FYOUUUCBIP6GD6EIOWV3V6V3/action/replication_record"}},"created_at":"2026-05-21T01:05:07.776143+00:00","updated_at":"2026-05-21T01:05:07.776143+00:00"}