{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:5FN2QWTUKBAPKOUJA3HK56MQVM","short_pith_number":"pith:5FN2QWTU","schema_version":"1.0","canonical_sha256":"e95ba85a745040f53a8906ceaef990ab23906f85bfde58209122e52d0a30865d","source":{"kind":"arxiv","id":"2408.07092","version":2},"attestation_state":"computed","paper":{"title":"Post-Training Sparse Attention with Double Sparsity","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Ion Stoica, Joseph E. Gonzalez, Lianmin Zheng, Shuo Yang, Ying Sheng","submitted_at":"2024-08-11T18:40:36Z","abstract_excerpt":"The inference process for large language models is slow and memory-intensive, with one of the most critical bottlenecks being excessive Key-Value (KV) cache accesses. This paper introduces \"Double Sparsity,\" a novel post-training sparse attention technique designed to alleviate this bottleneck by reducing KV cache access. Double Sparsity combines token sparsity, which focuses on utilizing only the important tokens for computing self-attention, with channel sparsity, an approach that uses important feature channels for identifying important tokens. Our key insight is that the pattern of channel"},"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":"2408.07092","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-08-11T18:40:36Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"1561c73eac3fb54b238b4c0ac5f7f6a38da1f64757242d2b0bba596991bf3e3f","abstract_canon_sha256":"61aa8748e2e77dac0ae940d127dd5866bc4ae0c110188003e9bd4d7d620902b6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:56:24.280069Z","signature_b64":"JYg08SJqh5nnqaHqpGOsA9/U35awsLXwNJCI3waPmsluB7ePrNHqF6/MrXn0rranWi/MajpP+bLAl+nv5FKKCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e95ba85a745040f53a8906ceaef990ab23906f85bfde58209122e52d0a30865d","last_reissued_at":"2026-07-05T08:56:24.279680Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:56:24.279680Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Post-Training Sparse Attention with Double Sparsity","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Ion Stoica, Joseph E. Gonzalez, Lianmin Zheng, Shuo Yang, Ying Sheng","submitted_at":"2024-08-11T18:40:36Z","abstract_excerpt":"The inference process for large language models is slow and memory-intensive, with one of the most critical bottlenecks being excessive Key-Value (KV) cache accesses. This paper introduces \"Double Sparsity,\" a novel post-training sparse attention technique designed to alleviate this bottleneck by reducing KV cache access. Double Sparsity combines token sparsity, which focuses on utilizing only the important tokens for computing self-attention, with channel sparsity, an approach that uses important feature channels for identifying important tokens. Our key insight is that the pattern of channel"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.07092","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/2408.07092/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":"2408.07092","created_at":"2026-07-05T08:56:24.279736+00:00"},{"alias_kind":"arxiv_version","alias_value":"2408.07092v2","created_at":"2026-07-05T08:56:24.279736+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.07092","created_at":"2026-07-05T08:56:24.279736+00:00"},{"alias_kind":"pith_short_12","alias_value":"5FN2QWTUKBAP","created_at":"2026-07-05T08:56:24.279736+00:00"},{"alias_kind":"pith_short_16","alias_value":"5FN2QWTUKBAPKOUJ","created_at":"2026-07-05T08:56:24.279736+00:00"},{"alias_kind":"pith_short_8","alias_value":"5FN2QWTU","created_at":"2026-07-05T08:56:24.279736+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.31519","citing_title":"RaBitQCache: Rotated Binary Quantization for KVCache in Long Context LLM Inference","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24168","citing_title":"Inference Time Context Sparsity: Illusion or Opportunity?","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19218","citing_title":"Rotation-Aligned Key Channel Pruning for Efficient Vision-Language Model Inference","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2602.06283","citing_title":"SOCKET: SOft Collision Kernel EsTimator for Sparse Attention","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24820","citing_title":"Salca: A Sparsity-Aware Hardware Accelerator for Efficient Long-Context Attention Decoding","ref_index":63,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18348","citing_title":"AdaCluster: Adaptive Query-Key Clustering for Sparse Attention in Video Generation","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16864","citing_title":"HieraSparse: Hierarchical Semi-Structured Sparse KV Attention","ref_index":33,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5FN2QWTUKBAPKOUJA3HK56MQVM","json":"https://pith.science/pith/5FN2QWTUKBAPKOUJA3HK56MQVM.json","graph_json":"https://pith.science/api/pith-number/5FN2QWTUKBAPKOUJA3HK56MQVM/graph.json","events_json":"https://pith.science/api/pith-number/5FN2QWTUKBAPKOUJA3HK56MQVM/events.json","paper":"https://pith.science/paper/5FN2QWTU"},"agent_actions":{"view_html":"https://pith.science/pith/5FN2QWTUKBAPKOUJA3HK56MQVM","download_json":"https://pith.science/pith/5FN2QWTUKBAPKOUJA3HK56MQVM.json","view_paper":"https://pith.science/paper/5FN2QWTU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2408.07092&json=true","fetch_graph":"https://pith.science/api/pith-number/5FN2QWTUKBAPKOUJA3HK56MQVM/graph.json","fetch_events":"https://pith.science/api/pith-number/5FN2QWTUKBAPKOUJA3HK56MQVM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5FN2QWTUKBAPKOUJA3HK56MQVM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5FN2QWTUKBAPKOUJA3HK56MQVM/action/storage_attestation","attest_author":"https://pith.science/pith/5FN2QWTUKBAPKOUJA3HK56MQVM/action/author_attestation","sign_citation":"https://pith.science/pith/5FN2QWTUKBAPKOUJA3HK56MQVM/action/citation_signature","submit_replication":"https://pith.science/pith/5FN2QWTUKBAPKOUJA3HK56MQVM/action/replication_record"}},"created_at":"2026-07-05T08:56:24.279736+00:00","updated_at":"2026-07-05T08:56:24.279736+00:00"}