{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:LLIL342SEOICJPBN4KYR2L5CB5","short_pith_number":"pith:LLIL342S","schema_version":"1.0","canonical_sha256":"5ad0bdf352239024bc2de2b11d2fa20f642509a19d5023b8351935cd96470b84","source":{"kind":"arxiv","id":"2406.02542","version":2},"attestation_state":"computed","paper":{"title":"Loki: Low-rank Keys for Efficient Sparse Attention","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Abhinav Bhatele, Prajwal Singhania, Shwai He, Siddharth Singh, Soheil Feizi","submitted_at":"2024-06-04T17:58:03Z","abstract_excerpt":"Inference on large language models (LLMs) can be expensive in terms of the compute and memory costs involved, especially when long sequence lengths are used. In particular, the self-attention mechanism used in LLM inference contributes significantly to these costs, which has sparked an interest in approximating the self-attention computation to reduce such costs. In this work, we propose to approximate self-attention by focusing on the dimensionality of key vectors computed in the attention block. Our analysis reveals that key vectors lie in a significantly lower-dimensional space, consistentl"},"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":"2406.02542","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-06-04T17:58:03Z","cross_cats_sorted":[],"title_canon_sha256":"27741bd88aacec9c1367ace099ee6f04cff8eb8dad68423a870703d07e178139","abstract_canon_sha256":"5a64d36478056b129832501fbd5b730fef894e1d9241e7efe25504bf9454da48"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:32:44.704526Z","signature_b64":"8KYiVlyd/SPYH5EzWN6aJmFTeEECN2I2c0+A2k1Z6xenJZJEifxFIJjtD45+f3db5Ai7jYY/gOm27y1H3Ab1DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5ad0bdf352239024bc2de2b11d2fa20f642509a19d5023b8351935cd96470b84","last_reissued_at":"2026-07-05T09:32:44.703966Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:32:44.703966Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Loki: Low-rank Keys for Efficient Sparse Attention","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Abhinav Bhatele, Prajwal Singhania, Shwai He, Siddharth Singh, Soheil Feizi","submitted_at":"2024-06-04T17:58:03Z","abstract_excerpt":"Inference on large language models (LLMs) can be expensive in terms of the compute and memory costs involved, especially when long sequence lengths are used. In particular, the self-attention mechanism used in LLM inference contributes significantly to these costs, which has sparked an interest in approximating the self-attention computation to reduce such costs. In this work, we propose to approximate self-attention by focusing on the dimensionality of key vectors computed in the attention block. Our analysis reveals that key vectors lie in a significantly lower-dimensional space, consistentl"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.02542","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/2406.02542/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":"2406.02542","created_at":"2026-07-05T09:32:44.704036+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.02542v2","created_at":"2026-07-05T09:32:44.704036+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.02542","created_at":"2026-07-05T09:32:44.704036+00:00"},{"alias_kind":"pith_short_12","alias_value":"LLIL342SEOIC","created_at":"2026-07-05T09:32:44.704036+00:00"},{"alias_kind":"pith_short_16","alias_value":"LLIL342SEOICJPBN","created_at":"2026-07-05T09:32:44.704036+00:00"},{"alias_kind":"pith_short_8","alias_value":"LLIL342S","created_at":"2026-07-05T09:32:44.704036+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2409.10516","citing_title":"RetrievalAttention: Accelerating Long-Context LLM Inference via Vector Retrieval","ref_index":108,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03260","citing_title":"Why Attend to Everything? Focus is the Key","ref_index":37,"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":40,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16864","citing_title":"HieraSparse: Hierarchical Semi-Structured Sparse KV Attention","ref_index":32,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LLIL342SEOICJPBN4KYR2L5CB5","json":"https://pith.science/pith/LLIL342SEOICJPBN4KYR2L5CB5.json","graph_json":"https://pith.science/api/pith-number/LLIL342SEOICJPBN4KYR2L5CB5/graph.json","events_json":"https://pith.science/api/pith-number/LLIL342SEOICJPBN4KYR2L5CB5/events.json","paper":"https://pith.science/paper/LLIL342S"},"agent_actions":{"view_html":"https://pith.science/pith/LLIL342SEOICJPBN4KYR2L5CB5","download_json":"https://pith.science/pith/LLIL342SEOICJPBN4KYR2L5CB5.json","view_paper":"https://pith.science/paper/LLIL342S","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.02542&json=true","fetch_graph":"https://pith.science/api/pith-number/LLIL342SEOICJPBN4KYR2L5CB5/graph.json","fetch_events":"https://pith.science/api/pith-number/LLIL342SEOICJPBN4KYR2L5CB5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LLIL342SEOICJPBN4KYR2L5CB5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LLIL342SEOICJPBN4KYR2L5CB5/action/storage_attestation","attest_author":"https://pith.science/pith/LLIL342SEOICJPBN4KYR2L5CB5/action/author_attestation","sign_citation":"https://pith.science/pith/LLIL342SEOICJPBN4KYR2L5CB5/action/citation_signature","submit_replication":"https://pith.science/pith/LLIL342SEOICJPBN4KYR2L5CB5/action/replication_record"}},"created_at":"2026-07-05T09:32:44.704036+00:00","updated_at":"2026-07-05T09:32:44.704036+00:00"}