{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:3GWDRGDSWSUKMPROHMEA6JX73V","short_pith_number":"pith:3GWDRGDS","schema_version":"1.0","canonical_sha256":"d9ac389872b4a8a63e2e3b080f26ffdd7115d37b3085f6d06986b70d06ce0f60","source":{"kind":"arxiv","id":"2402.15220","version":4},"attestation_state":"computed","paper":{"title":"ChunkAttention: Efficient Self-Attention with Prefix-Aware KV Cache and Two-Phase Partition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Lu Ye, Yang Li, Yong Huang, Ze Tao","submitted_at":"2024-02-23T09:29:19Z","abstract_excerpt":"Self-attention is an essential component of large language models (LLM) but a significant source of inference latency for long sequences. In multi-tenant LLM serving scenarios, the compute and memory operation cost of self-attention can be optimized by using the probability that multiple LLM requests have shared system prompts in prefixes. In this paper, we introduce ChunkAttention, a prefix-aware self-attention module that can detect matching prompt prefixes across multiple requests and share their key/value tensors in memory at runtime to improve the memory utilization of KV cache. This is a"},"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":"2402.15220","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-23T09:29:19Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"18e0f512eee20ff73bacd1ab420b377f3a43c0e666d13cc57db0bb76be15c9d9","abstract_canon_sha256":"f8c7e03602fc4c92ec21fee827fbf7a7f3ec5bcc9f1d4c24785496076435b02a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:50:54.104308Z","signature_b64":"KhOpCmCG+fhl7+qONh3HiVget7NBOLS7bg9AN7A+tG3vxoy5cC0zuxdmaRgTJ1d/Zxon9OgUGB3vCb25UBwfAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d9ac389872b4a8a63e2e3b080f26ffdd7115d37b3085f6d06986b70d06ce0f60","last_reissued_at":"2026-07-05T08:50:54.103899Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:50:54.103899Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ChunkAttention: Efficient Self-Attention with Prefix-Aware KV Cache and Two-Phase Partition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Lu Ye, Yang Li, Yong Huang, Ze Tao","submitted_at":"2024-02-23T09:29:19Z","abstract_excerpt":"Self-attention is an essential component of large language models (LLM) but a significant source of inference latency for long sequences. In multi-tenant LLM serving scenarios, the compute and memory operation cost of self-attention can be optimized by using the probability that multiple LLM requests have shared system prompts in prefixes. In this paper, we introduce ChunkAttention, a prefix-aware self-attention module that can detect matching prompt prefixes across multiple requests and share their key/value tensors in memory at runtime to improve the memory utilization of KV cache. This is a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.15220","kind":"arxiv","version":4},"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/2402.15220/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":"2402.15220","created_at":"2026-07-05T08:50:54.103951+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.15220v4","created_at":"2026-07-05T08:50:54.103951+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.15220","created_at":"2026-07-05T08:50:54.103951+00:00"},{"alias_kind":"pith_short_12","alias_value":"3GWDRGDSWSUK","created_at":"2026-07-05T08:50:54.103951+00:00"},{"alias_kind":"pith_short_16","alias_value":"3GWDRGDSWSUKMPRO","created_at":"2026-07-05T08:50:54.103951+00:00"},{"alias_kind":"pith_short_8","alias_value":"3GWDRGDS","created_at":"2026-07-05T08:50:54.103951+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.09735","citing_title":"KV-RM: Regularizing KV-Cache Movement for Static-Graph LLM Serving","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2504.15965","citing_title":"From Human Memory to AI Memory: A Survey on Memory Mechanisms in the Era of LLMs","ref_index":130,"is_internal_anchor":false},{"citing_arxiv_id":"2602.09725","citing_title":"Efficient Remote KV Cache Reuse with GPU-native Video Codec","ref_index":73,"is_internal_anchor":false},{"citing_arxiv_id":"2312.07104","citing_title":"SGLang: Efficient Execution of Structured Language Model Programs","ref_index":58,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09735","citing_title":"KV-RM: Regularizing KV-Cache Movement for Static-Graph LLM Serving","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03375","citing_title":"Tutti: Making SSD-Backed KV Cache Practical for Long-Context LLM Serving","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05219","citing_title":"Sparse Prefix Caching for Hybrid and Recurrent LLM Serving","ref_index":35,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3GWDRGDSWSUKMPROHMEA6JX73V","json":"https://pith.science/pith/3GWDRGDSWSUKMPROHMEA6JX73V.json","graph_json":"https://pith.science/api/pith-number/3GWDRGDSWSUKMPROHMEA6JX73V/graph.json","events_json":"https://pith.science/api/pith-number/3GWDRGDSWSUKMPROHMEA6JX73V/events.json","paper":"https://pith.science/paper/3GWDRGDS"},"agent_actions":{"view_html":"https://pith.science/pith/3GWDRGDSWSUKMPROHMEA6JX73V","download_json":"https://pith.science/pith/3GWDRGDSWSUKMPROHMEA6JX73V.json","view_paper":"https://pith.science/paper/3GWDRGDS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.15220&json=true","fetch_graph":"https://pith.science/api/pith-number/3GWDRGDSWSUKMPROHMEA6JX73V/graph.json","fetch_events":"https://pith.science/api/pith-number/3GWDRGDSWSUKMPROHMEA6JX73V/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3GWDRGDSWSUKMPROHMEA6JX73V/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3GWDRGDSWSUKMPROHMEA6JX73V/action/storage_attestation","attest_author":"https://pith.science/pith/3GWDRGDSWSUKMPROHMEA6JX73V/action/author_attestation","sign_citation":"https://pith.science/pith/3GWDRGDSWSUKMPROHMEA6JX73V/action/citation_signature","submit_replication":"https://pith.science/pith/3GWDRGDSWSUKMPROHMEA6JX73V/action/replication_record"}},"created_at":"2026-07-05T08:50:54.103951+00:00","updated_at":"2026-07-05T08:50:54.103951+00:00"}