{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:DDH2PVQMJI4IXU4722ZING4MKM","short_pith_number":"pith:DDH2PVQM","schema_version":"1.0","canonical_sha256":"18cfa7d60c4a388bd39fd6b2869b8c530d061b72c33d20d850007bafdd5bf9c9","source":{"kind":"arxiv","id":"2502.15294","version":3},"attestation_state":"computed","paper":{"title":"Round Attention: A Novel Round-Level Attention Mechanism to Accelerate LLM Inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Fan Mo, Hua Wang, Kun Cheng, Qiheng Lv, Yaohua Tang, Zhi Chen, Zhicheng Hu","submitted_at":"2025-02-21T08:40:07Z","abstract_excerpt":"The increasing context window size in large language models (LLMs) has improved their ability to handle complex, long-text tasks. However, as the conversation rounds continue, it is required to store a large amount of KV cache in GPU memory, which significantly affects the efficiency and even availability of the model serving systems. This paper analyzes dialogue data from real users on the granularity of round and discovers that the LLM inference manifests a watershed layer, after which the distribution of round-level attention shows notable similarity. Based on this, we propose Round Attenti"},"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":"2502.15294","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-02-21T08:40:07Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"22e5c2f975b8a53c124b2a2dbfc78380cd1227a6cb9715de601d1bec4c7d0808","abstract_canon_sha256":"cee816f58beacde35afee618f551e72c41877d185796e5216aad9828cb505fc0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:27:52.777528Z","signature_b64":"16hYeiIMfD3PNgSWCI1m0nzKqbzSXitjMmQMkdBG2riAAR9g0DgPyeGAavi89NlmKvrAd5c9CP40ZyuSMFmiBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"18cfa7d60c4a388bd39fd6b2869b8c530d061b72c33d20d850007bafdd5bf9c9","last_reissued_at":"2026-07-05T11:27:52.777012Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:27:52.777012Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Round Attention: A Novel Round-Level Attention Mechanism to Accelerate LLM Inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Fan Mo, Hua Wang, Kun Cheng, Qiheng Lv, Yaohua Tang, Zhi Chen, Zhicheng Hu","submitted_at":"2025-02-21T08:40:07Z","abstract_excerpt":"The increasing context window size in large language models (LLMs) has improved their ability to handle complex, long-text tasks. However, as the conversation rounds continue, it is required to store a large amount of KV cache in GPU memory, which significantly affects the efficiency and even availability of the model serving systems. This paper analyzes dialogue data from real users on the granularity of round and discovers that the LLM inference manifests a watershed layer, after which the distribution of round-level attention shows notable similarity. Based on this, we propose Round Attenti"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.15294","kind":"arxiv","version":3},"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/2502.15294/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":"2502.15294","created_at":"2026-07-05T11:27:52.777076+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.15294v3","created_at":"2026-07-05T11:27:52.777076+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.15294","created_at":"2026-07-05T11:27:52.777076+00:00"},{"alias_kind":"pith_short_12","alias_value":"DDH2PVQMJI4I","created_at":"2026-07-05T11:27:52.777076+00:00"},{"alias_kind":"pith_short_16","alias_value":"DDH2PVQMJI4IXU47","created_at":"2026-07-05T11:27:52.777076+00:00"},{"alias_kind":"pith_short_8","alias_value":"DDH2PVQM","created_at":"2026-07-05T11:27:52.777076+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/DDH2PVQMJI4IXU4722ZING4MKM","json":"https://pith.science/pith/DDH2PVQMJI4IXU4722ZING4MKM.json","graph_json":"https://pith.science/api/pith-number/DDH2PVQMJI4IXU4722ZING4MKM/graph.json","events_json":"https://pith.science/api/pith-number/DDH2PVQMJI4IXU4722ZING4MKM/events.json","paper":"https://pith.science/paper/DDH2PVQM"},"agent_actions":{"view_html":"https://pith.science/pith/DDH2PVQMJI4IXU4722ZING4MKM","download_json":"https://pith.science/pith/DDH2PVQMJI4IXU4722ZING4MKM.json","view_paper":"https://pith.science/paper/DDH2PVQM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.15294&json=true","fetch_graph":"https://pith.science/api/pith-number/DDH2PVQMJI4IXU4722ZING4MKM/graph.json","fetch_events":"https://pith.science/api/pith-number/DDH2PVQMJI4IXU4722ZING4MKM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DDH2PVQMJI4IXU4722ZING4MKM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DDH2PVQMJI4IXU4722ZING4MKM/action/storage_attestation","attest_author":"https://pith.science/pith/DDH2PVQMJI4IXU4722ZING4MKM/action/author_attestation","sign_citation":"https://pith.science/pith/DDH2PVQMJI4IXU4722ZING4MKM/action/citation_signature","submit_replication":"https://pith.science/pith/DDH2PVQMJI4IXU4722ZING4MKM/action/replication_record"}},"created_at":"2026-07-05T11:27:52.777076+00:00","updated_at":"2026-07-05T11:27:52.777076+00:00"}