{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:RBZF6FO757MOLOM7DAE6HQZRET","short_pith_number":"pith:RBZF6FO7","schema_version":"1.0","canonical_sha256":"88725f15dfefd8e5b99f1809e3c33124f2687403b66c51a19ad2b51f5b2e5cdf","source":{"kind":"arxiv","id":"2507.19823","version":1},"attestation_state":"computed","paper":{"title":"HCAttention: Extreme KV Cache Compression via Heterogeneous Attention Computing for LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Dongquan Yang, Rong Xiao, Xianbiao Qi, Xiaotian Yu, Yifan Yang","submitted_at":"2025-07-26T06:43:14Z","abstract_excerpt":"Processing long-context inputs with large language models presents a significant challenge due to the enormous memory requirements of the Key-Value (KV) cache during inference. Existing KV cache compression methods exhibit noticeable performance degradation when memory is reduced by more than 85%. Additionally, strategies that leverage GPU-CPU collaboration for approximate attention remain underexplored in this setting. We propose HCAttention, a heterogeneous attention computation framework that integrates key quantization, value offloading, and dynamic KV eviction to enable efficient inferenc"},"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":"2507.19823","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-07-26T06:43:14Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"94fd42ea5f7adcd0d2c3840aeb2512a1b0af1d4af7efae8f89bd05ea195eec67","abstract_canon_sha256":"3c02d56b6ca32635aff58691eb35b3586b227b47e076c48fba5d7dbe26529972"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:43:42.702404Z","signature_b64":"DzxRyt1oh3Jlk+stSCtGVTHnbyMii8tJdoV1vDjEE+S2zvRPFh1Idc6r9l3867gsoBdlcnU5XJmD5gf3kadyBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"88725f15dfefd8e5b99f1809e3c33124f2687403b66c51a19ad2b51f5b2e5cdf","last_reissued_at":"2026-07-05T11:43:42.701948Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:43:42.701948Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"HCAttention: Extreme KV Cache Compression via Heterogeneous Attention Computing for LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Dongquan Yang, Rong Xiao, Xianbiao Qi, Xiaotian Yu, Yifan Yang","submitted_at":"2025-07-26T06:43:14Z","abstract_excerpt":"Processing long-context inputs with large language models presents a significant challenge due to the enormous memory requirements of the Key-Value (KV) cache during inference. Existing KV cache compression methods exhibit noticeable performance degradation when memory is reduced by more than 85%. Additionally, strategies that leverage GPU-CPU collaboration for approximate attention remain underexplored in this setting. We propose HCAttention, a heterogeneous attention computation framework that integrates key quantization, value offloading, and dynamic KV eviction to enable efficient inferenc"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.19823","kind":"arxiv","version":1},"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/2507.19823/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":"2507.19823","created_at":"2026-07-05T11:43:42.702005+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.19823v1","created_at":"2026-07-05T11:43:42.702005+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.19823","created_at":"2026-07-05T11:43:42.702005+00:00"},{"alias_kind":"pith_short_12","alias_value":"RBZF6FO757MO","created_at":"2026-07-05T11:43:42.702005+00:00"},{"alias_kind":"pith_short_16","alias_value":"RBZF6FO757MOLOM7","created_at":"2026-07-05T11:43:42.702005+00:00"},{"alias_kind":"pith_short_8","alias_value":"RBZF6FO7","created_at":"2026-07-05T11:43:42.702005+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/RBZF6FO757MOLOM7DAE6HQZRET","json":"https://pith.science/pith/RBZF6FO757MOLOM7DAE6HQZRET.json","graph_json":"https://pith.science/api/pith-number/RBZF6FO757MOLOM7DAE6HQZRET/graph.json","events_json":"https://pith.science/api/pith-number/RBZF6FO757MOLOM7DAE6HQZRET/events.json","paper":"https://pith.science/paper/RBZF6FO7"},"agent_actions":{"view_html":"https://pith.science/pith/RBZF6FO757MOLOM7DAE6HQZRET","download_json":"https://pith.science/pith/RBZF6FO757MOLOM7DAE6HQZRET.json","view_paper":"https://pith.science/paper/RBZF6FO7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.19823&json=true","fetch_graph":"https://pith.science/api/pith-number/RBZF6FO757MOLOM7DAE6HQZRET/graph.json","fetch_events":"https://pith.science/api/pith-number/RBZF6FO757MOLOM7DAE6HQZRET/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RBZF6FO757MOLOM7DAE6HQZRET/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RBZF6FO757MOLOM7DAE6HQZRET/action/storage_attestation","attest_author":"https://pith.science/pith/RBZF6FO757MOLOM7DAE6HQZRET/action/author_attestation","sign_citation":"https://pith.science/pith/RBZF6FO757MOLOM7DAE6HQZRET/action/citation_signature","submit_replication":"https://pith.science/pith/RBZF6FO757MOLOM7DAE6HQZRET/action/replication_record"}},"created_at":"2026-07-05T11:43:42.702005+00:00","updated_at":"2026-07-05T11:43:42.702005+00:00"}