{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:CWGSNJML7ALEXV75DRW2WP553K","short_pith_number":"pith:CWGSNJML","schema_version":"1.0","canonical_sha256":"158d26a58bf8164bd7fd1c6dab3fbddaa7c08e3c106e7f7724ca576a78f5a196","source":{"kind":"arxiv","id":"2502.14051","version":3},"attestation_state":"computed","paper":{"title":"RocketKV: Accelerating Long-Context LLM Inference via Two-Stage KV Cache Compression","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Alexey Tumanov, Payman Behnam, Po-An Tsai, Ritchie Zhao, Yaosheng Fu, Zhiding Yu","submitted_at":"2025-02-19T19:12:46Z","abstract_excerpt":"Transformer-based Large Language Models rely critically on the KV cache to efficiently handle extended contexts during the decode phase. Yet, the size of the KV cache grows proportionally with the input length, burdening both memory bandwidth and capacity as decoding progresses. To address this challenge, we present RocketKV, a training-free KV cache compression strategy containing two consecutive stages. In the first stage, it performs coarse-grain permanent KV cache eviction on the input sequence tokens. In the second stage, it adopts a hybrid sparse attention method to conduct fine-grain to"},"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":"2502.14051","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-02-19T19:12:46Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"666b0681301b83ef25951109ee4cc8db730a27fea8fad840edaedf5f4cbe7231","abstract_canon_sha256":"a4882c984161c891a15011ab7c53468f238ac4e539bb50ce1c0573e1a15b5e08"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:52:59.566369Z","signature_b64":"ya7UzIjKKh++4zjwHgON6qEhd4J2DPvjFSPz+wuqDBHmNPwKkMYvgdMgSVByxHwJIbphIJkS2RuniUdLcu3rAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"158d26a58bf8164bd7fd1c6dab3fbddaa7c08e3c106e7f7724ca576a78f5a196","last_reissued_at":"2026-07-05T11:52:59.565892Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:52:59.565892Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RocketKV: Accelerating Long-Context LLM Inference via Two-Stage KV Cache Compression","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Alexey Tumanov, Payman Behnam, Po-An Tsai, Ritchie Zhao, Yaosheng Fu, Zhiding Yu","submitted_at":"2025-02-19T19:12:46Z","abstract_excerpt":"Transformer-based Large Language Models rely critically on the KV cache to efficiently handle extended contexts during the decode phase. Yet, the size of the KV cache grows proportionally with the input length, burdening both memory bandwidth and capacity as decoding progresses. To address this challenge, we present RocketKV, a training-free KV cache compression strategy containing two consecutive stages. In the first stage, it performs coarse-grain permanent KV cache eviction on the input sequence tokens. In the second stage, it adopts a hybrid sparse attention method to conduct fine-grain to"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.14051","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/2502.14051/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":"2502.14051","created_at":"2026-07-05T11:52:59.565957+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.14051v3","created_at":"2026-07-05T11:52:59.565957+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.14051","created_at":"2026-07-05T11:52:59.565957+00:00"},{"alias_kind":"pith_short_12","alias_value":"CWGSNJML7ALE","created_at":"2026-07-05T11:52:59.565957+00:00"},{"alias_kind":"pith_short_16","alias_value":"CWGSNJML7ALEXV75","created_at":"2026-07-05T11:52:59.565957+00:00"},{"alias_kind":"pith_short_8","alias_value":"CWGSNJML","created_at":"2026-07-05T11:52:59.565957+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.28831","citing_title":"HARD-KV: Head-Adaptive Regularization for Decoding-time KV Compression","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2512.12087","citing_title":"BLASST: Dynamic BLocked Attention Sparsity via Softmax Thresholding","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08580","citing_title":"Slipstream: Trajectory-Grounded Compaction Validation for Long-Horizon Agents","ref_index":62,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CWGSNJML7ALEXV75DRW2WP553K","json":"https://pith.science/pith/CWGSNJML7ALEXV75DRW2WP553K.json","graph_json":"https://pith.science/api/pith-number/CWGSNJML7ALEXV75DRW2WP553K/graph.json","events_json":"https://pith.science/api/pith-number/CWGSNJML7ALEXV75DRW2WP553K/events.json","paper":"https://pith.science/paper/CWGSNJML"},"agent_actions":{"view_html":"https://pith.science/pith/CWGSNJML7ALEXV75DRW2WP553K","download_json":"https://pith.science/pith/CWGSNJML7ALEXV75DRW2WP553K.json","view_paper":"https://pith.science/paper/CWGSNJML","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.14051&json=true","fetch_graph":"https://pith.science/api/pith-number/CWGSNJML7ALEXV75DRW2WP553K/graph.json","fetch_events":"https://pith.science/api/pith-number/CWGSNJML7ALEXV75DRW2WP553K/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CWGSNJML7ALEXV75DRW2WP553K/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CWGSNJML7ALEXV75DRW2WP553K/action/storage_attestation","attest_author":"https://pith.science/pith/CWGSNJML7ALEXV75DRW2WP553K/action/author_attestation","sign_citation":"https://pith.science/pith/CWGSNJML7ALEXV75DRW2WP553K/action/citation_signature","submit_replication":"https://pith.science/pith/CWGSNJML7ALEXV75DRW2WP553K/action/replication_record"}},"created_at":"2026-07-05T11:52:59.565957+00:00","updated_at":"2026-07-05T11:52:59.565957+00:00"}