{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:W7PANH42X2LAYOAA6ZLN7HYEU6","short_pith_number":"pith:W7PANH42","schema_version":"1.0","canonical_sha256":"b7de069f9abe960c3800f656df9f04a7a6ba8ec44d65badb98e76505a422027a","source":{"kind":"arxiv","id":"2605.24786","version":1},"attestation_state":"computed","paper":{"title":"CONF-KV: Confidence-Aware KV Cache Eviction with Mixed-Precision Storage for Long-Horizon LLM","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Yidi Miao, Yubo Li","submitted_at":"2026-05-24T00:15:34Z","abstract_excerpt":"Long-horizon LLM inference turns the key--value (KV) cache into the dominant GPU memory consumer and makes per-token attention increasingly expensive. Many common eviction policies use static recency windows or historical attention, leaving unused a signal computed on every decoding step: the model's current uncertainty. We introduce CONF-KV, a KV-cache manager that converts the next-token distribution into a scalar confidence score and uses it to choose the per-step cache budget, retaining more context when the model is uncertain and pruning aggressively when it is confident. Within each budg"},"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":"2605.24786","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-24T00:15:34Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"4a3d11716a3208320d9c739b6f68f333cde3a98117c118e580b507fcd70cca54","abstract_canon_sha256":"1b5cd4e63fb65f32541b0c1047349734e932c36e4d5532645fbdfb0f822cb3e8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T01:03:58.042547Z","signature_b64":"HWBwqi7Z4222rcmFlmJVrRE4TIQWZGozXNxdM6A4w17lQbhyw65TpRyJvz6UfMMJnE5joc//PjcAp5XV+3ISBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b7de069f9abe960c3800f656df9f04a7a6ba8ec44d65badb98e76505a422027a","last_reissued_at":"2026-05-26T01:03:58.041750Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T01:03:58.041750Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"CONF-KV: Confidence-Aware KV Cache Eviction with Mixed-Precision Storage for Long-Horizon LLM","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Yidi Miao, Yubo Li","submitted_at":"2026-05-24T00:15:34Z","abstract_excerpt":"Long-horizon LLM inference turns the key--value (KV) cache into the dominant GPU memory consumer and makes per-token attention increasingly expensive. Many common eviction policies use static recency windows or historical attention, leaving unused a signal computed on every decoding step: the model's current uncertainty. We introduce CONF-KV, a KV-cache manager that converts the next-token distribution into a scalar confidence score and uses it to choose the per-step cache budget, retaining more context when the model is uncertain and pruning aggressively when it is confident. Within each budg"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.24786","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/2605.24786/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":"2605.24786","created_at":"2026-05-26T01:03:58.041878+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.24786v1","created_at":"2026-05-26T01:03:58.041878+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.24786","created_at":"2026-05-26T01:03:58.041878+00:00"},{"alias_kind":"pith_short_12","alias_value":"W7PANH42X2LA","created_at":"2026-05-26T01:03:58.041878+00:00"},{"alias_kind":"pith_short_16","alias_value":"W7PANH42X2LAYOAA","created_at":"2026-05-26T01:03:58.041878+00:00"},{"alias_kind":"pith_short_8","alias_value":"W7PANH42","created_at":"2026-05-26T01:03:58.041878+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/W7PANH42X2LAYOAA6ZLN7HYEU6","json":"https://pith.science/pith/W7PANH42X2LAYOAA6ZLN7HYEU6.json","graph_json":"https://pith.science/api/pith-number/W7PANH42X2LAYOAA6ZLN7HYEU6/graph.json","events_json":"https://pith.science/api/pith-number/W7PANH42X2LAYOAA6ZLN7HYEU6/events.json","paper":"https://pith.science/paper/W7PANH42"},"agent_actions":{"view_html":"https://pith.science/pith/W7PANH42X2LAYOAA6ZLN7HYEU6","download_json":"https://pith.science/pith/W7PANH42X2LAYOAA6ZLN7HYEU6.json","view_paper":"https://pith.science/paper/W7PANH42","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.24786&json=true","fetch_graph":"https://pith.science/api/pith-number/W7PANH42X2LAYOAA6ZLN7HYEU6/graph.json","fetch_events":"https://pith.science/api/pith-number/W7PANH42X2LAYOAA6ZLN7HYEU6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/W7PANH42X2LAYOAA6ZLN7HYEU6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/W7PANH42X2LAYOAA6ZLN7HYEU6/action/storage_attestation","attest_author":"https://pith.science/pith/W7PANH42X2LAYOAA6ZLN7HYEU6/action/author_attestation","sign_citation":"https://pith.science/pith/W7PANH42X2LAYOAA6ZLN7HYEU6/action/citation_signature","submit_replication":"https://pith.science/pith/W7PANH42X2LAYOAA6ZLN7HYEU6/action/replication_record"}},"created_at":"2026-05-26T01:03:58.041878+00:00","updated_at":"2026-05-26T01:03:58.041878+00:00"}