{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:VOZ4QDN7NXQO42JDYCLFYYSNL2","short_pith_number":"pith:VOZ4QDN7","schema_version":"1.0","canonical_sha256":"abb3c80dbf6de0ee6923c0965c624d5eaf904741479b3b6d84082669e951cef8","source":{"kind":"arxiv","id":"2606.03928","version":1},"attestation_state":"computed","paper":{"title":"Value-Aware Stochastic KV Cache Eviction for Reasoning Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Chenghao Yang, Deqing Fu, Harvey Yiyun Fu, Jesse Thomason, Robin Jia, Ting-Yun Chang","submitted_at":"2026-06-02T17:16:33Z","abstract_excerpt":"Reasoning models improve accuracy through extended chains of thought, but their long outputs create a memory and compute bottleneck. KV cache eviction methods reduce this cost by evicting unimportant key-value pairs from the cache, yet they often yield worse accuracy than selection-based sparse attention alternatives, which keep the full KV cache. We identify key factors crucial to KV cache eviction accuracy. First, a small fraction of value states have abnormally large magnitudes, and evicting them causes catastrophic failure where models enter repetitive reasoning loops. Second, introducing "},"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":"2606.03928","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-02T17:16:33Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"67684df13cd138e2c4503ad0542296756520316ef52765c3c8aa02e5ea863473","abstract_canon_sha256":"16ed3cddccdfcad5caa9f09d2642964107c1be5055c23bf4d088d6713e813466"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-03T02:06:07.188815Z","signature_b64":"49KrcFyNt1rBBnmGAh7vbwCCAHmScBk8VnvA03Qlzs8NzqULqNApE0Vl+2FVIoiGC+9vfCs6aDgBlQjaEqnFBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"abb3c80dbf6de0ee6923c0965c624d5eaf904741479b3b6d84082669e951cef8","last_reissued_at":"2026-06-03T02:06:07.188432Z","signature_status":"signed_v1","first_computed_at":"2026-06-03T02:06:07.188432Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Value-Aware Stochastic KV Cache Eviction for Reasoning Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Chenghao Yang, Deqing Fu, Harvey Yiyun Fu, Jesse Thomason, Robin Jia, Ting-Yun Chang","submitted_at":"2026-06-02T17:16:33Z","abstract_excerpt":"Reasoning models improve accuracy through extended chains of thought, but their long outputs create a memory and compute bottleneck. KV cache eviction methods reduce this cost by evicting unimportant key-value pairs from the cache, yet they often yield worse accuracy than selection-based sparse attention alternatives, which keep the full KV cache. We identify key factors crucial to KV cache eviction accuracy. First, a small fraction of value states have abnormally large magnitudes, and evicting them causes catastrophic failure where models enter repetitive reasoning loops. Second, introducing "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.03928","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/2606.03928/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":"2606.03928","created_at":"2026-06-03T02:06:07.188486+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.03928v1","created_at":"2026-06-03T02:06:07.188486+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.03928","created_at":"2026-06-03T02:06:07.188486+00:00"},{"alias_kind":"pith_short_12","alias_value":"VOZ4QDN7NXQO","created_at":"2026-06-03T02:06:07.188486+00:00"},{"alias_kind":"pith_short_16","alias_value":"VOZ4QDN7NXQO42JD","created_at":"2026-06-03T02:06:07.188486+00:00"},{"alias_kind":"pith_short_8","alias_value":"VOZ4QDN7","created_at":"2026-06-03T02:06:07.188486+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/VOZ4QDN7NXQO42JDYCLFYYSNL2","json":"https://pith.science/pith/VOZ4QDN7NXQO42JDYCLFYYSNL2.json","graph_json":"https://pith.science/api/pith-number/VOZ4QDN7NXQO42JDYCLFYYSNL2/graph.json","events_json":"https://pith.science/api/pith-number/VOZ4QDN7NXQO42JDYCLFYYSNL2/events.json","paper":"https://pith.science/paper/VOZ4QDN7"},"agent_actions":{"view_html":"https://pith.science/pith/VOZ4QDN7NXQO42JDYCLFYYSNL2","download_json":"https://pith.science/pith/VOZ4QDN7NXQO42JDYCLFYYSNL2.json","view_paper":"https://pith.science/paper/VOZ4QDN7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.03928&json=true","fetch_graph":"https://pith.science/api/pith-number/VOZ4QDN7NXQO42JDYCLFYYSNL2/graph.json","fetch_events":"https://pith.science/api/pith-number/VOZ4QDN7NXQO42JDYCLFYYSNL2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VOZ4QDN7NXQO42JDYCLFYYSNL2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VOZ4QDN7NXQO42JDYCLFYYSNL2/action/storage_attestation","attest_author":"https://pith.science/pith/VOZ4QDN7NXQO42JDYCLFYYSNL2/action/author_attestation","sign_citation":"https://pith.science/pith/VOZ4QDN7NXQO42JDYCLFYYSNL2/action/citation_signature","submit_replication":"https://pith.science/pith/VOZ4QDN7NXQO42JDYCLFYYSNL2/action/replication_record"}},"created_at":"2026-06-03T02:06:07.188486+00:00","updated_at":"2026-06-03T02:06:07.188486+00:00"}