{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:NRVSJVMVLDSZQKTOIGB6WKQD6M","short_pith_number":"pith:NRVSJVMV","schema_version":"1.0","canonical_sha256":"6c6b24d59558e5982a6e4183eb2a03f3328758744d8cc75c137f9254f2cc51e3","source":{"kind":"arxiv","id":"2607.16213","version":1},"attestation_state":"computed","paper":{"title":"SelKV: Selective KV Cache Merging with Per-Token Merge-or-Drop and Attention Compensation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Aicha Boutorh, Manel Kara laouar, Mohamed Hadj Ameur, Soumia Bouyahiaoui","submitted_at":"2026-05-19T17:44:56Z","abstract_excerpt":"Large Language Models (LLMs) generate text autoregressively, relying on a key-value (KV) cache whose memory footprint grows linearly with context length, creating a major bottleneck. Recent compression methods mitigate this cost via token merging; however, these approaches often rely on indiscriminate aggregation, which degrades representations and introduces attention sag, a mismatch where merged tokens receive the same softmax mass as individual tokens despite encoding multiple inputs. We propose a training-free, dual-component framework for KV cache compression that addresses these limitati"},"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":"2607.16213","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-19T17:44:56Z","cross_cats_sorted":[],"title_canon_sha256":"555674cb21c98630b17b1ad510dddd0767aa711e5e146317e6f5377ce8d2919a","abstract_canon_sha256":"34b2975ddc76ec229dabf8a2ed476ab1701326070811cada3a63c6523af95d1d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-21T00:20:06.079867Z","signature_b64":"BQZ9dbpXPQ40f/oiM0RdC/EjVQaqFFQP7HWiWPqkAWb1xjDyr22VaXZiymnua7KYVI3bf/dKtfbq42zcKCVCAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6c6b24d59558e5982a6e4183eb2a03f3328758744d8cc75c137f9254f2cc51e3","last_reissued_at":"2026-07-21T00:20:06.078959Z","signature_status":"signed_v1","first_computed_at":"2026-07-21T00:20:06.078959Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SelKV: Selective KV Cache Merging with Per-Token Merge-or-Drop and Attention Compensation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Aicha Boutorh, Manel Kara laouar, Mohamed Hadj Ameur, Soumia Bouyahiaoui","submitted_at":"2026-05-19T17:44:56Z","abstract_excerpt":"Large Language Models (LLMs) generate text autoregressively, relying on a key-value (KV) cache whose memory footprint grows linearly with context length, creating a major bottleneck. Recent compression methods mitigate this cost via token merging; however, these approaches often rely on indiscriminate aggregation, which degrades representations and introduces attention sag, a mismatch where merged tokens receive the same softmax mass as individual tokens despite encoding multiple inputs. We propose a training-free, dual-component framework for KV cache compression that addresses these limitati"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.16213","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/2607.16213/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":"2607.16213","created_at":"2026-07-21T00:20:06.079440+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.16213v1","created_at":"2026-07-21T00:20:06.079440+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.16213","created_at":"2026-07-21T00:20:06.079440+00:00"},{"alias_kind":"pith_short_12","alias_value":"NRVSJVMVLDSZ","created_at":"2026-07-21T00:20:06.079440+00:00"},{"alias_kind":"pith_short_16","alias_value":"NRVSJVMVLDSZQKTO","created_at":"2026-07-21T00:20:06.079440+00:00"},{"alias_kind":"pith_short_8","alias_value":"NRVSJVMV","created_at":"2026-07-21T00:20:06.079440+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/NRVSJVMVLDSZQKTOIGB6WKQD6M","json":"https://pith.science/pith/NRVSJVMVLDSZQKTOIGB6WKQD6M.json","graph_json":"https://pith.science/api/pith-number/NRVSJVMVLDSZQKTOIGB6WKQD6M/graph.json","events_json":"https://pith.science/api/pith-number/NRVSJVMVLDSZQKTOIGB6WKQD6M/events.json","paper":"https://pith.science/paper/NRVSJVMV"},"agent_actions":{"view_html":"https://pith.science/pith/NRVSJVMVLDSZQKTOIGB6WKQD6M","download_json":"https://pith.science/pith/NRVSJVMVLDSZQKTOIGB6WKQD6M.json","view_paper":"https://pith.science/paper/NRVSJVMV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.16213&json=true","fetch_graph":"https://pith.science/api/pith-number/NRVSJVMVLDSZQKTOIGB6WKQD6M/graph.json","fetch_events":"https://pith.science/api/pith-number/NRVSJVMVLDSZQKTOIGB6WKQD6M/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NRVSJVMVLDSZQKTOIGB6WKQD6M/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NRVSJVMVLDSZQKTOIGB6WKQD6M/action/storage_attestation","attest_author":"https://pith.science/pith/NRVSJVMVLDSZQKTOIGB6WKQD6M/action/author_attestation","sign_citation":"https://pith.science/pith/NRVSJVMVLDSZQKTOIGB6WKQD6M/action/citation_signature","submit_replication":"https://pith.science/pith/NRVSJVMVLDSZQKTOIGB6WKQD6M/action/replication_record"}},"created_at":"2026-07-21T00:20:06.079440+00:00","updated_at":"2026-07-21T00:20:06.079440+00:00"}