{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:35IMJO3M5NWDG6O4MQP6PFWRIL","short_pith_number":"pith:35IMJO3M","schema_version":"1.0","canonical_sha256":"df50c4bb6ceb6c3379dc641fe796d142ed71f5f9df2ff547c192d056859f78b1","source":{"kind":"arxiv","id":"2607.22586","version":1},"attestation_state":"computed","paper":{"title":"MM-ShiftKV: Decode-Aware Prefill-Stage KV Selection for Multimodal Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.AI","authors_text":"Baokun Wang, Chenyang Wu, Jinsong Shu, Lidan Shou, Zhongle Xie","submitted_at":"2026-06-09T04:56:59Z","abstract_excerpt":"Key-Value (KV) caching is essential for efficient inference in multimodal large language models (MLLMs), yet its memory footprint grows linearly with context length and becomes a major bottleneck due to the large number of visual tokens. Recent prefill-stage KV selection methods estimate KV importance from prefilling statistics, implicitly assuming that prefilling-time queries are representative of those encountered during decoding. We show that this assumption breaks down in multimodal inference, where decoding-time queries exhibit substantially larger variance than prefilling-stage represent"},"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":"2607.22586","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-06-09T04:56:59Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"7d81c94bd5bbaffa07f572d78df2f6be915a365d3559036558d6033955629ba3","abstract_canon_sha256":"705360536bd27735f90aa5d042bb10293e9d179221ce61f9362a8bd22753d824"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-28T00:21:43.607679Z","signature_b64":"e+DY0LWkOAoZxBEcKeP1VooRYoop+bpRQ+rF15Yg09c5NJKanHfUm7SxuO7iGaclKFSlEgACWcLG9Lx81rnKDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"df50c4bb6ceb6c3379dc641fe796d142ed71f5f9df2ff547c192d056859f78b1","last_reissued_at":"2026-07-28T00:21:43.606801Z","signature_status":"signed_v1","first_computed_at":"2026-07-28T00:21:43.606801Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MM-ShiftKV: Decode-Aware Prefill-Stage KV Selection for Multimodal Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.AI","authors_text":"Baokun Wang, Chenyang Wu, Jinsong Shu, Lidan Shou, Zhongle Xie","submitted_at":"2026-06-09T04:56:59Z","abstract_excerpt":"Key-Value (KV) caching is essential for efficient inference in multimodal large language models (MLLMs), yet its memory footprint grows linearly with context length and becomes a major bottleneck due to the large number of visual tokens. Recent prefill-stage KV selection methods estimate KV importance from prefilling statistics, implicitly assuming that prefilling-time queries are representative of those encountered during decoding. We show that this assumption breaks down in multimodal inference, where decoding-time queries exhibit substantially larger variance than prefilling-stage represent"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.22586","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/2607.22586/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":"2607.22586","created_at":"2026-07-28T00:21:43.607202+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.22586v1","created_at":"2026-07-28T00:21:43.607202+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.22586","created_at":"2026-07-28T00:21:43.607202+00:00"},{"alias_kind":"pith_short_12","alias_value":"35IMJO3M5NWD","created_at":"2026-07-28T00:21:43.607202+00:00"},{"alias_kind":"pith_short_16","alias_value":"35IMJO3M5NWDG6O4","created_at":"2026-07-28T00:21:43.607202+00:00"},{"alias_kind":"pith_short_8","alias_value":"35IMJO3M","created_at":"2026-07-28T00:21:43.607202+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/35IMJO3M5NWDG6O4MQP6PFWRIL","json":"https://pith.science/pith/35IMJO3M5NWDG6O4MQP6PFWRIL.json","graph_json":"https://pith.science/api/pith-number/35IMJO3M5NWDG6O4MQP6PFWRIL/graph.json","events_json":"https://pith.science/api/pith-number/35IMJO3M5NWDG6O4MQP6PFWRIL/events.json","paper":"https://pith.science/paper/35IMJO3M"},"agent_actions":{"view_html":"https://pith.science/pith/35IMJO3M5NWDG6O4MQP6PFWRIL","download_json":"https://pith.science/pith/35IMJO3M5NWDG6O4MQP6PFWRIL.json","view_paper":"https://pith.science/paper/35IMJO3M","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.22586&json=true","fetch_graph":"https://pith.science/api/pith-number/35IMJO3M5NWDG6O4MQP6PFWRIL/graph.json","fetch_events":"https://pith.science/api/pith-number/35IMJO3M5NWDG6O4MQP6PFWRIL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/35IMJO3M5NWDG6O4MQP6PFWRIL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/35IMJO3M5NWDG6O4MQP6PFWRIL/action/storage_attestation","attest_author":"https://pith.science/pith/35IMJO3M5NWDG6O4MQP6PFWRIL/action/author_attestation","sign_citation":"https://pith.science/pith/35IMJO3M5NWDG6O4MQP6PFWRIL/action/citation_signature","submit_replication":"https://pith.science/pith/35IMJO3M5NWDG6O4MQP6PFWRIL/action/replication_record"}},"created_at":"2026-07-28T00:21:43.607202+00:00","updated_at":"2026-07-28T00:21:43.607202+00:00"}