{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:Y3B5GQDPTUO73I6I7KADHF5YJ2","short_pith_number":"pith:Y3B5GQDP","schema_version":"1.0","canonical_sha256":"c6c3d3406f9d1dfda3c8fa803397b84eaefd7c3e2afcb5e467e7047d4ded6558","source":{"kind":"arxiv","id":"2605.19218","version":1},"attestation_state":"computed","paper":{"title":"Rotation-Aligned Key Channel Pruning for Efficient Vision-Language Model Inference","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Beomseok Kang, Donghwee Son, Dongwon Jo, Jae-Joon Kim, Jiwon Song","submitted_at":"2026-05-19T00:45:00Z","abstract_excerpt":"Vision-Language Models suffer severe KV cache pressure at inference, as a single image often encodes into thousands of tokens. Most existing methods exploit token sparsity through token pruning, but permanently discarding visual content causes substantial degradation on fine-grained perception tasks. This motivates a complementary axis, feature sparsity: under a fixed KV cache budget, compressing the channel dimension preserves more visual tokens at the same memory cost. Prior Key channel pruning methods, however, face a structural trade-off: token-wise channel pruning is expressive but unstru"},"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":"2605.19218","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-19T00:45:00Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"3101ff104d698e9ba521a9468c6e8dfbdd93a007eff48cff30dac2ed4e5567b8","abstract_canon_sha256":"8bae728b7e00286168a436e39e3b5eb7980bcbaa4cf9c99fe7fb09303e54e53c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T01:05:34.045271Z","signature_b64":"9KRvQugTZfkfk4e67Wl8hB90Kc8tmnmyaMIZlXWkI7m8U5fYySY3TxoO6fvQKLi+xHyL+zr5C58H/KEmZFYXAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c6c3d3406f9d1dfda3c8fa803397b84eaefd7c3e2afcb5e467e7047d4ded6558","last_reissued_at":"2026-05-20T01:05:34.044651Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T01:05:34.044651Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Rotation-Aligned Key Channel Pruning for Efficient Vision-Language Model Inference","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Beomseok Kang, Donghwee Son, Dongwon Jo, Jae-Joon Kim, Jiwon Song","submitted_at":"2026-05-19T00:45:00Z","abstract_excerpt":"Vision-Language Models suffer severe KV cache pressure at inference, as a single image often encodes into thousands of tokens. Most existing methods exploit token sparsity through token pruning, but permanently discarding visual content causes substantial degradation on fine-grained perception tasks. This motivates a complementary axis, feature sparsity: under a fixed KV cache budget, compressing the channel dimension preserves more visual tokens at the same memory cost. Prior Key channel pruning methods, however, face a structural trade-off: token-wise channel pruning is expressive but unstru"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.19218","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/2605.19218/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":"2605.19218","created_at":"2026-05-20T01:05:34.044767+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.19218v1","created_at":"2026-05-20T01:05:34.044767+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.19218","created_at":"2026-05-20T01:05:34.044767+00:00"},{"alias_kind":"pith_short_12","alias_value":"Y3B5GQDPTUO7","created_at":"2026-05-20T01:05:34.044767+00:00"},{"alias_kind":"pith_short_16","alias_value":"Y3B5GQDPTUO73I6I","created_at":"2026-05-20T01:05:34.044767+00:00"},{"alias_kind":"pith_short_8","alias_value":"Y3B5GQDP","created_at":"2026-05-20T01:05:34.044767+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/Y3B5GQDPTUO73I6I7KADHF5YJ2","json":"https://pith.science/pith/Y3B5GQDPTUO73I6I7KADHF5YJ2.json","graph_json":"https://pith.science/api/pith-number/Y3B5GQDPTUO73I6I7KADHF5YJ2/graph.json","events_json":"https://pith.science/api/pith-number/Y3B5GQDPTUO73I6I7KADHF5YJ2/events.json","paper":"https://pith.science/paper/Y3B5GQDP"},"agent_actions":{"view_html":"https://pith.science/pith/Y3B5GQDPTUO73I6I7KADHF5YJ2","download_json":"https://pith.science/pith/Y3B5GQDPTUO73I6I7KADHF5YJ2.json","view_paper":"https://pith.science/paper/Y3B5GQDP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.19218&json=true","fetch_graph":"https://pith.science/api/pith-number/Y3B5GQDPTUO73I6I7KADHF5YJ2/graph.json","fetch_events":"https://pith.science/api/pith-number/Y3B5GQDPTUO73I6I7KADHF5YJ2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Y3B5GQDPTUO73I6I7KADHF5YJ2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Y3B5GQDPTUO73I6I7KADHF5YJ2/action/storage_attestation","attest_author":"https://pith.science/pith/Y3B5GQDPTUO73I6I7KADHF5YJ2/action/author_attestation","sign_citation":"https://pith.science/pith/Y3B5GQDPTUO73I6I7KADHF5YJ2/action/citation_signature","submit_replication":"https://pith.science/pith/Y3B5GQDPTUO73I6I7KADHF5YJ2/action/replication_record"}},"created_at":"2026-05-20T01:05:34.044767+00:00","updated_at":"2026-05-20T01:05:34.044767+00:00"}