{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:WVFESS5RGKUIUV3BPHPOQYL3MC","short_pith_number":"pith:WVFESS5R","schema_version":"1.0","canonical_sha256":"b54a494bb132a88a576179dee8617b6095c4aba513897f170d1a0fe981e8b3df","source":{"kind":"arxiv","id":"2407.15891","version":1},"attestation_state":"computed","paper":{"title":"RazorAttention: Efficient KV Cache Compression Through Retrieval Heads","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Gongyi Wang, Hanlin Tang, Jing Lin, Qingsen Han, Shikuan Hong, Yang Lin, Yiwu Yao","submitted_at":"2024-07-22T01:12:23Z","abstract_excerpt":"The memory and computational demands of Key-Value (KV) cache present significant challenges for deploying long-context language models. Previous approaches attempt to mitigate this issue by selectively dropping tokens, which irreversibly erases critical information that might be needed for future queries. In this paper, we propose a novel compression technique for KV cache that preserves all token information. Our investigation reveals that: i) Most attention heads primarily focus on the local context; ii) Only a few heads, denoted as retrieval heads, can essentially pay attention to all input"},"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":"2407.15891","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-07-22T01:12:23Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"d937b3121906bf9c32e55e299c06591c21c0141665832eaf2c790b4b6dc91e8d","abstract_canon_sha256":"458676bf91e3834af77f49142206e99aee8604a04f160f3aef8667e257b4be92"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:47:23.987835Z","signature_b64":"dr5T2hfLY811zfpbsv53gq38CfPjhMe0YJ/rdR+Hr0SqtlMy1GNJYjGFdVWWenNHMBqW2ZQuVxGZBqtoYsuYAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b54a494bb132a88a576179dee8617b6095c4aba513897f170d1a0fe981e8b3df","last_reissued_at":"2026-07-05T08:47:23.987355Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:47:23.987355Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RazorAttention: Efficient KV Cache Compression Through Retrieval Heads","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Gongyi Wang, Hanlin Tang, Jing Lin, Qingsen Han, Shikuan Hong, Yang Lin, Yiwu Yao","submitted_at":"2024-07-22T01:12:23Z","abstract_excerpt":"The memory and computational demands of Key-Value (KV) cache present significant challenges for deploying long-context language models. Previous approaches attempt to mitigate this issue by selectively dropping tokens, which irreversibly erases critical information that might be needed for future queries. In this paper, we propose a novel compression technique for KV cache that preserves all token information. Our investigation reveals that: i) Most attention heads primarily focus on the local context; ii) Only a few heads, denoted as retrieval heads, can essentially pay attention to all input"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.15891","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/2407.15891/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":"2407.15891","created_at":"2026-07-05T08:47:23.987413+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.15891v1","created_at":"2026-07-05T08:47:23.987413+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.15891","created_at":"2026-07-05T08:47:23.987413+00:00"},{"alias_kind":"pith_short_12","alias_value":"WVFESS5RGKUI","created_at":"2026-07-05T08:47:23.987413+00:00"},{"alias_kind":"pith_short_16","alias_value":"WVFESS5RGKUIUV3B","created_at":"2026-07-05T08:47:23.987413+00:00"},{"alias_kind":"pith_short_8","alias_value":"WVFESS5R","created_at":"2026-07-05T08:47:23.987413+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":11,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.08032","citing_title":"What to Keep, What to Forget: A Rate--Distortion View of Memory Compaction in LLMs and Agents","ref_index":112,"is_internal_anchor":true},{"citing_arxiv_id":"2606.24467","citing_title":"CompressKV: Semantic-Retrieval-Guided KV-Cache Compression for Resource-Efficient Long-Context LLM Inference","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28831","citing_title":"HARD-KV: Head-Adaptive Regularization for Decoding-time KV Compression","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06840","citing_title":"Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09508","citing_title":"From Rigid to Dynamic: Entropy-Guided Adaptive Inference for Long-Context LLMs","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23640","citing_title":"CachePrune: Privacy-Aware and Fine-Grained KV Cache Sharing for Efficient LLM Inference","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18053","citing_title":"Protection Is (Nearly) All You Need: Structural Protection Dominates Scoring in Globally Capped KV Eviction","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2410.10819","citing_title":"DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14037","citing_title":"Self-Pruned Key-Value Attention: Learning When to Write by Predicting Future Utility","ref_index":74,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08840","citing_title":"ReST-KV: Robust KV Cache Eviction with Layer-wise Output Reconstruction and Spatial-Temporal Smoothing","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07234","citing_title":"Reformulating KV Cache Eviction Problem for Long-Context LLM Inference","ref_index":38,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WVFESS5RGKUIUV3BPHPOQYL3MC","json":"https://pith.science/pith/WVFESS5RGKUIUV3BPHPOQYL3MC.json","graph_json":"https://pith.science/api/pith-number/WVFESS5RGKUIUV3BPHPOQYL3MC/graph.json","events_json":"https://pith.science/api/pith-number/WVFESS5RGKUIUV3BPHPOQYL3MC/events.json","paper":"https://pith.science/paper/WVFESS5R"},"agent_actions":{"view_html":"https://pith.science/pith/WVFESS5RGKUIUV3BPHPOQYL3MC","download_json":"https://pith.science/pith/WVFESS5RGKUIUV3BPHPOQYL3MC.json","view_paper":"https://pith.science/paper/WVFESS5R","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.15891&json=true","fetch_graph":"https://pith.science/api/pith-number/WVFESS5RGKUIUV3BPHPOQYL3MC/graph.json","fetch_events":"https://pith.science/api/pith-number/WVFESS5RGKUIUV3BPHPOQYL3MC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WVFESS5RGKUIUV3BPHPOQYL3MC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WVFESS5RGKUIUV3BPHPOQYL3MC/action/storage_attestation","attest_author":"https://pith.science/pith/WVFESS5RGKUIUV3BPHPOQYL3MC/action/author_attestation","sign_citation":"https://pith.science/pith/WVFESS5RGKUIUV3BPHPOQYL3MC/action/citation_signature","submit_replication":"https://pith.science/pith/WVFESS5RGKUIUV3BPHPOQYL3MC/action/replication_record"}},"created_at":"2026-07-05T08:47:23.987413+00:00","updated_at":"2026-07-05T08:47:23.987413+00:00"}