{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:J44I5WJCLMJF7E5GRSE2V4WFFS","short_pith_number":"pith:J44I5WJC","schema_version":"1.0","canonical_sha256":"4f388ed9225b125f93a68c89aaf2c52ca9c8227b3bb0c16202fb75d3c84f6b9f","source":{"kind":"arxiv","id":"2503.04973","version":1},"attestation_state":"computed","paper":{"title":"Beyond RAG: Task-Aware KV Cache Compression for Comprehensive Knowledge Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.IR","cs.LG"],"primary_cat":"cs.CL","authors_text":"Fabio Petroni, Giulio Corallo, Orion Weller, Paolo Papotti","submitted_at":"2025-03-06T21:07:41Z","abstract_excerpt":"Incorporating external knowledge in large language models (LLMs) enhances their utility across diverse applications, but existing methods have trade-offs. Retrieval-Augmented Generation (RAG) fetches evidence via similarity search, but key information may fall outside top ranked results. Long-context models can process multiple documents but are computationally expensive and limited by context window size. Inspired by students condensing study material for open-book exams, we propose task-aware key-value (KV) cache compression, which compresses external knowledge in a zero- or few-shot setup. "},"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":"2503.04973","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-03-06T21:07:41Z","cross_cats_sorted":["cs.AI","cs.IR","cs.LG"],"title_canon_sha256":"2f39510ea485a750cc52c26c4f7992bf1b6ecb71855c00c1cdb3f7922c381f87","abstract_canon_sha256":"adcbfbdb850c6b2c73107f67846691b26d8ec438fa24099fce7dce2b46855b0e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:26:17.657233Z","signature_b64":"bd/pTwDLEpm5axcTY9Q+fgeW3uy8Qjc+/E/8d58teTHh+CX3Jw4hlS5hoxiVrP/AnOE9zhTd4xW0xiGwO823Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4f388ed9225b125f93a68c89aaf2c52ca9c8227b3bb0c16202fb75d3c84f6b9f","last_reissued_at":"2026-07-05T10:26:17.656773Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:26:17.656773Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Beyond RAG: Task-Aware KV Cache Compression for Comprehensive Knowledge Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.IR","cs.LG"],"primary_cat":"cs.CL","authors_text":"Fabio Petroni, Giulio Corallo, Orion Weller, Paolo Papotti","submitted_at":"2025-03-06T21:07:41Z","abstract_excerpt":"Incorporating external knowledge in large language models (LLMs) enhances their utility across diverse applications, but existing methods have trade-offs. Retrieval-Augmented Generation (RAG) fetches evidence via similarity search, but key information may fall outside top ranked results. Long-context models can process multiple documents but are computationally expensive and limited by context window size. Inspired by students condensing study material for open-book exams, we propose task-aware key-value (KV) cache compression, which compresses external knowledge in a zero- or few-shot setup. "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.04973","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/2503.04973/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":"2503.04973","created_at":"2026-07-05T10:26:17.656824+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.04973v1","created_at":"2026-07-05T10:26:17.656824+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.04973","created_at":"2026-07-05T10:26:17.656824+00:00"},{"alias_kind":"pith_short_12","alias_value":"J44I5WJCLMJF","created_at":"2026-07-05T10:26:17.656824+00:00"},{"alias_kind":"pith_short_16","alias_value":"J44I5WJCLMJF7E5G","created_at":"2026-07-05T10:26:17.656824+00:00"},{"alias_kind":"pith_short_8","alias_value":"J44I5WJC","created_at":"2026-07-05T10:26:17.656824+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2508.15717","citing_title":"StreamMem: Query-Agnostic KV Cache Memory for Streaming Video Understanding","ref_index":3,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/J44I5WJCLMJF7E5GRSE2V4WFFS","json":"https://pith.science/pith/J44I5WJCLMJF7E5GRSE2V4WFFS.json","graph_json":"https://pith.science/api/pith-number/J44I5WJCLMJF7E5GRSE2V4WFFS/graph.json","events_json":"https://pith.science/api/pith-number/J44I5WJCLMJF7E5GRSE2V4WFFS/events.json","paper":"https://pith.science/paper/J44I5WJC"},"agent_actions":{"view_html":"https://pith.science/pith/J44I5WJCLMJF7E5GRSE2V4WFFS","download_json":"https://pith.science/pith/J44I5WJCLMJF7E5GRSE2V4WFFS.json","view_paper":"https://pith.science/paper/J44I5WJC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.04973&json=true","fetch_graph":"https://pith.science/api/pith-number/J44I5WJCLMJF7E5GRSE2V4WFFS/graph.json","fetch_events":"https://pith.science/api/pith-number/J44I5WJCLMJF7E5GRSE2V4WFFS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/J44I5WJCLMJF7E5GRSE2V4WFFS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/J44I5WJCLMJF7E5GRSE2V4WFFS/action/storage_attestation","attest_author":"https://pith.science/pith/J44I5WJCLMJF7E5GRSE2V4WFFS/action/author_attestation","sign_citation":"https://pith.science/pith/J44I5WJCLMJF7E5GRSE2V4WFFS/action/citation_signature","submit_replication":"https://pith.science/pith/J44I5WJCLMJF7E5GRSE2V4WFFS/action/replication_record"}},"created_at":"2026-07-05T10:26:17.656824+00:00","updated_at":"2026-07-05T10:26:17.656824+00:00"}