{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:3Y5E44A47FHW35X6RLHSKBKC36","short_pith_number":"pith:3Y5E44A4","schema_version":"1.0","canonical_sha256":"de3a4e701cf94f6df6fe8acf250542dfad27aca310e34defa8cb2615c4bc11dc","source":{"kind":"arxiv","id":"2503.08727","version":4},"attestation_state":"computed","paper":{"title":"Training Plug-n-Play Knowledge Modules with Deep Context Distillation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alan Ansell, Alessandro Sordoni, Edoardo Ponti, Ivan Vuli\\'c, Lucas Caccia","submitted_at":"2025-03-11T01:07:57Z","abstract_excerpt":"Dynamically integrating new or rapidly evolving information after (Large) Language Model pre-training remains challenging, particularly in low-data scenarios or when dealing with private and specialized documents. In-context learning and retrieval-augmented generation (RAG) face limitations, including their high inference costs and their inability to capture global document information. In this paper, we propose a way of modularizing knowledge by training document-level Knowledge Modules (KMs). KMs are lightweight components implemented as parameter-efficient LoRA modules, which are trained to"},"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":"2503.08727","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-03-11T01:07:57Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"5b574cc996a757c7f9fd63a884a88b280711829c6e83d4e0bbd912a03863f15c","abstract_canon_sha256":"6e8cb3e3a3e096e247413321346735d4f07835baf968f6c9b1d03bda2ea63ba1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:50:33.869588Z","signature_b64":"b1emykneo2qpoF3GccnuNKim8busl9oUCWBzN7NDrwhhD+SsXeiWJTWDQK8FDKXKPuYgFYHqG4wKo01qkhIaCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"de3a4e701cf94f6df6fe8acf250542dfad27aca310e34defa8cb2615c4bc11dc","last_reissued_at":"2026-07-05T11:50:33.869080Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:50:33.869080Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Training Plug-n-Play Knowledge Modules with Deep Context Distillation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Alan Ansell, Alessandro Sordoni, Edoardo Ponti, Ivan Vuli\\'c, Lucas Caccia","submitted_at":"2025-03-11T01:07:57Z","abstract_excerpt":"Dynamically integrating new or rapidly evolving information after (Large) Language Model pre-training remains challenging, particularly in low-data scenarios or when dealing with private and specialized documents. In-context learning and retrieval-augmented generation (RAG) face limitations, including their high inference costs and their inability to capture global document information. In this paper, we propose a way of modularizing knowledge by training document-level Knowledge Modules (KMs). KMs are lightweight components implemented as parameter-efficient LoRA modules, which are trained to"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.08727","kind":"arxiv","version":4},"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/2503.08727/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":"2503.08727","created_at":"2026-07-05T11:50:33.869138+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.08727v4","created_at":"2026-07-05T11:50:33.869138+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.08727","created_at":"2026-07-05T11:50:33.869138+00:00"},{"alias_kind":"pith_short_12","alias_value":"3Y5E44A47FHW","created_at":"2026-07-05T11:50:33.869138+00:00"},{"alias_kind":"pith_short_16","alias_value":"3Y5E44A47FHW35X6","created_at":"2026-07-05T11:50:33.869138+00:00"},{"alias_kind":"pith_short_8","alias_value":"3Y5E44A4","created_at":"2026-07-05T11:50:33.869138+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.04351","citing_title":"Frames2LoRA: Parametric Video Internalization for Vision-Language Models","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03979","citing_title":"Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories","ref_index":109,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26099","citing_title":"Do Language Models Need Sleep? Offline Recurrence for Improved Online Inference","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09778","citing_title":"Nectar: Neural Estimation of Cached-Token Attention via Regression","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06377","citing_title":"The Master Key Hypothesis: Unlocking Cross-Model Capability Transfer via Linear Subspace Alignment","ref_index":12,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3Y5E44A47FHW35X6RLHSKBKC36","json":"https://pith.science/pith/3Y5E44A47FHW35X6RLHSKBKC36.json","graph_json":"https://pith.science/api/pith-number/3Y5E44A47FHW35X6RLHSKBKC36/graph.json","events_json":"https://pith.science/api/pith-number/3Y5E44A47FHW35X6RLHSKBKC36/events.json","paper":"https://pith.science/paper/3Y5E44A4"},"agent_actions":{"view_html":"https://pith.science/pith/3Y5E44A47FHW35X6RLHSKBKC36","download_json":"https://pith.science/pith/3Y5E44A47FHW35X6RLHSKBKC36.json","view_paper":"https://pith.science/paper/3Y5E44A4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.08727&json=true","fetch_graph":"https://pith.science/api/pith-number/3Y5E44A47FHW35X6RLHSKBKC36/graph.json","fetch_events":"https://pith.science/api/pith-number/3Y5E44A47FHW35X6RLHSKBKC36/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3Y5E44A47FHW35X6RLHSKBKC36/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3Y5E44A47FHW35X6RLHSKBKC36/action/storage_attestation","attest_author":"https://pith.science/pith/3Y5E44A47FHW35X6RLHSKBKC36/action/author_attestation","sign_citation":"https://pith.science/pith/3Y5E44A47FHW35X6RLHSKBKC36/action/citation_signature","submit_replication":"https://pith.science/pith/3Y5E44A47FHW35X6RLHSKBKC36/action/replication_record"}},"created_at":"2026-07-05T11:50:33.869138+00:00","updated_at":"2026-07-05T11:50:33.869138+00:00"}