{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:GJWP2GLQSAHMGGIZKOVMTOWHUU","short_pith_number":"pith:GJWP2GLQ","schema_version":"1.0","canonical_sha256":"326cfd1970900ec3191953aac9bac7a52ede436f3063eab22b67ec8fdad6b59b","source":{"kind":"arxiv","id":"2502.07365","version":3},"attestation_state":"computed","paper":{"title":"LongReD: Mitigating Short-Text Degradation of Long-Context Large Language Models via Restoration Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bingning Wang, Jinhao Jiang, Junyi Li, Mingyu Xu, Wayne Xin Zhao, Weipeng Chen, Zican Dong","submitted_at":"2025-02-11T08:37:16Z","abstract_excerpt":"Large language models (LLMs) have gained extended context windows through scaling positional encodings and lightweight continual pre-training. However, this often leads to degraded performance on short-text tasks, while the reasons for this degradation remain insufficiently explored. In this work, we identify two primary factors contributing to this issue: distribution drift in hidden states and attention scores, and catastrophic forgetting during continual pre-training. To address these challenges, we propose Long Context Pre-training with Restoration Distillation (LongReD), a novel approach "},"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":"2502.07365","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-02-11T08:37:16Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"9901e50ff91e9afb635e669075fabc6988641191b65569655feb2f2957d4f746","abstract_canon_sha256":"828229cdad89b294ab9d32ad25ccc57645cead651c209d7af6afa58cdcdca094"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:10:52.278853Z","signature_b64":"X9qJDuNM4yfBvXL7Ocqg0d6aa+B6NSmyPoApcWwonBXvn418LZZKbTU7I+1OdnaYF58VfZE2dt4DZuRC1xMkCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"326cfd1970900ec3191953aac9bac7a52ede436f3063eab22b67ec8fdad6b59b","last_reissued_at":"2026-07-05T11:10:52.278369Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:10:52.278369Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LongReD: Mitigating Short-Text Degradation of Long-Context Large Language Models via Restoration Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Bingning Wang, Jinhao Jiang, Junyi Li, Mingyu Xu, Wayne Xin Zhao, Weipeng Chen, Zican Dong","submitted_at":"2025-02-11T08:37:16Z","abstract_excerpt":"Large language models (LLMs) have gained extended context windows through scaling positional encodings and lightweight continual pre-training. However, this often leads to degraded performance on short-text tasks, while the reasons for this degradation remain insufficiently explored. In this work, we identify two primary factors contributing to this issue: distribution drift in hidden states and attention scores, and catastrophic forgetting during continual pre-training. To address these challenges, we propose Long Context Pre-training with Restoration Distillation (LongReD), a novel approach "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.07365","kind":"arxiv","version":3},"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/2502.07365/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":"2502.07365","created_at":"2026-07-05T11:10:52.278425+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.07365v3","created_at":"2026-07-05T11:10:52.278425+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.07365","created_at":"2026-07-05T11:10:52.278425+00:00"},{"alias_kind":"pith_short_12","alias_value":"GJWP2GLQSAHM","created_at":"2026-07-05T11:10:52.278425+00:00"},{"alias_kind":"pith_short_16","alias_value":"GJWP2GLQSAHMGGIZ","created_at":"2026-07-05T11:10:52.278425+00:00"},{"alias_kind":"pith_short_8","alias_value":"GJWP2GLQ","created_at":"2026-07-05T11:10:52.278425+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.23414","citing_title":"When Planning Fails Despite Correct Execution: On Epistemic Calibration for LLM-Based Multi-Agent Systems","ref_index":71,"is_internal_anchor":false},{"citing_arxiv_id":"2505.06708","citing_title":"Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free","ref_index":10,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GJWP2GLQSAHMGGIZKOVMTOWHUU","json":"https://pith.science/pith/GJWP2GLQSAHMGGIZKOVMTOWHUU.json","graph_json":"https://pith.science/api/pith-number/GJWP2GLQSAHMGGIZKOVMTOWHUU/graph.json","events_json":"https://pith.science/api/pith-number/GJWP2GLQSAHMGGIZKOVMTOWHUU/events.json","paper":"https://pith.science/paper/GJWP2GLQ"},"agent_actions":{"view_html":"https://pith.science/pith/GJWP2GLQSAHMGGIZKOVMTOWHUU","download_json":"https://pith.science/pith/GJWP2GLQSAHMGGIZKOVMTOWHUU.json","view_paper":"https://pith.science/paper/GJWP2GLQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.07365&json=true","fetch_graph":"https://pith.science/api/pith-number/GJWP2GLQSAHMGGIZKOVMTOWHUU/graph.json","fetch_events":"https://pith.science/api/pith-number/GJWP2GLQSAHMGGIZKOVMTOWHUU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GJWP2GLQSAHMGGIZKOVMTOWHUU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GJWP2GLQSAHMGGIZKOVMTOWHUU/action/storage_attestation","attest_author":"https://pith.science/pith/GJWP2GLQSAHMGGIZKOVMTOWHUU/action/author_attestation","sign_citation":"https://pith.science/pith/GJWP2GLQSAHMGGIZKOVMTOWHUU/action/citation_signature","submit_replication":"https://pith.science/pith/GJWP2GLQSAHMGGIZKOVMTOWHUU/action/replication_record"}},"created_at":"2026-07-05T11:10:52.278425+00:00","updated_at":"2026-07-05T11:10:52.278425+00:00"}