{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:52X73VVIXQVDMSOOXEXA5NOLLN","short_pith_number":"pith:52X73VVI","schema_version":"1.0","canonical_sha256":"eeaffdd6a8bc2a3649ceb92e0eb5cb5b42f43720e0feb931f283726494cb39f0","source":{"kind":"arxiv","id":"2502.02659","version":2},"attestation_state":"computed","paper":{"title":"A Training-Free Length Extrapolation Approach for LLMs: Greedy Attention Logit Interpolation (GALI)","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Soyeon Caren Han, Tianyi Zhang, Yan Li, Zechuan Li","submitted_at":"2025-02-04T19:01:24Z","abstract_excerpt":"Transformer-based Large Language Models (LLMs) struggle with inputs exceeding their training context window due to positional out-of-distribution (O.O.D.) issues that disrupt attention. Existing solutions, including fine-tuning and training-free methods, face challenges like inefficiency, redundant interpolation, logit outliers, or loss of local positional information. We propose Greedy Attention Logit Interpolation (GALI), a training-free method that improves length extrapolation by greedily reusing pretrained positional intervals and interpolating attention logit to eliminate outliers. GALI "},"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":"2502.02659","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-02-04T19:01:24Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"554f66920673a97e7908bd43cfe149fe4975bd2319d44c1882d95c7dfd429e5d","abstract_canon_sha256":"acedab77ba333b28cc9d99f2d48c869873e73c70e505afefe90381234db7badc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:12:25.392594Z","signature_b64":"ZAtwHcdLTJ/ceZzbc+flzQS8gJbFReUYpjADUJ9PHJfYVh3D/eSjsqypP65mWW13TgpRzJTsIxovqcnX9OCbAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"eeaffdd6a8bc2a3649ceb92e0eb5cb5b42f43720e0feb931f283726494cb39f0","last_reissued_at":"2026-07-05T11:12:25.391748Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:12:25.391748Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Training-Free Length Extrapolation Approach for LLMs: Greedy Attention Logit Interpolation (GALI)","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Soyeon Caren Han, Tianyi Zhang, Yan Li, Zechuan Li","submitted_at":"2025-02-04T19:01:24Z","abstract_excerpt":"Transformer-based Large Language Models (LLMs) struggle with inputs exceeding their training context window due to positional out-of-distribution (O.O.D.) issues that disrupt attention. Existing solutions, including fine-tuning and training-free methods, face challenges like inefficiency, redundant interpolation, logit outliers, or loss of local positional information. We propose Greedy Attention Logit Interpolation (GALI), a training-free method that improves length extrapolation by greedily reusing pretrained positional intervals and interpolating attention logit to eliminate outliers. GALI "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.02659","kind":"arxiv","version":2},"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/2502.02659/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":"2502.02659","created_at":"2026-07-05T11:12:25.392092+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.02659v2","created_at":"2026-07-05T11:12:25.392092+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.02659","created_at":"2026-07-05T11:12:25.392092+00:00"},{"alias_kind":"pith_short_12","alias_value":"52X73VVIXQVD","created_at":"2026-07-05T11:12:25.392092+00:00"},{"alias_kind":"pith_short_16","alias_value":"52X73VVIXQVDMSOO","created_at":"2026-07-05T11:12:25.392092+00:00"},{"alias_kind":"pith_short_8","alias_value":"52X73VVI","created_at":"2026-07-05T11:12:25.392092+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07740","citing_title":"Jet-Long: Efficient Long-Context Extension with Dynamic Bifocal RoPE","ref_index":41,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/52X73VVIXQVDMSOOXEXA5NOLLN","json":"https://pith.science/pith/52X73VVIXQVDMSOOXEXA5NOLLN.json","graph_json":"https://pith.science/api/pith-number/52X73VVIXQVDMSOOXEXA5NOLLN/graph.json","events_json":"https://pith.science/api/pith-number/52X73VVIXQVDMSOOXEXA5NOLLN/events.json","paper":"https://pith.science/paper/52X73VVI"},"agent_actions":{"view_html":"https://pith.science/pith/52X73VVIXQVDMSOOXEXA5NOLLN","download_json":"https://pith.science/pith/52X73VVIXQVDMSOOXEXA5NOLLN.json","view_paper":"https://pith.science/paper/52X73VVI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.02659&json=true","fetch_graph":"https://pith.science/api/pith-number/52X73VVIXQVDMSOOXEXA5NOLLN/graph.json","fetch_events":"https://pith.science/api/pith-number/52X73VVIXQVDMSOOXEXA5NOLLN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/52X73VVIXQVDMSOOXEXA5NOLLN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/52X73VVIXQVDMSOOXEXA5NOLLN/action/storage_attestation","attest_author":"https://pith.science/pith/52X73VVIXQVDMSOOXEXA5NOLLN/action/author_attestation","sign_citation":"https://pith.science/pith/52X73VVIXQVDMSOOXEXA5NOLLN/action/citation_signature","submit_replication":"https://pith.science/pith/52X73VVIXQVDMSOOXEXA5NOLLN/action/replication_record"}},"created_at":"2026-07-05T11:12:25.392092+00:00","updated_at":"2026-07-05T11:12:25.392092+00:00"}