{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:EQJV3FMDTUZ2YKDLQUPSZES56A","short_pith_number":"pith:EQJV3FMD","schema_version":"1.0","canonical_sha256":"24135d95839d33ac286b851f2c925df012cb7f52c4e72dabf41a58b666cfbfca","source":{"kind":"arxiv","id":"2410.13077","version":2},"attestation_state":"computed","paper":{"title":"Tuning Language Models by Mixture-of-Depths Ensemble","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Haoyan Luo, Lucia Specia","submitted_at":"2024-10-16T22:51:45Z","abstract_excerpt":"Transformer-based Large Language Models (LLMs) traditionally rely on final-layer loss for finetuning and final-layer representations for predictions, potentially overlooking the predictive power embedded in late layers. Interpretability tools such as the logit lens show that late-layer representations already carry largely formed, task-relevant predictions; here we ask whether that observation can be turned into an actionable training signal. We find that focusing tuning effort on these layers can yield losses comparable to those of the final layer, with complementary test-time behaviour. Buil"},"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":"2410.13077","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-16T22:51:45Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"41936874a5bc7483887f4f45ea23ecb75e16907b13dae23756aa0e74f60e4e51","abstract_canon_sha256":"2c2a95f8372945ef012aa2cad9818229b6fcb48b9f7164c8cc5355889ea8ba1d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-26T00:15:21.713372Z","signature_b64":"8Oa2434t4wfBkgjy79F0n9ziFv7WWhSmeu5BKenf03MB8vu6fs2L562GD2HehBvuG2CKTyE9oyrFleOV8OzPAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"24135d95839d33ac286b851f2c925df012cb7f52c4e72dabf41a58b666cfbfca","last_reissued_at":"2026-06-26T00:15:21.712722Z","signature_status":"signed_v1","first_computed_at":"2026-06-26T00:15:21.712722Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Tuning Language Models by Mixture-of-Depths Ensemble","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Haoyan Luo, Lucia Specia","submitted_at":"2024-10-16T22:51:45Z","abstract_excerpt":"Transformer-based Large Language Models (LLMs) traditionally rely on final-layer loss for finetuning and final-layer representations for predictions, potentially overlooking the predictive power embedded in late layers. Interpretability tools such as the logit lens show that late-layer representations already carry largely formed, task-relevant predictions; here we ask whether that observation can be turned into an actionable training signal. We find that focusing tuning effort on these layers can yield losses comparable to those of the final layer, with complementary test-time behaviour. Buil"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.13077","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/2410.13077/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":"2410.13077","created_at":"2026-06-26T00:15:21.712809+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.13077v2","created_at":"2026-06-26T00:15:21.712809+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.13077","created_at":"2026-06-26T00:15:21.712809+00:00"},{"alias_kind":"pith_short_12","alias_value":"EQJV3FMDTUZ2","created_at":"2026-06-26T00:15:21.712809+00:00"},{"alias_kind":"pith_short_16","alias_value":"EQJV3FMDTUZ2YKDL","created_at":"2026-06-26T00:15:21.712809+00:00"},{"alias_kind":"pith_short_8","alias_value":"EQJV3FMD","created_at":"2026-06-26T00:15:21.712809+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EQJV3FMDTUZ2YKDLQUPSZES56A","json":"https://pith.science/pith/EQJV3FMDTUZ2YKDLQUPSZES56A.json","graph_json":"https://pith.science/api/pith-number/EQJV3FMDTUZ2YKDLQUPSZES56A/graph.json","events_json":"https://pith.science/api/pith-number/EQJV3FMDTUZ2YKDLQUPSZES56A/events.json","paper":"https://pith.science/paper/EQJV3FMD"},"agent_actions":{"view_html":"https://pith.science/pith/EQJV3FMDTUZ2YKDLQUPSZES56A","download_json":"https://pith.science/pith/EQJV3FMDTUZ2YKDLQUPSZES56A.json","view_paper":"https://pith.science/paper/EQJV3FMD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.13077&json=true","fetch_graph":"https://pith.science/api/pith-number/EQJV3FMDTUZ2YKDLQUPSZES56A/graph.json","fetch_events":"https://pith.science/api/pith-number/EQJV3FMDTUZ2YKDLQUPSZES56A/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EQJV3FMDTUZ2YKDLQUPSZES56A/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EQJV3FMDTUZ2YKDLQUPSZES56A/action/storage_attestation","attest_author":"https://pith.science/pith/EQJV3FMDTUZ2YKDLQUPSZES56A/action/author_attestation","sign_citation":"https://pith.science/pith/EQJV3FMDTUZ2YKDLQUPSZES56A/action/citation_signature","submit_replication":"https://pith.science/pith/EQJV3FMDTUZ2YKDLQUPSZES56A/action/replication_record"}},"created_at":"2026-06-26T00:15:21.712809+00:00","updated_at":"2026-06-26T00:15:21.712809+00:00"}