{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:N6EHFQ4SAMOO2YAEHDLKBYNFCO","short_pith_number":"pith:N6EHFQ4S","schema_version":"1.0","canonical_sha256":"6f8872c392031ced600438d6a0e1a513b29115d92ccf4f6265e3e839fcd66158","source":{"kind":"arxiv","id":"2606.05516","version":1},"attestation_state":"computed","paper":{"title":"Dominant-Layer ZO: A Single Layer Dominates Zeroth-Order Fine-Tuning of LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Abeer Matar Almalky, Adnan Siraj Rakin, Deliang Fan, Li Yang, Sen Lin, Shuteng Niu, Wanhao Yu, Yihang Zuo, Zheng Wang, Ziyan Wang","submitted_at":"2026-06-03T23:42:09Z","abstract_excerpt":"Zeroth-order (ZO) optimization enables memory-efficient fine-tuning of large language models (LLMs) using only forward passes, but it remains unclear how useful adaptation is distributed across layers. In this work, we reveal a surprising phenomenon: ZO fine-tuning is sharply dominated by a single decoding layer. Across multiple LLM families and downstream tasks, fine-tuning this dominant layer alone consistently matches or even exceeds full-model ZO fine-tuning. We further show that the dominant layer is task-agnostic but model-specific, and can be identified before training through a simple "},"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":"2606.05516","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-03T23:42:09Z","cross_cats_sorted":[],"title_canon_sha256":"2a1d9091cd71edc5c696910b20f29a5fad700b2edbc9e4b48efb43bc913a3ad2","abstract_canon_sha256":"11cbbd0a7f762409cb14dd72e1f7e55575ef69fc655441291014f8c46f1278cf"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-05T01:14:53.456150Z","signature_b64":"iqmit6sMYZm8fw+qVHR+9bJHS320Qkz9emY7m1c1aNU532tM5MYjSYatleEZicL47yJ8mDscOjXOQU5s+CNnCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6f8872c392031ced600438d6a0e1a513b29115d92ccf4f6265e3e839fcd66158","last_reissued_at":"2026-06-05T01:14:53.455631Z","signature_status":"signed_v1","first_computed_at":"2026-06-05T01:14:53.455631Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Dominant-Layer ZO: A Single Layer Dominates Zeroth-Order Fine-Tuning of LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Abeer Matar Almalky, Adnan Siraj Rakin, Deliang Fan, Li Yang, Sen Lin, Shuteng Niu, Wanhao Yu, Yihang Zuo, Zheng Wang, Ziyan Wang","submitted_at":"2026-06-03T23:42:09Z","abstract_excerpt":"Zeroth-order (ZO) optimization enables memory-efficient fine-tuning of large language models (LLMs) using only forward passes, but it remains unclear how useful adaptation is distributed across layers. In this work, we reveal a surprising phenomenon: ZO fine-tuning is sharply dominated by a single decoding layer. Across multiple LLM families and downstream tasks, fine-tuning this dominant layer alone consistently matches or even exceeds full-model ZO fine-tuning. We further show that the dominant layer is task-agnostic but model-specific, and can be identified before training through a simple "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.05516","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/2606.05516/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":"2606.05516","created_at":"2026-06-05T01:14:53.455709+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.05516v1","created_at":"2026-06-05T01:14:53.455709+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.05516","created_at":"2026-06-05T01:14:53.455709+00:00"},{"alias_kind":"pith_short_12","alias_value":"N6EHFQ4SAMOO","created_at":"2026-06-05T01:14:53.455709+00:00"},{"alias_kind":"pith_short_16","alias_value":"N6EHFQ4SAMOO2YAE","created_at":"2026-06-05T01:14:53.455709+00:00"},{"alias_kind":"pith_short_8","alias_value":"N6EHFQ4S","created_at":"2026-06-05T01:14:53.455709+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/N6EHFQ4SAMOO2YAEHDLKBYNFCO","json":"https://pith.science/pith/N6EHFQ4SAMOO2YAEHDLKBYNFCO.json","graph_json":"https://pith.science/api/pith-number/N6EHFQ4SAMOO2YAEHDLKBYNFCO/graph.json","events_json":"https://pith.science/api/pith-number/N6EHFQ4SAMOO2YAEHDLKBYNFCO/events.json","paper":"https://pith.science/paper/N6EHFQ4S"},"agent_actions":{"view_html":"https://pith.science/pith/N6EHFQ4SAMOO2YAEHDLKBYNFCO","download_json":"https://pith.science/pith/N6EHFQ4SAMOO2YAEHDLKBYNFCO.json","view_paper":"https://pith.science/paper/N6EHFQ4S","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.05516&json=true","fetch_graph":"https://pith.science/api/pith-number/N6EHFQ4SAMOO2YAEHDLKBYNFCO/graph.json","fetch_events":"https://pith.science/api/pith-number/N6EHFQ4SAMOO2YAEHDLKBYNFCO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/N6EHFQ4SAMOO2YAEHDLKBYNFCO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/N6EHFQ4SAMOO2YAEHDLKBYNFCO/action/storage_attestation","attest_author":"https://pith.science/pith/N6EHFQ4SAMOO2YAEHDLKBYNFCO/action/author_attestation","sign_citation":"https://pith.science/pith/N6EHFQ4SAMOO2YAEHDLKBYNFCO/action/citation_signature","submit_replication":"https://pith.science/pith/N6EHFQ4SAMOO2YAEHDLKBYNFCO/action/replication_record"}},"created_at":"2026-06-05T01:14:53.455709+00:00","updated_at":"2026-06-05T01:14:53.455709+00:00"}