{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:CND7DI2DCDBK2FRKURIR6F5KE5","short_pith_number":"pith:CND7DI2D","schema_version":"1.0","canonical_sha256":"1347f1a34310c2ad162aa4511f17aa27542cb1d60ae3a032a6e9972dd855988c","source":{"kind":"arxiv","id":"2605.26484","version":1},"attestation_state":"computed","paper":{"title":"Extra-Merge: Tracing the Rank-1 Subspace of Model Merging in Language Model Pre-Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Bohan Wang, Chenxi Jia, Hongtao Zhang, Wei Chen, Wenjie Zhou, Xueqi Cheng","submitted_at":"2026-05-26T02:48:34Z","abstract_excerpt":"Model merging has emerged as a lightweight paradigm for enhancing Large Language Models (LLMs), yet its underlying mechanisms remain poorly understood. In this work, we analyze late-stage pre-training trajectories and uncover a \\textbf{Rank-1 Subspace} phenomenon: while raw optimization steps oscillate violently, consecutive \\emph{merged} checkpoints collapse onto a stable, approximately one-dimensional linear manifold. We theoretically ground this observation in a \\emph{river-valley} landscape analysis: averaging acts as a geometric low-pass filter that dampens high-curvature noise to reveal "},"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":"2605.26484","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-26T02:48:34Z","cross_cats_sorted":[],"title_canon_sha256":"1edf592ebe906467095d72d2b7e18b924eb32d1cda3e777836f32e2f2e7c9feb","abstract_canon_sha256":"1877203c180f67f281e97407f3551776c6fc94c7f2b7b021fd851074d74d739a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-27T01:05:20.640290Z","signature_b64":"7Rjqe4V3GkIfVCPrg8m0a5tkf89lzLIWt2q7Uqc/vtMYGpDM6snRn24EoRE8f7mb1pJJ2cUIvMA4wTXjF7sPCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1347f1a34310c2ad162aa4511f17aa27542cb1d60ae3a032a6e9972dd855988c","last_reissued_at":"2026-05-27T01:05:20.639643Z","signature_status":"signed_v1","first_computed_at":"2026-05-27T01:05:20.639643Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Extra-Merge: Tracing the Rank-1 Subspace of Model Merging in Language Model Pre-Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Bohan Wang, Chenxi Jia, Hongtao Zhang, Wei Chen, Wenjie Zhou, Xueqi Cheng","submitted_at":"2026-05-26T02:48:34Z","abstract_excerpt":"Model merging has emerged as a lightweight paradigm for enhancing Large Language Models (LLMs), yet its underlying mechanisms remain poorly understood. In this work, we analyze late-stage pre-training trajectories and uncover a \\textbf{Rank-1 Subspace} phenomenon: while raw optimization steps oscillate violently, consecutive \\emph{merged} checkpoints collapse onto a stable, approximately one-dimensional linear manifold. We theoretically ground this observation in a \\emph{river-valley} landscape analysis: averaging acts as a geometric low-pass filter that dampens high-curvature noise to reveal "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.26484","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/2605.26484/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":"2605.26484","created_at":"2026-05-27T01:05:20.639755+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.26484v1","created_at":"2026-05-27T01:05:20.639755+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.26484","created_at":"2026-05-27T01:05:20.639755+00:00"},{"alias_kind":"pith_short_12","alias_value":"CND7DI2DCDBK","created_at":"2026-05-27T01:05:20.639755+00:00"},{"alias_kind":"pith_short_16","alias_value":"CND7DI2DCDBK2FRK","created_at":"2026-05-27T01:05:20.639755+00:00"},{"alias_kind":"pith_short_8","alias_value":"CND7DI2D","created_at":"2026-05-27T01:05:20.639755+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/CND7DI2DCDBK2FRKURIR6F5KE5","json":"https://pith.science/pith/CND7DI2DCDBK2FRKURIR6F5KE5.json","graph_json":"https://pith.science/api/pith-number/CND7DI2DCDBK2FRKURIR6F5KE5/graph.json","events_json":"https://pith.science/api/pith-number/CND7DI2DCDBK2FRKURIR6F5KE5/events.json","paper":"https://pith.science/paper/CND7DI2D"},"agent_actions":{"view_html":"https://pith.science/pith/CND7DI2DCDBK2FRKURIR6F5KE5","download_json":"https://pith.science/pith/CND7DI2DCDBK2FRKURIR6F5KE5.json","view_paper":"https://pith.science/paper/CND7DI2D","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.26484&json=true","fetch_graph":"https://pith.science/api/pith-number/CND7DI2DCDBK2FRKURIR6F5KE5/graph.json","fetch_events":"https://pith.science/api/pith-number/CND7DI2DCDBK2FRKURIR6F5KE5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CND7DI2DCDBK2FRKURIR6F5KE5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CND7DI2DCDBK2FRKURIR6F5KE5/action/storage_attestation","attest_author":"https://pith.science/pith/CND7DI2DCDBK2FRKURIR6F5KE5/action/author_attestation","sign_citation":"https://pith.science/pith/CND7DI2DCDBK2FRKURIR6F5KE5/action/citation_signature","submit_replication":"https://pith.science/pith/CND7DI2DCDBK2FRKURIR6F5KE5/action/replication_record"}},"created_at":"2026-05-27T01:05:20.639755+00:00","updated_at":"2026-05-27T01:05:20.639755+00:00"}