{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:OLZW4GZTUAAWC3GDHCF55WBHY5","short_pith_number":"pith:OLZW4GZT","schema_version":"1.0","canonical_sha256":"72f36e1b33a001616cc3388bded827c773ffa621f9ebd9a085efa69e80875111","source":{"kind":"arxiv","id":"2409.07787","version":1},"attestation_state":"computed","paper":{"title":"Stable Language Model Pre-training by Reducing Embedding Variability","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"James Thorne, Jiwoo Hong, Na Min An, Se-Young Yun, WooJin Chung","submitted_at":"2024-09-12T06:37:46Z","abstract_excerpt":"Stable pre-training is essential for achieving better-performing language models. However, tracking pre-training stability by calculating gradient variance at every step is impractical due to the significant computational costs. We explore Token Embedding Variability (TEV) as a simple and efficient proxy for assessing pre-training stability in language models with pre-layer normalization, given that shallower layers are more prone to gradient explosion (section 2.2). Moreover, we propose Multi-head Low-Rank Attention (MLRA) as an architecture to alleviate such instability by limiting the expon"},"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":"2409.07787","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-09-12T06:37:46Z","cross_cats_sorted":[],"title_canon_sha256":"68e6ffe3e224d79e5828b189bb7d7d590106590e1c3397581653210a5839786b","abstract_canon_sha256":"1dca235b03ed985e8365e9792eb8325e4717c63b1ba6e35ee6dd0e93aa02e426"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:06:15.047556Z","signature_b64":"JnVHQJQOIbZHIQ+OGPz/xRkINZqVoFDmh3KATyrQpxgw7ZRr2WMNhrmNxO3HhyJz+lL6NmVXYeo12nmwogZ0Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"72f36e1b33a001616cc3388bded827c773ffa621f9ebd9a085efa69e80875111","last_reissued_at":"2026-07-05T09:06:15.047201Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:06:15.047201Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Stable Language Model Pre-training by Reducing Embedding Variability","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"James Thorne, Jiwoo Hong, Na Min An, Se-Young Yun, WooJin Chung","submitted_at":"2024-09-12T06:37:46Z","abstract_excerpt":"Stable pre-training is essential for achieving better-performing language models. However, tracking pre-training stability by calculating gradient variance at every step is impractical due to the significant computational costs. We explore Token Embedding Variability (TEV) as a simple and efficient proxy for assessing pre-training stability in language models with pre-layer normalization, given that shallower layers are more prone to gradient explosion (section 2.2). Moreover, we propose Multi-head Low-Rank Attention (MLRA) as an architecture to alleviate such instability by limiting the expon"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.07787","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/2409.07787/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":"2409.07787","created_at":"2026-07-05T09:06:15.047256+00:00"},{"alias_kind":"arxiv_version","alias_value":"2409.07787v1","created_at":"2026-07-05T09:06:15.047256+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.07787","created_at":"2026-07-05T09:06:15.047256+00:00"},{"alias_kind":"pith_short_12","alias_value":"OLZW4GZTUAAW","created_at":"2026-07-05T09:06:15.047256+00:00"},{"alias_kind":"pith_short_16","alias_value":"OLZW4GZTUAAWC3GD","created_at":"2026-07-05T09:06:15.047256+00:00"},{"alias_kind":"pith_short_8","alias_value":"OLZW4GZT","created_at":"2026-07-05T09:06:15.047256+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/OLZW4GZTUAAWC3GDHCF55WBHY5","json":"https://pith.science/pith/OLZW4GZTUAAWC3GDHCF55WBHY5.json","graph_json":"https://pith.science/api/pith-number/OLZW4GZTUAAWC3GDHCF55WBHY5/graph.json","events_json":"https://pith.science/api/pith-number/OLZW4GZTUAAWC3GDHCF55WBHY5/events.json","paper":"https://pith.science/paper/OLZW4GZT"},"agent_actions":{"view_html":"https://pith.science/pith/OLZW4GZTUAAWC3GDHCF55WBHY5","download_json":"https://pith.science/pith/OLZW4GZTUAAWC3GDHCF55WBHY5.json","view_paper":"https://pith.science/paper/OLZW4GZT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2409.07787&json=true","fetch_graph":"https://pith.science/api/pith-number/OLZW4GZTUAAWC3GDHCF55WBHY5/graph.json","fetch_events":"https://pith.science/api/pith-number/OLZW4GZTUAAWC3GDHCF55WBHY5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OLZW4GZTUAAWC3GDHCF55WBHY5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OLZW4GZTUAAWC3GDHCF55WBHY5/action/storage_attestation","attest_author":"https://pith.science/pith/OLZW4GZTUAAWC3GDHCF55WBHY5/action/author_attestation","sign_citation":"https://pith.science/pith/OLZW4GZTUAAWC3GDHCF55WBHY5/action/citation_signature","submit_replication":"https://pith.science/pith/OLZW4GZTUAAWC3GDHCF55WBHY5/action/replication_record"}},"created_at":"2026-07-05T09:06:15.047256+00:00","updated_at":"2026-07-05T09:06:15.047256+00:00"}