{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:SAPHDR3RTV7JA7AOOA3IFA7A2V","short_pith_number":"pith:SAPHDR3R","schema_version":"1.0","canonical_sha256":"901e71c7719d7e907c0e70368283e0d54ef163f6c0192cebe8280fe74d8d9ab2","source":{"kind":"arxiv","id":"2002.04745","version":2},"attestation_state":"computed","paper":{"title":"On Layer Normalization in the Transformer Architecture","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Chen Xing, Di He, Huishuai Zhang, Kai Zheng, Liwei Wang, Ruibin Xiong, Shuxin Zheng, Tie-Yan Liu, Yanyan Lan, Yunchang Yang","submitted_at":"2020-02-12T00:33:03Z","abstract_excerpt":"The Transformer is widely used in natural language processing tasks. To train a Transformer however, one usually needs a carefully designed learning rate warm-up stage, which is shown to be crucial to the final performance but will slow down the optimization and bring more hyper-parameter tunings. In this paper, we first study theoretically why the learning rate warm-up stage is essential and show that the location of layer normalization matters. Specifically, we prove with mean field theory that at initialization, for the original-designed Post-LN Transformer, which places the layer normaliza"},"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":"2002.04745","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2020-02-12T00:33:03Z","cross_cats_sorted":["cs.CL","stat.ML"],"title_canon_sha256":"96b527f13f4517567d030e3b983a85804127c2c03c3dde2625963c5de93683f9","abstract_canon_sha256":"5cd7a634e3eaad5549936455628f9e05555e72d47c1c18f517c33dec2ecf310f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T01:13:57.975666Z","signature_b64":"5xX1oAt4/sILf78oOBOHF1lwdPlySU3H76XNRoVhv+EuJxSz3c4Q6ApK0MogW9vb08zU7sRwXHOMUSUSZIMrCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"901e71c7719d7e907c0e70368283e0d54ef163f6c0192cebe8280fe74d8d9ab2","last_reissued_at":"2026-07-05T01:13:57.975174Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T01:13:57.975174Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On Layer Normalization in the Transformer Architecture","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Chen Xing, Di He, Huishuai Zhang, Kai Zheng, Liwei Wang, Ruibin Xiong, Shuxin Zheng, Tie-Yan Liu, Yanyan Lan, Yunchang Yang","submitted_at":"2020-02-12T00:33:03Z","abstract_excerpt":"The Transformer is widely used in natural language processing tasks. To train a Transformer however, one usually needs a carefully designed learning rate warm-up stage, which is shown to be crucial to the final performance but will slow down the optimization and bring more hyper-parameter tunings. In this paper, we first study theoretically why the learning rate warm-up stage is essential and show that the location of layer normalization matters. Specifically, we prove with mean field theory that at initialization, for the original-designed Post-LN Transformer, which places the layer normaliza"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2002.04745","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/2002.04745/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":"2002.04745","created_at":"2026-07-05T01:13:57.975233+00:00"},{"alias_kind":"arxiv_version","alias_value":"2002.04745v2","created_at":"2026-07-05T01:13:57.975233+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2002.04745","created_at":"2026-07-05T01:13:57.975233+00:00"},{"alias_kind":"pith_short_12","alias_value":"SAPHDR3RTV7J","created_at":"2026-07-05T01:13:57.975233+00:00"},{"alias_kind":"pith_short_16","alias_value":"SAPHDR3RTV7JA7AO","created_at":"2026-07-05T01:13:57.975233+00:00"},{"alias_kind":"pith_short_8","alias_value":"SAPHDR3R","created_at":"2026-07-05T01:13:57.975233+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":19,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2607.06629","citing_title":"STST-JEPA: Shallow-Target Spatio-Temporal Joint Embedding Prediction Architecture For EEG Self-Supervised Learning","ref_index":22,"is_internal_anchor":true},{"citing_arxiv_id":"2607.08175","citing_title":"Transformer-based machine learning using low-level calorimeter signals for collimated photon identification at collider experiments","ref_index":59,"is_internal_anchor":true},{"citing_arxiv_id":"2606.25971","citing_title":"Improving Neural Network Training by Decoupling the Magnitude and Direction of Weight Vectors","ref_index":155,"is_internal_anchor":false},{"citing_arxiv_id":"2606.25519","citing_title":"Quantization Inflates Reasoning: Token Inflation as a Hidden Cost of Low-Bit Reasoning Models","ref_index":150,"is_internal_anchor":false},{"citing_arxiv_id":"2606.26894","citing_title":"Modeling Local, Global, and Cross-Modal Context in Multimodal 3D MRI","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2606.13168","citing_title":"When Does Routing Become Interpretable? Causal Probes on Block Attention Residuals","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2606.25519","citing_title":"Quantization Inflates Reasoning: Token Inflation as a Hidden Cost of Low-Bit Reasoning Models","ref_index":150,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31859","citing_title":"Review Residuals: Update-Conditioned Residual Gating for Transformers","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23259","citing_title":"Multi-Gate Residuals","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2602.10408","citing_title":"Gated Normalization Removal and Scale Anchoring in Pre-Norm Transformers","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2603.15031","citing_title":"Attention Residuals","ref_index":61,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02075","citing_title":"Graph Transformers and Stabilized Reinforcement Learning for Large-Scale Dynamic Routing Modulation and Spectrum Allocation in Elastic Optical Networks","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18933","citing_title":"A Geometric Analysis of Sign-Magnitude Asymmetry in a ReLU + RMSNorm Block under Ternary Quantization","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2601.14053","citing_title":"LLMOrbit: A Circular Taxonomy of Large Language Models -From Scaling Walls to Agentic AI Systems","ref_index":167,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07097","citing_title":"Every Feedforward Neural Network Definable in an o-Minimal Structure Has Finite Sample Complexity","ref_index":92,"is_internal_anchor":false},{"citing_arxiv_id":"2004.05150","citing_title":"Longformer: The Long-Document Transformer","ref_index":127,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15259","citing_title":"Stability and Generalization in Looped Transformers","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21184","citing_title":"Predicting the thermodynamics in the chromosphere from the translation of SDO data into the IRIS$^{2}$ inversion results using a visual transformer model","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02075","citing_title":"Graph Transformers and Stabilized Reinforcement Learning for Large-Scale Dynamic Routing Modulation and Spectrum Allocation in Elastic Optical Networks","ref_index":10,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SAPHDR3RTV7JA7AOOA3IFA7A2V","json":"https://pith.science/pith/SAPHDR3RTV7JA7AOOA3IFA7A2V.json","graph_json":"https://pith.science/api/pith-number/SAPHDR3RTV7JA7AOOA3IFA7A2V/graph.json","events_json":"https://pith.science/api/pith-number/SAPHDR3RTV7JA7AOOA3IFA7A2V/events.json","paper":"https://pith.science/paper/SAPHDR3R"},"agent_actions":{"view_html":"https://pith.science/pith/SAPHDR3RTV7JA7AOOA3IFA7A2V","download_json":"https://pith.science/pith/SAPHDR3RTV7JA7AOOA3IFA7A2V.json","view_paper":"https://pith.science/paper/SAPHDR3R","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2002.04745&json=true","fetch_graph":"https://pith.science/api/pith-number/SAPHDR3RTV7JA7AOOA3IFA7A2V/graph.json","fetch_events":"https://pith.science/api/pith-number/SAPHDR3RTV7JA7AOOA3IFA7A2V/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SAPHDR3RTV7JA7AOOA3IFA7A2V/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SAPHDR3RTV7JA7AOOA3IFA7A2V/action/storage_attestation","attest_author":"https://pith.science/pith/SAPHDR3RTV7JA7AOOA3IFA7A2V/action/author_attestation","sign_citation":"https://pith.science/pith/SAPHDR3RTV7JA7AOOA3IFA7A2V/action/citation_signature","submit_replication":"https://pith.science/pith/SAPHDR3RTV7JA7AOOA3IFA7A2V/action/replication_record"}},"created_at":"2026-07-05T01:13:57.975233+00:00","updated_at":"2026-07-05T01:13:57.975233+00:00"}