{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:25FYT5LUFOAWVASWFYGAFV56S7","short_pith_number":"pith:25FYT5LU","schema_version":"1.0","canonical_sha256":"d74b89f5742b816a82562e0c02d7be97c67681df2cd2eb65ae0cfdd5427a179a","source":{"kind":"arxiv","id":"2305.14342","version":4},"attestation_state":"computed","paper":{"title":"Sophia: A Scalable Stochastic Second-order Optimizer for Language Model Pre-training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","math.OC"],"primary_cat":"cs.LG","authors_text":"David Hall, Hong Liu, Percy Liang, Tengyu Ma, Zhiyuan Li","submitted_at":"2023-05-23T17:59:21Z","abstract_excerpt":"Given the massive cost of language model pre-training, a non-trivial improvement of the optimization algorithm would lead to a material reduction on the time and cost of training. Adam and its variants have been state-of-the-art for years, and more sophisticated second-order (Hessian-based) optimizers often incur too much per-step overhead. In this paper, we propose Sophia, Second-order Clipped Stochastic Optimization, a simple scalable second-order optimizer that uses a light-weight estimate of the diagonal Hessian as the pre-conditioner. The update is the moving average of the gradients divi"},"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":"2305.14342","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-05-23T17:59:21Z","cross_cats_sorted":["cs.CL","math.OC"],"title_canon_sha256":"16bfd4c87e981757e2910c52ba9416938cbca2518d9c25375bc4bd78f52c7b32","abstract_canon_sha256":"e38861e0c7ee14d28cdfb5ad1d0ab65e50b316d796b3741676369641014c8265"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:52:09.876325Z","signature_b64":"S2pEko+VrSWiyD7AAPwxpuniHj2UeHwgyX2jF7X7+dPXUk9cNaqMxw5nXiFPpprRRMMGUWhzCqs7kUH7osSkAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d74b89f5742b816a82562e0c02d7be97c67681df2cd2eb65ae0cfdd5427a179a","last_reissued_at":"2026-07-05T07:52:09.875823Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:52:09.875823Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Sophia: A Scalable Stochastic Second-order Optimizer for Language Model Pre-training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","math.OC"],"primary_cat":"cs.LG","authors_text":"David Hall, Hong Liu, Percy Liang, Tengyu Ma, Zhiyuan Li","submitted_at":"2023-05-23T17:59:21Z","abstract_excerpt":"Given the massive cost of language model pre-training, a non-trivial improvement of the optimization algorithm would lead to a material reduction on the time and cost of training. Adam and its variants have been state-of-the-art for years, and more sophisticated second-order (Hessian-based) optimizers often incur too much per-step overhead. In this paper, we propose Sophia, Second-order Clipped Stochastic Optimization, a simple scalable second-order optimizer that uses a light-weight estimate of the diagonal Hessian as the pre-conditioner. The update is the moving average of the gradients divi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2305.14342","kind":"arxiv","version":4},"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/2305.14342/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":"2305.14342","created_at":"2026-07-05T07:52:09.875883+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.14342v4","created_at":"2026-07-05T07:52:09.875883+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.14342","created_at":"2026-07-05T07:52:09.875883+00:00"},{"alias_kind":"pith_short_12","alias_value":"25FYT5LUFOAW","created_at":"2026-07-05T07:52:09.875883+00:00"},{"alias_kind":"pith_short_16","alias_value":"25FYT5LUFOAWVASW","created_at":"2026-07-05T07:52:09.875883+00:00"},{"alias_kind":"pith_short_8","alias_value":"25FYT5LU","created_at":"2026-07-05T07:52:09.875883+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":27,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07206","citing_title":"Causal Optimizer Interaction Calculus: Hidden Geometric Relaxation and Identifiable Interventions","ref_index":78,"is_internal_anchor":true},{"citing_arxiv_id":"2606.25086","citing_title":"Training for the Model You Return: Improving Optimization for Iterate-Averaged Language Models","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2606.14187","citing_title":"Zeta: Dual Whitening for Matrix Optimization via Coordinate-Adaptive Preconditioning","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06412","citing_title":"Nonreversible Gauge Fields in Fokker--Planck Dynamics: Supersymmetric Hamiltonians and Learned Finite Forces","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04662","citing_title":"Why Muon Outperforms Adam: A Curvature Perspective","ref_index":103,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03904","citing_title":"MAdam: Metric-Aware Multi-Objective Adam","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2606.25086","citing_title":"Training for the Model You Return: Improving Optimization for Iterate-Averaged Language Models","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30455","citing_title":"Curvature-Weighted Gradient Diversity: A Noise Measure for Geometry-Adaptive SGD Schedules","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2603.10067","citing_title":"HTMuon: Improving Muon via Heavy-Tailed Spectral Correction","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2505.13196","citing_title":"A Physics-Inspired Optimizer: Velocity Regularized Adam","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21724","citing_title":"TBP-mHC: full expressivity for manifold-constrained hyper connections through transportation polytopes","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22644","citing_title":"Why SGD is not Brownian Motion: A New Perspective on Stochastic Dynamics","ref_index":153,"is_internal_anchor":false},{"citing_arxiv_id":"2511.19078","citing_title":"GraphMind: Theorem Selection and Conclusion Generation Framework with Dynamic GNN for LLM Reasoning","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2602.00545","citing_title":"Depth, Not Data: An Analysis of Hessian Spectral Bifurcation","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15622","citing_title":"Position: Zeroth-Order Optimization in Deep Learning Is Underexplored, Not Underpowered","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2509.15816","citing_title":"On the Convergence of Muon and Beyond","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2509.18993","citing_title":"CR-Net: Scaling Parameter-Efficient Training with Cross-Layer Low-Rank Structure","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2510.00761","citing_title":"Downgrade to Upgrade: Optimizer Simplification Enhances Robustness in LLM Unlearning","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2601.06540","citing_title":"Self-Organizing Dual-Buffer Adaptive Clustering Experience Replay (SODACER) for Safe Reinforcement Learning in Optimal Control","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06081","citing_title":"Fast Gauss-Newton for Multiclass Cross-Entropy","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02317","citing_title":"Anon: Extrapolating Adaptivity Beyond SGD and Adam","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00650","citing_title":"AdaMeZO: Adam-style Zeroth-Order Optimizer for LLM Fine-tuning Without Maintaining the Moments","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06868","citing_title":"When Descent Is Too Stable: Event-Triggered Hamiltonian Learning to Optimize","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07815","citing_title":"OrScale: Orthogonalised Optimization with Layer-Wise Trust-Ratio Scaling","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06836","citing_title":"STQuant: Spatio-Temporal Adaptive Framework for Optimizer Quantization in Large Multimodal Model Training","ref_index":12,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/25FYT5LUFOAWVASWFYGAFV56S7","json":"https://pith.science/pith/25FYT5LUFOAWVASWFYGAFV56S7.json","graph_json":"https://pith.science/api/pith-number/25FYT5LUFOAWVASWFYGAFV56S7/graph.json","events_json":"https://pith.science/api/pith-number/25FYT5LUFOAWVASWFYGAFV56S7/events.json","paper":"https://pith.science/paper/25FYT5LU"},"agent_actions":{"view_html":"https://pith.science/pith/25FYT5LUFOAWVASWFYGAFV56S7","download_json":"https://pith.science/pith/25FYT5LUFOAWVASWFYGAFV56S7.json","view_paper":"https://pith.science/paper/25FYT5LU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.14342&json=true","fetch_graph":"https://pith.science/api/pith-number/25FYT5LUFOAWVASWFYGAFV56S7/graph.json","fetch_events":"https://pith.science/api/pith-number/25FYT5LUFOAWVASWFYGAFV56S7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/25FYT5LUFOAWVASWFYGAFV56S7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/25FYT5LUFOAWVASWFYGAFV56S7/action/storage_attestation","attest_author":"https://pith.science/pith/25FYT5LUFOAWVASWFYGAFV56S7/action/author_attestation","sign_citation":"https://pith.science/pith/25FYT5LUFOAWVASWFYGAFV56S7/action/citation_signature","submit_replication":"https://pith.science/pith/25FYT5LUFOAWVASWFYGAFV56S7/action/replication_record"}},"created_at":"2026-07-05T07:52:09.875883+00:00","updated_at":"2026-07-05T07:52:09.875883+00:00"}