{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:HA5XVOEFS7OGRSA6GC2WONPP5L","short_pith_number":"pith:HA5XVOEF","schema_version":"1.0","canonical_sha256":"383b7ab88597dc68c81e30b56735efeadf2f96ef5a0d5ec3c8965779973a4e73","source":{"kind":"arxiv","id":"2505.02222","version":4},"attestation_state":"computed","paper":{"title":"Practical Efficiency of Muon for Pretraining","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Adarsh Chaluvaraju, Andrew Hojel, Andrew Ma, Anil Thomas, Anthony M. Polloreno, Ashish Tanwer, Ashish Vaswani, Darsh J Shah, Essential AI: Ishaan Shah, Karl Stratos, Khoi Nguyen, Kurt Smith, Michael Callahan, Michael Pust, Mohit Parmar, Peter Rushton, Philip Monk, Platon Mazarakis, Ritvik Kapila, Saurabh Srivastava, Somanshu Singla, Tim Romanski, Yash Vanjani","submitted_at":"2025-05-04T19:14:43Z","abstract_excerpt":"We demonstrate that Muon, the simplest instantiation of a second-order optimizer, explicitly expands the Pareto frontier over AdamW on the compute-time tradeoff. We find that Muon is more effective than AdamW in retaining data efficiency at large batch sizes, far beyond the so-called critical batch size, while remaining computationally efficient, thus enabling more economical training. We study the combination of Muon and the maximal update parameterization (muP) for efficient hyperparameter transfer and present a simple telescoping algorithm that accounts for all sources of error in muP while"},"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":"2505.02222","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-04T19:14:43Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"225355c010c0ea7a454a71167f030d5ce1477f0ae8bd5d012da4e130e5c59225","abstract_canon_sha256":"09d3d8a263be5cf636cf709b1318f37eb7dd5aab024777b8bef9f998551496c8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:05:43.778509Z","signature_b64":"ZZGIbYjL7bUZpK4aXl+bg1akztlkAns5USWoxbg9nuXu/TvAlimsp5I0rcX3a3JwzKDAIOsGYCcKJAJOGTL7Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"383b7ab88597dc68c81e30b56735efeadf2f96ef5a0d5ec3c8965779973a4e73","last_reissued_at":"2026-07-05T11:05:43.778023Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:05:43.778023Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Practical Efficiency of Muon for Pretraining","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Adarsh Chaluvaraju, Andrew Hojel, Andrew Ma, Anil Thomas, Anthony M. Polloreno, Ashish Tanwer, Ashish Vaswani, Darsh J Shah, Essential AI: Ishaan Shah, Karl Stratos, Khoi Nguyen, Kurt Smith, Michael Callahan, Michael Pust, Mohit Parmar, Peter Rushton, Philip Monk, Platon Mazarakis, Ritvik Kapila, Saurabh Srivastava, Somanshu Singla, Tim Romanski, Yash Vanjani","submitted_at":"2025-05-04T19:14:43Z","abstract_excerpt":"We demonstrate that Muon, the simplest instantiation of a second-order optimizer, explicitly expands the Pareto frontier over AdamW on the compute-time tradeoff. We find that Muon is more effective than AdamW in retaining data efficiency at large batch sizes, far beyond the so-called critical batch size, while remaining computationally efficient, thus enabling more economical training. We study the combination of Muon and the maximal update parameterization (muP) for efficient hyperparameter transfer and present a simple telescoping algorithm that accounts for all sources of error in muP while"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.02222","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/2505.02222/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":"2505.02222","created_at":"2026-07-05T11:05:43.778083+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.02222v4","created_at":"2026-07-05T11:05:43.778083+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.02222","created_at":"2026-07-05T11:05:43.778083+00:00"},{"alias_kind":"pith_short_12","alias_value":"HA5XVOEFS7OG","created_at":"2026-07-05T11:05:43.778083+00:00"},{"alias_kind":"pith_short_16","alias_value":"HA5XVOEFS7OGRSA6","created_at":"2026-07-05T11:05:43.778083+00:00"},{"alias_kind":"pith_short_8","alias_value":"HA5XVOEF","created_at":"2026-07-05T11:05:43.778083+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":20,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23676","citing_title":"Open Problem: Is AdamW Effective Under Heavy-Tailed Noise?","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2606.21514","citing_title":"Towards Understanding the Power and Limits of the Muon Optimizer: A River-Valley Perspective","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01487","citing_title":"How to Allocate Your Tokens? Scaling Laws with Training Steps and Batch Size","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09658","citing_title":"Muon Learns More Robust and Transferable Features than Adam","ref_index":132,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08783","citing_title":"OptMuon: Closed-Loop Orthogonalized Momentum Methods for Stochastic Optimization with Zero-Noise Optimality","ref_index":57,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04662","citing_title":"Why Muon Outperforms Adam: A Curvature Perspective","ref_index":182,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18106","citing_title":"Symmetry-Compatible Principle for Optimizer Design: Embeddings, LM Heads, SwiGLU MLPs, and MoE Routers","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26977","citing_title":"Convergence of Spectral Descent for Non-smooth Optimization","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08783","citing_title":"OptMuon: Closed-Loop Orthogonalized Momentum Methods for Stochastic Optimization with Zero-Noise Optimality","ref_index":63,"is_internal_anchor":false},{"citing_arxiv_id":"2603.10067","citing_title":"HTMuon: Improving Muon via Heavy-Tailed Spectral Correction","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15622","citing_title":"Position: Zeroth-Order Optimization in Deep Learning Is Underexplored, Not Underpowered","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18106","citing_title":"Symmetry-Compatible Principle for Optimizer Design: Embeddings, LM Heads, SwiGLU MLPs, and MoE Routers","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2509.15816","citing_title":"On the Convergence of Muon and Beyond","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2509.11983","citing_title":"Low-rank Orthogonalization for Large-scale Matrix Optimization with Applications to Foundation Model Training","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2603.28254","citing_title":"MuonEq: Balancing Before Orthogonalization with Lightweight Equilibration","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13405","citing_title":"When is Warmstarting Effective for Scaling Language Models?","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11838","citing_title":"Gradient Clipping Beyond Vector Norms: A Spectral Approach for Matrix-Valued Parameters","ref_index":55,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10468","citing_title":"Can Muon Fine-tune Adam-Pretrained Models?","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09552","citing_title":"Phases of Muon: When Muon Eclipses SignSGD","ref_index":64,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06615","citing_title":"When and Why SignSGD Outperforms SGD: A Theoretical Study Based on $\\ell_1$-norm Lower Bounds","ref_index":31,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HA5XVOEFS7OGRSA6GC2WONPP5L","json":"https://pith.science/pith/HA5XVOEFS7OGRSA6GC2WONPP5L.json","graph_json":"https://pith.science/api/pith-number/HA5XVOEFS7OGRSA6GC2WONPP5L/graph.json","events_json":"https://pith.science/api/pith-number/HA5XVOEFS7OGRSA6GC2WONPP5L/events.json","paper":"https://pith.science/paper/HA5XVOEF"},"agent_actions":{"view_html":"https://pith.science/pith/HA5XVOEFS7OGRSA6GC2WONPP5L","download_json":"https://pith.science/pith/HA5XVOEFS7OGRSA6GC2WONPP5L.json","view_paper":"https://pith.science/paper/HA5XVOEF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.02222&json=true","fetch_graph":"https://pith.science/api/pith-number/HA5XVOEFS7OGRSA6GC2WONPP5L/graph.json","fetch_events":"https://pith.science/api/pith-number/HA5XVOEFS7OGRSA6GC2WONPP5L/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HA5XVOEFS7OGRSA6GC2WONPP5L/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HA5XVOEFS7OGRSA6GC2WONPP5L/action/storage_attestation","attest_author":"https://pith.science/pith/HA5XVOEFS7OGRSA6GC2WONPP5L/action/author_attestation","sign_citation":"https://pith.science/pith/HA5XVOEFS7OGRSA6GC2WONPP5L/action/citation_signature","submit_replication":"https://pith.science/pith/HA5XVOEFS7OGRSA6GC2WONPP5L/action/replication_record"}},"created_at":"2026-07-05T11:05:43.778083+00:00","updated_at":"2026-07-05T11:05:43.778083+00:00"}