{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:ORN6RXF5POLONYGJKMAPBBCSMN","short_pith_number":"pith:ORN6RXF5","schema_version":"1.0","canonical_sha256":"745be8dcbd7b96e6e0c95300f08452635ba710fa0914afd46a27eecee7453e65","source":{"kind":"arxiv","id":"2507.08472","version":2},"attestation_state":"computed","paper":{"title":"Pre-Training LLMs on a budget: A comparison of three optimizers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chris Hinze, Christian Kroos, Fabian K\\\"uch, Joel Schlotthauer, Luzian Hahn, Viktor Hangya","submitted_at":"2025-07-11T10:29:04Z","abstract_excerpt":"Optimizers play a decisive role in reducing pre-training times for LLMs and achieving better-performing models. In this study, we compare three major variants: the de-facto standard AdamW, the simpler Lion, developed through an evolutionary search, and the second-order optimizer Sophia. For better generalization, we train with two different base architectures and use a single- and a multiple-epoch approach while keeping the number of tokens constant. Using the Maximal Update Parametrization and smaller proxy models, we tune relevant hyperparameters separately for each combination of base archi"},"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":"2507.08472","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-07-11T10:29:04Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"087c1947ae967f740ab3d2afab97453c6c675043b1246b084b03ede8ffd5b103","abstract_canon_sha256":"55f6926034d4ef950e86ae4a03f36f80637e958b92111b19fb6e58a483ef3092"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:40:56.579264Z","signature_b64":"Pko+WB/CHOs+IvpDL4vQOJOx4sftMyMfHdN7wVH8PwnfnwQEOBqeAKpFQGuaZWNa6/zIwBoI946rZDUEO8UEBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"745be8dcbd7b96e6e0c95300f08452635ba710fa0914afd46a27eecee7453e65","last_reissued_at":"2026-07-05T11:40:56.577889Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:40:56.577889Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Pre-Training LLMs on a budget: A comparison of three optimizers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chris Hinze, Christian Kroos, Fabian K\\\"uch, Joel Schlotthauer, Luzian Hahn, Viktor Hangya","submitted_at":"2025-07-11T10:29:04Z","abstract_excerpt":"Optimizers play a decisive role in reducing pre-training times for LLMs and achieving better-performing models. In this study, we compare three major variants: the de-facto standard AdamW, the simpler Lion, developed through an evolutionary search, and the second-order optimizer Sophia. For better generalization, we train with two different base architectures and use a single- and a multiple-epoch approach while keeping the number of tokens constant. Using the Maximal Update Parametrization and smaller proxy models, we tune relevant hyperparameters separately for each combination of base archi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.08472","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/2507.08472/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":"2507.08472","created_at":"2026-07-05T11:40:56.578338+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.08472v2","created_at":"2026-07-05T11:40:56.578338+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.08472","created_at":"2026-07-05T11:40:56.578338+00:00"},{"alias_kind":"pith_short_12","alias_value":"ORN6RXF5POLO","created_at":"2026-07-05T11:40:56.578338+00:00"},{"alias_kind":"pith_short_16","alias_value":"ORN6RXF5POLONYGJ","created_at":"2026-07-05T11:40:56.578338+00:00"},{"alias_kind":"pith_short_8","alias_value":"ORN6RXF5","created_at":"2026-07-05T11:40:56.578338+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.15416","citing_title":"StoSignSGD: Unbiased Structural Stochasticity Fixes SignSGD for Training Large Language Models","ref_index":35,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ORN6RXF5POLONYGJKMAPBBCSMN","json":"https://pith.science/pith/ORN6RXF5POLONYGJKMAPBBCSMN.json","graph_json":"https://pith.science/api/pith-number/ORN6RXF5POLONYGJKMAPBBCSMN/graph.json","events_json":"https://pith.science/api/pith-number/ORN6RXF5POLONYGJKMAPBBCSMN/events.json","paper":"https://pith.science/paper/ORN6RXF5"},"agent_actions":{"view_html":"https://pith.science/pith/ORN6RXF5POLONYGJKMAPBBCSMN","download_json":"https://pith.science/pith/ORN6RXF5POLONYGJKMAPBBCSMN.json","view_paper":"https://pith.science/paper/ORN6RXF5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.08472&json=true","fetch_graph":"https://pith.science/api/pith-number/ORN6RXF5POLONYGJKMAPBBCSMN/graph.json","fetch_events":"https://pith.science/api/pith-number/ORN6RXF5POLONYGJKMAPBBCSMN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ORN6RXF5POLONYGJKMAPBBCSMN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ORN6RXF5POLONYGJKMAPBBCSMN/action/storage_attestation","attest_author":"https://pith.science/pith/ORN6RXF5POLONYGJKMAPBBCSMN/action/author_attestation","sign_citation":"https://pith.science/pith/ORN6RXF5POLONYGJKMAPBBCSMN/action/citation_signature","submit_replication":"https://pith.science/pith/ORN6RXF5POLONYGJKMAPBBCSMN/action/replication_record"}},"created_at":"2026-07-05T11:40:56.578338+00:00","updated_at":"2026-07-05T11:40:56.578338+00:00"}