{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:ZHE5LMQIXJRVCGUN44IBNH7JYS","short_pith_number":"pith:ZHE5LMQI","schema_version":"1.0","canonical_sha256":"c9c9d5b208ba63511a8de710169fe9c4a4f5c8e43069d52c0dbac029028625bc","source":{"kind":"arxiv","id":"2507.03526","version":1},"attestation_state":"computed","paper":{"title":"Decoupled Relative Learning Rate Schedules","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Jakub Krajewski, Jan Ludziejewski, Jan Ma{\\l}a\\'snicki, Kamil Adamczewski, Kamil Ciebiera, Maciej Pi\\'oro, Maciej Stefaniak, Marek Cygan, Micha{\\l} Krutul, Piotr Sankowski, Sebastian Jaszczur","submitted_at":"2025-07-04T12:23:45Z","abstract_excerpt":"In this work, we introduce a novel approach for optimizing LLM training by adjusting learning rates across weights of different components in Transformer models. Traditional methods often apply a uniform learning rate across all network layers, potentially overlooking the unique dynamics of each part. Remarkably, our introduced relative learning rates, RLRS, method accelerates the training process by up to $23\\%$, particularly in complex models such as Mixture of Experts (MoE). Hyperparameters of RLRS can be efficiently tuned on smaller models and then effectively reused on models up to $27\\ti"},"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":"2507.03526","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-07-04T12:23:45Z","cross_cats_sorted":[],"title_canon_sha256":"7dc1113d1d7d9597abebce7c80c490290dffb51e8213ae08c2cefacc8face931","abstract_canon_sha256":"a46894127806282325edbf1e79d6540e8d737459528ccc3177635a50e0bf86eb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:32:01.001834Z","signature_b64":"O1EqSwC59+WOh3vg1U6u1Hswd4J2j1MnnmTgnSBj5rYC+DGYUUbOtmZ/q23IxhOaTZm9aH2WzJFFTP0lck65Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c9c9d5b208ba63511a8de710169fe9c4a4f5c8e43069d52c0dbac029028625bc","last_reissued_at":"2026-07-05T11:32:01.001335Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:32:01.001335Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Decoupled Relative Learning Rate Schedules","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Jakub Krajewski, Jan Ludziejewski, Jan Ma{\\l}a\\'snicki, Kamil Adamczewski, Kamil Ciebiera, Maciej Pi\\'oro, Maciej Stefaniak, Marek Cygan, Micha{\\l} Krutul, Piotr Sankowski, Sebastian Jaszczur","submitted_at":"2025-07-04T12:23:45Z","abstract_excerpt":"In this work, we introduce a novel approach for optimizing LLM training by adjusting learning rates across weights of different components in Transformer models. Traditional methods often apply a uniform learning rate across all network layers, potentially overlooking the unique dynamics of each part. Remarkably, our introduced relative learning rates, RLRS, method accelerates the training process by up to $23\\%$, particularly in complex models such as Mixture of Experts (MoE). Hyperparameters of RLRS can be efficiently tuned on smaller models and then effectively reused on models up to $27\\ti"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.03526","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/2507.03526/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":"2507.03526","created_at":"2026-07-05T11:32:01.001398+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.03526v1","created_at":"2026-07-05T11:32:01.001398+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.03526","created_at":"2026-07-05T11:32:01.001398+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZHE5LMQIXJRV","created_at":"2026-07-05T11:32:01.001398+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZHE5LMQIXJRVCGUN","created_at":"2026-07-05T11:32:01.001398+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZHE5LMQI","created_at":"2026-07-05T11:32:01.001398+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.27133","citing_title":"Deep-layer limit and stability analysis of the basic forward-backward-splitting induced network (II): learning problems","ref_index":45,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZHE5LMQIXJRVCGUN44IBNH7JYS","json":"https://pith.science/pith/ZHE5LMQIXJRVCGUN44IBNH7JYS.json","graph_json":"https://pith.science/api/pith-number/ZHE5LMQIXJRVCGUN44IBNH7JYS/graph.json","events_json":"https://pith.science/api/pith-number/ZHE5LMQIXJRVCGUN44IBNH7JYS/events.json","paper":"https://pith.science/paper/ZHE5LMQI"},"agent_actions":{"view_html":"https://pith.science/pith/ZHE5LMQIXJRVCGUN44IBNH7JYS","download_json":"https://pith.science/pith/ZHE5LMQIXJRVCGUN44IBNH7JYS.json","view_paper":"https://pith.science/paper/ZHE5LMQI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.03526&json=true","fetch_graph":"https://pith.science/api/pith-number/ZHE5LMQIXJRVCGUN44IBNH7JYS/graph.json","fetch_events":"https://pith.science/api/pith-number/ZHE5LMQIXJRVCGUN44IBNH7JYS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZHE5LMQIXJRVCGUN44IBNH7JYS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZHE5LMQIXJRVCGUN44IBNH7JYS/action/storage_attestation","attest_author":"https://pith.science/pith/ZHE5LMQIXJRVCGUN44IBNH7JYS/action/author_attestation","sign_citation":"https://pith.science/pith/ZHE5LMQIXJRVCGUN44IBNH7JYS/action/citation_signature","submit_replication":"https://pith.science/pith/ZHE5LMQIXJRVCGUN44IBNH7JYS/action/replication_record"}},"created_at":"2026-07-05T11:32:01.001398+00:00","updated_at":"2026-07-05T11:32:01.001398+00:00"}