{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:XGIFGUCBHUAJCURRXNGTUG7MD3","short_pith_number":"pith:XGIFGUCB","schema_version":"1.0","canonical_sha256":"b9905350413d00915231bb4d3a1bec1ee2922e56684e48ded90d079796161159","source":{"kind":"arxiv","id":"2508.01483","version":1},"attestation_state":"computed","paper":{"title":"Training Dynamics of the Cooldown Stage in Warmup-Stable-Decay Learning Rate Scheduler","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aleksandr Dremov, Alexander H\\\"agele, Atli Kosson, Martin Jaggi","submitted_at":"2025-08-02T20:36:52Z","abstract_excerpt":"Learning rate scheduling is essential in transformer training, where the final annealing plays a crucial role in getting the best performance. However, the mechanisms behind this cooldown phase, with its characteristic drop in loss, remain poorly understood. To address this, we provide a comprehensive analysis focusing solely on the cooldown phase in the Warmup-Stable-Decay (WSD) learning rate scheduler. Our analysis reveals that different cooldown shapes reveal a fundamental bias-variance trade-off in the resulting models, with shapes that balance exploration and exploitation consistently out"},"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":"2508.01483","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-08-02T20:36:52Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"c886c815d3c7c86a2078c720b5de28017de2dd113244cc394199ce5ff604e990","abstract_canon_sha256":"18fb02672897095a7cd3eef5a209e4b25f423dcbdb5bad38a9974b8a0b2d5a88"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:49:58.866313Z","signature_b64":"6vuYxKrRONRvOjbYf97rCHscUouEtX3usQvxbXWT6d58vnnvENSkv3WO8190o/zc9T7BXVXQkebtsG3f6ehNDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b9905350413d00915231bb4d3a1bec1ee2922e56684e48ded90d079796161159","last_reissued_at":"2026-07-05T11:49:58.865839Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:49:58.865839Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Training Dynamics of the Cooldown Stage in Warmup-Stable-Decay Learning Rate Scheduler","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aleksandr Dremov, Alexander H\\\"agele, Atli Kosson, Martin Jaggi","submitted_at":"2025-08-02T20:36:52Z","abstract_excerpt":"Learning rate scheduling is essential in transformer training, where the final annealing plays a crucial role in getting the best performance. However, the mechanisms behind this cooldown phase, with its characteristic drop in loss, remain poorly understood. To address this, we provide a comprehensive analysis focusing solely on the cooldown phase in the Warmup-Stable-Decay (WSD) learning rate scheduler. Our analysis reveals that different cooldown shapes reveal a fundamental bias-variance trade-off in the resulting models, with shapes that balance exploration and exploitation consistently out"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.01483","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/2508.01483/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":"2508.01483","created_at":"2026-07-05T11:49:58.865901+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.01483v1","created_at":"2026-07-05T11:49:58.865901+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.01483","created_at":"2026-07-05T11:49:58.865901+00:00"},{"alias_kind":"pith_short_12","alias_value":"XGIFGUCBHUAJ","created_at":"2026-07-05T11:49:58.865901+00:00"},{"alias_kind":"pith_short_16","alias_value":"XGIFGUCBHUAJCURR","created_at":"2026-07-05T11:49:58.865901+00:00"},{"alias_kind":"pith_short_8","alias_value":"XGIFGUCB","created_at":"2026-07-05T11:49:58.865901+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.09165","citing_title":"Sparse Layers are Critical to Scaling Looped Language Models","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09165","citing_title":"Sparse Layers are Critical to Scaling Looped Language Models","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XGIFGUCBHUAJCURRXNGTUG7MD3","json":"https://pith.science/pith/XGIFGUCBHUAJCURRXNGTUG7MD3.json","graph_json":"https://pith.science/api/pith-number/XGIFGUCBHUAJCURRXNGTUG7MD3/graph.json","events_json":"https://pith.science/api/pith-number/XGIFGUCBHUAJCURRXNGTUG7MD3/events.json","paper":"https://pith.science/paper/XGIFGUCB"},"agent_actions":{"view_html":"https://pith.science/pith/XGIFGUCBHUAJCURRXNGTUG7MD3","download_json":"https://pith.science/pith/XGIFGUCBHUAJCURRXNGTUG7MD3.json","view_paper":"https://pith.science/paper/XGIFGUCB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.01483&json=true","fetch_graph":"https://pith.science/api/pith-number/XGIFGUCBHUAJCURRXNGTUG7MD3/graph.json","fetch_events":"https://pith.science/api/pith-number/XGIFGUCBHUAJCURRXNGTUG7MD3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XGIFGUCBHUAJCURRXNGTUG7MD3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XGIFGUCBHUAJCURRXNGTUG7MD3/action/storage_attestation","attest_author":"https://pith.science/pith/XGIFGUCBHUAJCURRXNGTUG7MD3/action/author_attestation","sign_citation":"https://pith.science/pith/XGIFGUCBHUAJCURRXNGTUG7MD3/action/citation_signature","submit_replication":"https://pith.science/pith/XGIFGUCBHUAJCURRXNGTUG7MD3/action/replication_record"}},"created_at":"2026-07-05T11:49:58.865901+00:00","updated_at":"2026-07-05T11:49:58.865901+00:00"}