{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:IFVTDPZTO6AEIO6HB5JRLHNNHG","short_pith_number":"pith:IFVTDPZT","schema_version":"1.0","canonical_sha256":"416b31bf337780443bc70f53159dad39a947f211452fae54654094fe32b977d0","source":{"kind":"arxiv","id":"2410.05192","version":3},"attestation_state":"computed","paper":{"title":"Understanding Warmup-Stable-Decay Learning Rates: A River Valley Loss Landscape Perspective","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"David Hall, Jason Wang, Kaiyue Wen, Percy Liang, Tengyu Ma, Zhiyuan Li","submitted_at":"2024-10-07T16:49:39Z","abstract_excerpt":"Training language models currently requires pre-determining a fixed compute budget because the typical cosine learning rate schedule depends on the total number of steps. In contrast, the Warmup-Stable-Decay (WSD) schedule uses a constant learning rate to produce a main branch of iterates that can in principle continue indefinitely without a pre-specified compute budget. Then, given any compute budget, one can branch out from the main branch at a proper time with a rapidly decaying learning rate to produce a strong model. Empirically, WSD generates a non-traditional loss curve: the loss remain"},"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":"2410.05192","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-10-07T16:49:39Z","cross_cats_sorted":["cs.CL","stat.ML"],"title_canon_sha256":"33605dfa59d979df32cadd1abaf66118f6f971502ab3017b3a02fad62f57297d","abstract_canon_sha256":"70fbca32c7a9f54608dcab11941a1076e50121d13b6a31adc370bed763bbcb2c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:43:35.049877Z","signature_b64":"fDYI1+JjMf693wg+9pC1M0uMiWU7UGEBhXg8r4aR2cjxHAF0dFSVBuXcyrWnqJ+vJ3XhL5nRbPZM4LsQb2YeAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"416b31bf337780443bc70f53159dad39a947f211452fae54654094fe32b977d0","last_reissued_at":"2026-07-05T09:43:35.049340Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:43:35.049340Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding Warmup-Stable-Decay Learning Rates: A River Valley Loss Landscape Perspective","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"David Hall, Jason Wang, Kaiyue Wen, Percy Liang, Tengyu Ma, Zhiyuan Li","submitted_at":"2024-10-07T16:49:39Z","abstract_excerpt":"Training language models currently requires pre-determining a fixed compute budget because the typical cosine learning rate schedule depends on the total number of steps. In contrast, the Warmup-Stable-Decay (WSD) schedule uses a constant learning rate to produce a main branch of iterates that can in principle continue indefinitely without a pre-specified compute budget. Then, given any compute budget, one can branch out from the main branch at a proper time with a rapidly decaying learning rate to produce a strong model. Empirically, WSD generates a non-traditional loss curve: the loss remain"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.05192","kind":"arxiv","version":3},"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/2410.05192/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":"2410.05192","created_at":"2026-07-05T09:43:35.049406+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.05192v3","created_at":"2026-07-05T09:43:35.049406+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.05192","created_at":"2026-07-05T09:43:35.049406+00:00"},{"alias_kind":"pith_short_12","alias_value":"IFVTDPZTO6AE","created_at":"2026-07-05T09:43:35.049406+00:00"},{"alias_kind":"pith_short_16","alias_value":"IFVTDPZTO6AEIO6H","created_at":"2026-07-05T09:43:35.049406+00:00"},{"alias_kind":"pith_short_8","alias_value":"IFVTDPZT","created_at":"2026-07-05T09:43:35.049406+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":20,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25086","citing_title":"Training for the Model You Return: Improving Optimization for Iterate-Averaged Language Models","ref_index":37,"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":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24956","citing_title":"NITP: Next Implicit Token Prediction for LLM Pre-training","ref_index":53,"is_internal_anchor":false},{"citing_arxiv_id":"2606.18524","citing_title":"On the Residual Scaling of Looped Transformers: Stability and Transferability","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05610","citing_title":"Predictable Scaling Laws of Optimal Hyperparameters for LLM Continued Pre-training","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04327","citing_title":"A Geometric Characterization of the Stationary Plateau for Two-Layer Neural Networks","ref_index":33,"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":37,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24956","citing_title":"NITP: Next Implicit Token Prediction for LLM Pre-training","ref_index":53,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00439","citing_title":"Physical Object Understanding with a Physically Controllable World Model","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2606.18524","citing_title":"On the Residual Scaling of Looped Transformers: Stability and Transferability","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06546","citing_title":"Efficient Pre-Training with Token Superposition","ref_index":59,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13652","citing_title":"Beyond Perplexity: A Geometric and Spectral Study of Low-Rank Pre-Training","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2602.04774","citing_title":"Theory of Optimal Learning Rate Schedules and Scaling Laws for a Random Feature Model","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2510.25741","citing_title":"Scaling Latent Reasoning via Looped Language Models","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13652","citing_title":"Beyond Perplexity: A Geometric and Spectral Study of Low-Rank Pre-Training","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06366","citing_title":"Layer Collapse in Diffusion Language Models","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06546","citing_title":"Efficient Pre-Training with Token Superposition","ref_index":57,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06366","citing_title":"Layer Collapse in Diffusion Language Models","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18827","citing_title":"OmniMouse: Scaling properties of multi-modal, multi-task Brain Models on 150B Neural Tokens","ref_index":111,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02105","citing_title":"Sharpness-Aware Pretraining Mitigates Catastrophic Forgetting","ref_index":77,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IFVTDPZTO6AEIO6HB5JRLHNNHG","json":"https://pith.science/pith/IFVTDPZTO6AEIO6HB5JRLHNNHG.json","graph_json":"https://pith.science/api/pith-number/IFVTDPZTO6AEIO6HB5JRLHNNHG/graph.json","events_json":"https://pith.science/api/pith-number/IFVTDPZTO6AEIO6HB5JRLHNNHG/events.json","paper":"https://pith.science/paper/IFVTDPZT"},"agent_actions":{"view_html":"https://pith.science/pith/IFVTDPZTO6AEIO6HB5JRLHNNHG","download_json":"https://pith.science/pith/IFVTDPZTO6AEIO6HB5JRLHNNHG.json","view_paper":"https://pith.science/paper/IFVTDPZT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.05192&json=true","fetch_graph":"https://pith.science/api/pith-number/IFVTDPZTO6AEIO6HB5JRLHNNHG/graph.json","fetch_events":"https://pith.science/api/pith-number/IFVTDPZTO6AEIO6HB5JRLHNNHG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IFVTDPZTO6AEIO6HB5JRLHNNHG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IFVTDPZTO6AEIO6HB5JRLHNNHG/action/storage_attestation","attest_author":"https://pith.science/pith/IFVTDPZTO6AEIO6HB5JRLHNNHG/action/author_attestation","sign_citation":"https://pith.science/pith/IFVTDPZTO6AEIO6HB5JRLHNNHG/action/citation_signature","submit_replication":"https://pith.science/pith/IFVTDPZTO6AEIO6HB5JRLHNNHG/action/replication_record"}},"created_at":"2026-07-05T09:43:35.049406+00:00","updated_at":"2026-07-05T09:43:35.049406+00:00"}