{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:3JKBJ6XXKLPWAYFEI2OINYRHTN","short_pith_number":"pith:3JKBJ6XX","schema_version":"1.0","canonical_sha256":"da5414faf752df6060a4469c86e2279b48773390bd662160c64ad6e6c848e052","source":{"kind":"arxiv","id":"2505.24275","version":3},"attestation_state":"computed","paper":{"title":"GradPower: Powering Gradients for Faster Language Model Pre-Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Jiaqi Zhang, Jinbo Wang, Lei Wu, Mingze Wang, Peng Pei, Weinan E, Wei Wang, Xunliang Cai","submitted_at":"2025-05-30T06:49:57Z","abstract_excerpt":"We propose GradPower, a lightweight gradient-transformation technique for accelerating language model pre-training. Given a gradient vector $g=(g_i)_i$, GradPower first applies the elementwise sign-power transformation: $\\varphi_p(g)=({\\rm sign}(g_i)|g_i|^p)_{i}$ for a fixed $p>0$, and then feeds the transformed gradient into a base optimizer. Notably, GradPower requires only a single-line code change and no modifications to the base optimizer's internal logic, including the hyperparameters. When applied to Adam (termed AdamPower), GradPower consistently achieves lower terminal loss across div"},"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":"2505.24275","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-30T06:49:57Z","cross_cats_sorted":["math.OC","stat.ML"],"title_canon_sha256":"16f073554ab6719c55fdcc2ee950d35597d354033ff95305a24b9ccc10506856","abstract_canon_sha256":"a2ec1e5d386e9cf493d836e8580907fce3e18c9884a5234937815862bc6da1bc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-21T01:04:14.252386Z","signature_b64":"OeGbVwBQ2exhfRzzvFxnapp2YWZxs/MzI42EPVHk604uQCkHXVtl4Y0QSzN22zTB4KjImuPbXWP9ZxYMnxFpDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"da5414faf752df6060a4469c86e2279b48773390bd662160c64ad6e6c848e052","last_reissued_at":"2026-05-21T01:04:14.251596Z","signature_status":"signed_v1","first_computed_at":"2026-05-21T01:04:14.251596Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"GradPower: Powering Gradients for Faster Language Model Pre-Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Jiaqi Zhang, Jinbo Wang, Lei Wu, Mingze Wang, Peng Pei, Weinan E, Wei Wang, Xunliang Cai","submitted_at":"2025-05-30T06:49:57Z","abstract_excerpt":"We propose GradPower, a lightweight gradient-transformation technique for accelerating language model pre-training. Given a gradient vector $g=(g_i)_i$, GradPower first applies the elementwise sign-power transformation: $\\varphi_p(g)=({\\rm sign}(g_i)|g_i|^p)_{i}$ for a fixed $p>0$, and then feeds the transformed gradient into a base optimizer. Notably, GradPower requires only a single-line code change and no modifications to the base optimizer's internal logic, including the hyperparameters. When applied to Adam (termed AdamPower), GradPower consistently achieves lower terminal loss across div"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.24275","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/2505.24275/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":"2505.24275","created_at":"2026-05-21T01:04:14.251716+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.24275v3","created_at":"2026-05-21T01:04:14.251716+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.24275","created_at":"2026-05-21T01:04:14.251716+00:00"},{"alias_kind":"pith_short_12","alias_value":"3JKBJ6XXKLPW","created_at":"2026-05-21T01:04:14.251716+00:00"},{"alias_kind":"pith_short_16","alias_value":"3JKBJ6XXKLPWAYFE","created_at":"2026-05-21T01:04:14.251716+00:00"},{"alias_kind":"pith_short_8","alias_value":"3JKBJ6XX","created_at":"2026-05-21T01:04:14.251716+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.05794","citing_title":"Revealing Modular Gradient Noise Imbalance in LLMs: Calibrating Adam via Signal-to-Noise Ratio","ref_index":38,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3JKBJ6XXKLPWAYFEI2OINYRHTN","json":"https://pith.science/pith/3JKBJ6XXKLPWAYFEI2OINYRHTN.json","graph_json":"https://pith.science/api/pith-number/3JKBJ6XXKLPWAYFEI2OINYRHTN/graph.json","events_json":"https://pith.science/api/pith-number/3JKBJ6XXKLPWAYFEI2OINYRHTN/events.json","paper":"https://pith.science/paper/3JKBJ6XX"},"agent_actions":{"view_html":"https://pith.science/pith/3JKBJ6XXKLPWAYFEI2OINYRHTN","download_json":"https://pith.science/pith/3JKBJ6XXKLPWAYFEI2OINYRHTN.json","view_paper":"https://pith.science/paper/3JKBJ6XX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.24275&json=true","fetch_graph":"https://pith.science/api/pith-number/3JKBJ6XXKLPWAYFEI2OINYRHTN/graph.json","fetch_events":"https://pith.science/api/pith-number/3JKBJ6XXKLPWAYFEI2OINYRHTN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3JKBJ6XXKLPWAYFEI2OINYRHTN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3JKBJ6XXKLPWAYFEI2OINYRHTN/action/storage_attestation","attest_author":"https://pith.science/pith/3JKBJ6XXKLPWAYFEI2OINYRHTN/action/author_attestation","sign_citation":"https://pith.science/pith/3JKBJ6XXKLPWAYFEI2OINYRHTN/action/citation_signature","submit_replication":"https://pith.science/pith/3JKBJ6XXKLPWAYFEI2OINYRHTN/action/replication_record"}},"created_at":"2026-05-21T01:04:14.251716+00:00","updated_at":"2026-05-21T01:04:14.251716+00:00"}