{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:67FL66SPUJLAWGM2NWN5SHM7ZP","short_pith_number":"pith:67FL66SP","schema_version":"1.0","canonical_sha256":"f7cabf7a4fa2560b199a6d9bd91d9fcbe925784b044dc9df9a72da33c89e8159","source":{"kind":"arxiv","id":"2405.18144","version":3},"attestation_state":"computed","paper":{"title":"4-bit Shampoo for Memory-Efficient Network Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Hua Huang, Jia Li, Pan Zhou, Sike Wang","submitted_at":"2024-05-28T13:02:56Z","abstract_excerpt":"Second-order optimizers, maintaining a matrix termed a preconditioner, are superior to first-order optimizers in both theory and practice. The states forming the preconditioner and its inverse root restrict the maximum size of models trained by second-order optimizers. To address this, compressing 32-bit optimizer states to lower bitwidths has shown promise in reducing memory usage. However, current approaches only pertain to first-order optimizers. In this paper, we propose the first 4-bit second-order optimizers, exemplified by 4-bit Shampoo, maintaining performance similar to that of 32-bit"},"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":"2405.18144","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-05-28T13:02:56Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"f9c96f77c9fce37019e13f1d9e9a5e0377163930247e242fb465f43df536e360","abstract_canon_sha256":"fbfbebe1fd3281269cd0c2e52b06e1bcd680a82cc97e55052d978261fab4c851"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:59:15.885566Z","signature_b64":"PfD6SuPHOwqMywDKnSsEZzGYJuSwOdIkC65UYNxIQMjBj6eocNVgqh9aBTBIX8D2ka9l439vsxq8l0s6eXEaDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f7cabf7a4fa2560b199a6d9bd91d9fcbe925784b044dc9df9a72da33c89e8159","last_reissued_at":"2026-07-05T09:59:15.885126Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:59:15.885126Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"4-bit Shampoo for Memory-Efficient Network Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Hua Huang, Jia Li, Pan Zhou, Sike Wang","submitted_at":"2024-05-28T13:02:56Z","abstract_excerpt":"Second-order optimizers, maintaining a matrix termed a preconditioner, are superior to first-order optimizers in both theory and practice. The states forming the preconditioner and its inverse root restrict the maximum size of models trained by second-order optimizers. To address this, compressing 32-bit optimizer states to lower bitwidths has shown promise in reducing memory usage. However, current approaches only pertain to first-order optimizers. In this paper, we propose the first 4-bit second-order optimizers, exemplified by 4-bit Shampoo, maintaining performance similar to that of 32-bit"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.18144","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/2405.18144/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":"2405.18144","created_at":"2026-07-05T09:59:15.885200+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.18144v3","created_at":"2026-07-05T09:59:15.885200+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.18144","created_at":"2026-07-05T09:59:15.885200+00:00"},{"alias_kind":"pith_short_12","alias_value":"67FL66SPUJLA","created_at":"2026-07-05T09:59:15.885200+00:00"},{"alias_kind":"pith_short_16","alias_value":"67FL66SPUJLAWGM2","created_at":"2026-07-05T09:59:15.885200+00:00"},{"alias_kind":"pith_short_8","alias_value":"67FL66SP","created_at":"2026-07-05T09:59:15.885200+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2505.22922","citing_title":"Scalable Parameter and Memory Efficient Pretraining for LLM: Recent Algorithmic Advances and Benchmarking","ref_index":45,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/67FL66SPUJLAWGM2NWN5SHM7ZP","json":"https://pith.science/pith/67FL66SPUJLAWGM2NWN5SHM7ZP.json","graph_json":"https://pith.science/api/pith-number/67FL66SPUJLAWGM2NWN5SHM7ZP/graph.json","events_json":"https://pith.science/api/pith-number/67FL66SPUJLAWGM2NWN5SHM7ZP/events.json","paper":"https://pith.science/paper/67FL66SP"},"agent_actions":{"view_html":"https://pith.science/pith/67FL66SPUJLAWGM2NWN5SHM7ZP","download_json":"https://pith.science/pith/67FL66SPUJLAWGM2NWN5SHM7ZP.json","view_paper":"https://pith.science/paper/67FL66SP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.18144&json=true","fetch_graph":"https://pith.science/api/pith-number/67FL66SPUJLAWGM2NWN5SHM7ZP/graph.json","fetch_events":"https://pith.science/api/pith-number/67FL66SPUJLAWGM2NWN5SHM7ZP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/67FL66SPUJLAWGM2NWN5SHM7ZP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/67FL66SPUJLAWGM2NWN5SHM7ZP/action/storage_attestation","attest_author":"https://pith.science/pith/67FL66SPUJLAWGM2NWN5SHM7ZP/action/author_attestation","sign_citation":"https://pith.science/pith/67FL66SPUJLAWGM2NWN5SHM7ZP/action/citation_signature","submit_replication":"https://pith.science/pith/67FL66SPUJLAWGM2NWN5SHM7ZP/action/replication_record"}},"created_at":"2026-07-05T09:59:15.885200+00:00","updated_at":"2026-07-05T09:59:15.885200+00:00"}