{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TOA7DL4O37XZDLS3O2RWWNAYIP","short_pith_number":"pith:TOA7DL4O","schema_version":"1.0","canonical_sha256":"9b81f1af8edfef91ae5b76a36b341843eac7268c36087df2fcb38e6202471acd","source":{"kind":"arxiv","id":"2506.19697","version":1},"attestation_state":"computed","paper":{"title":"Outlier-Safe Pre-Training for Robust 4-Bit Quantization of Large Language Models","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Chanwoong Yoon, Hyeon Hwang, Jaewoo Kang, Jungwoo Park, Taewhoo Lee","submitted_at":"2025-06-24T15:03:57Z","abstract_excerpt":"Extreme activation outliers in Large Language Models (LLMs) critically degrade quantization performance, hindering efficient on-device deployment. While channel-wise operations and adaptive gradient scaling are recognized causes, practical mitigation remains challenging. We introduce Outlier-Safe Pre-Training (OSP), a practical guideline that proactively prevents outlier formation rather than relying on post-hoc mitigation. OSP combines three key innovations: (1) the Muon optimizer, eliminating privileged bases while maintaining training efficiency; (2) Single-Scale RMSNorm, preventing channel"},"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":"2506.19697","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-24T15:03:57Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"de8bd4c76c7908506d39d8654a03c3f467f4c7036c4d135b91b1a1ad6d4ef395","abstract_canon_sha256":"b8001f6458e6962b671ab547efdf9694aa2b754b87e86764fe3feef3ee142a49"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:26:37.230114Z","signature_b64":"DhQGPf/7uQFKfR6lOy42v+fQSFBAT3cOQOstY6zxImre9aVeRRBwBORVB1MMKkM4ZHF9r6q1CtCLEFEAphjZBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9b81f1af8edfef91ae5b76a36b341843eac7268c36087df2fcb38e6202471acd","last_reissued_at":"2026-07-05T11:26:37.229615Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:26:37.229615Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Outlier-Safe Pre-Training for Robust 4-Bit Quantization of Large Language Models","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Chanwoong Yoon, Hyeon Hwang, Jaewoo Kang, Jungwoo Park, Taewhoo Lee","submitted_at":"2025-06-24T15:03:57Z","abstract_excerpt":"Extreme activation outliers in Large Language Models (LLMs) critically degrade quantization performance, hindering efficient on-device deployment. While channel-wise operations and adaptive gradient scaling are recognized causes, practical mitigation remains challenging. We introduce Outlier-Safe Pre-Training (OSP), a practical guideline that proactively prevents outlier formation rather than relying on post-hoc mitigation. OSP combines three key innovations: (1) the Muon optimizer, eliminating privileged bases while maintaining training efficiency; (2) Single-Scale RMSNorm, preventing channel"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.19697","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/2506.19697/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":"2506.19697","created_at":"2026-07-05T11:26:37.229671+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.19697v1","created_at":"2026-07-05T11:26:37.229671+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.19697","created_at":"2026-07-05T11:26:37.229671+00:00"},{"alias_kind":"pith_short_12","alias_value":"TOA7DL4O37XZ","created_at":"2026-07-05T11:26:37.229671+00:00"},{"alias_kind":"pith_short_16","alias_value":"TOA7DL4O37XZDLS3","created_at":"2026-07-05T11:26:37.229671+00:00"},{"alias_kind":"pith_short_8","alias_value":"TOA7DL4O","created_at":"2026-07-05T11:26:37.229671+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.19167","citing_title":"LBLLM: Lightweight Binarization of Large Language Models via Three-Stage Distillation","ref_index":96,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TOA7DL4O37XZDLS3O2RWWNAYIP","json":"https://pith.science/pith/TOA7DL4O37XZDLS3O2RWWNAYIP.json","graph_json":"https://pith.science/api/pith-number/TOA7DL4O37XZDLS3O2RWWNAYIP/graph.json","events_json":"https://pith.science/api/pith-number/TOA7DL4O37XZDLS3O2RWWNAYIP/events.json","paper":"https://pith.science/paper/TOA7DL4O"},"agent_actions":{"view_html":"https://pith.science/pith/TOA7DL4O37XZDLS3O2RWWNAYIP","download_json":"https://pith.science/pith/TOA7DL4O37XZDLS3O2RWWNAYIP.json","view_paper":"https://pith.science/paper/TOA7DL4O","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.19697&json=true","fetch_graph":"https://pith.science/api/pith-number/TOA7DL4O37XZDLS3O2RWWNAYIP/graph.json","fetch_events":"https://pith.science/api/pith-number/TOA7DL4O37XZDLS3O2RWWNAYIP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TOA7DL4O37XZDLS3O2RWWNAYIP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TOA7DL4O37XZDLS3O2RWWNAYIP/action/storage_attestation","attest_author":"https://pith.science/pith/TOA7DL4O37XZDLS3O2RWWNAYIP/action/author_attestation","sign_citation":"https://pith.science/pith/TOA7DL4O37XZDLS3O2RWWNAYIP/action/citation_signature","submit_replication":"https://pith.science/pith/TOA7DL4O37XZDLS3O2RWWNAYIP/action/replication_record"}},"created_at":"2026-07-05T11:26:37.229671+00:00","updated_at":"2026-07-05T11:26:37.229671+00:00"}