{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:CSWEHIKURWZBJ3RNXZXGBXMI46","short_pith_number":"pith:CSWEHIKU","schema_version":"1.0","canonical_sha256":"14ac43a1548db214ee2dbe6e60dd88e797b50293624b6ffb65434b0895808b6c","source":{"kind":"arxiv","id":"2605.17471","version":1},"attestation_state":"computed","paper":{"title":"WinQ: Accelerating Quantization-Aware Training of Language Models Around Saddle Points","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.NA","math.NA"],"primary_cat":"cs.LG","authors_text":"Changsheng Zhao, Dongyue Li, Harshit Khaitan, Hongyang R. Zhang, Kai Yi, Raghuraman Krishnamoorthi, Steven Li, Zechun Liu, Zhenshuo Zhang","submitted_at":"2026-05-17T14:20:51Z","abstract_excerpt":"Quantization-aware training (QAT) is widely adopted to quantize language models by training full-precision weights using gradients from the quantized model. The main bottleneck is its slow convergence and early performance plateau, particularly below 4-bit-widths. While this problem has been observed in prior work, its precise cause remains unclear. In this paper, we analyze the convergence of QAT by estimating the spectrum of the loss-surface Hessians. We find that the weights converge to flat regions around saddle points, where a large fraction of the Hessian eigenvalues are both positive an"},"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":"2605.17471","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-17T14:20:51Z","cross_cats_sorted":["cs.NA","math.NA"],"title_canon_sha256":"6a942f85a6ca1abc8ada0e8a577afedf6d003d37849b7aed7cad297e90443139","abstract_canon_sha256":"06889f7640dac651dfb09aedea84e5d876436f42bd8cf88b9f34be3fae857fbb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:04:40.698441Z","signature_b64":"1MocaXbtrvxxV5PmhzP4x/s97tECPGtezv0HCXnKXDiIBbjEmoe3CyVi9uD2vnLnBXSCSG/q0609E6RF7JmTCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"14ac43a1548db214ee2dbe6e60dd88e797b50293624b6ffb65434b0895808b6c","last_reissued_at":"2026-05-20T00:04:40.697667Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:04:40.697667Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"WinQ: Accelerating Quantization-Aware Training of Language Models Around Saddle Points","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.NA","math.NA"],"primary_cat":"cs.LG","authors_text":"Changsheng Zhao, Dongyue Li, Harshit Khaitan, Hongyang R. Zhang, Kai Yi, Raghuraman Krishnamoorthi, Steven Li, Zechun Liu, Zhenshuo Zhang","submitted_at":"2026-05-17T14:20:51Z","abstract_excerpt":"Quantization-aware training (QAT) is widely adopted to quantize language models by training full-precision weights using gradients from the quantized model. The main bottleneck is its slow convergence and early performance plateau, particularly below 4-bit-widths. While this problem has been observed in prior work, its precise cause remains unclear. In this paper, we analyze the convergence of QAT by estimating the spectrum of the loss-surface Hessians. We find that the weights converge to flat regions around saddle points, where a large fraction of the Hessian eigenvalues are both positive an"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.17471","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/2605.17471/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-19T21:41:57.697233Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T21:33:23.654040Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"91ecb93a39a4ec77857f2f637d83a5935afed225a46e974cb663fb4ddc55c8cc"},"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":"2605.17471","created_at":"2026-05-20T00:04:40.697808+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.17471v1","created_at":"2026-05-20T00:04:40.697808+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.17471","created_at":"2026-05-20T00:04:40.697808+00:00"},{"alias_kind":"pith_short_12","alias_value":"CSWEHIKURWZB","created_at":"2026-05-20T00:04:40.697808+00:00"},{"alias_kind":"pith_short_16","alias_value":"CSWEHIKURWZBJ3RN","created_at":"2026-05-20T00:04:40.697808+00:00"},{"alias_kind":"pith_short_8","alias_value":"CSWEHIKU","created_at":"2026-05-20T00:04:40.697808+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CSWEHIKURWZBJ3RNXZXGBXMI46","json":"https://pith.science/pith/CSWEHIKURWZBJ3RNXZXGBXMI46.json","graph_json":"https://pith.science/api/pith-number/CSWEHIKURWZBJ3RNXZXGBXMI46/graph.json","events_json":"https://pith.science/api/pith-number/CSWEHIKURWZBJ3RNXZXGBXMI46/events.json","paper":"https://pith.science/paper/CSWEHIKU"},"agent_actions":{"view_html":"https://pith.science/pith/CSWEHIKURWZBJ3RNXZXGBXMI46","download_json":"https://pith.science/pith/CSWEHIKURWZBJ3RNXZXGBXMI46.json","view_paper":"https://pith.science/paper/CSWEHIKU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.17471&json=true","fetch_graph":"https://pith.science/api/pith-number/CSWEHIKURWZBJ3RNXZXGBXMI46/graph.json","fetch_events":"https://pith.science/api/pith-number/CSWEHIKURWZBJ3RNXZXGBXMI46/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CSWEHIKURWZBJ3RNXZXGBXMI46/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CSWEHIKURWZBJ3RNXZXGBXMI46/action/storage_attestation","attest_author":"https://pith.science/pith/CSWEHIKURWZBJ3RNXZXGBXMI46/action/author_attestation","sign_citation":"https://pith.science/pith/CSWEHIKURWZBJ3RNXZXGBXMI46/action/citation_signature","submit_replication":"https://pith.science/pith/CSWEHIKURWZBJ3RNXZXGBXMI46/action/replication_record"}},"created_at":"2026-05-20T00:04:40.697808+00:00","updated_at":"2026-05-20T00:04:40.697808+00:00"}