{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:WS4AZZ2D6LOA2IIP4AOP5BT4ZE","short_pith_number":"pith:WS4AZZ2D","schema_version":"1.0","canonical_sha256":"b4b80ce743f2dc0d210fe01cfe867cc9340354f7daf38f72127d12a9b2c99d60","source":{"kind":"arxiv","id":"2512.00956","version":3},"attestation_state":"computed","paper":{"title":"WUSH: Near-Optimal Adaptive Transforms for LLM Quantization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Dan Alistarh, Jiale Chen, Roberto L. Castro, Torsten Hoefler, Vage Egiazarian","submitted_at":"2025-11-30T16:17:34Z","abstract_excerpt":"Quantizing LLM weights and activations is a standard approach for efficient deployment, but a few extreme outliers can stretch the dynamic range and amplify low-bit quantization errors. Prior transform-based mitigations (e.g., Hadamard rotations) are fixed and data-agnostic, and their optimality for quantization has remained unclear. We derive closed-form optimal linear blockwise transforms for joint weight-activation quantization under standard RTN AbsMax-scaled block quantizers, covering both integer and floating-point formats. The resulting construction, WUSH, combines a Hadamard backbone w"},"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":"2512.00956","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-11-30T16:17:34Z","cross_cats_sorted":[],"title_canon_sha256":"1bfc0e1bdc6179330c6a8793a09db3421f0afc92c9091e17146ccd8405f20ed3","abstract_canon_sha256":"713026ed4061b255b1e121748ef3c5e75756a455edcd62c364bd04af1a9d05e5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T01:03:38.901336Z","signature_b64":"p8wMTcoJOCSFurRsHY4TzqqmgYwq8LupPxaR/lwS4kgbqdGGYT7qIIFR4fq/hyJLPtp/M8iQZzKu4sB78zhVCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b4b80ce743f2dc0d210fe01cfe867cc9340354f7daf38f72127d12a9b2c99d60","last_reissued_at":"2026-06-02T01:03:38.900748Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T01:03:38.900748Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"WUSH: Near-Optimal Adaptive Transforms for LLM Quantization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Dan Alistarh, Jiale Chen, Roberto L. Castro, Torsten Hoefler, Vage Egiazarian","submitted_at":"2025-11-30T16:17:34Z","abstract_excerpt":"Quantizing LLM weights and activations is a standard approach for efficient deployment, but a few extreme outliers can stretch the dynamic range and amplify low-bit quantization errors. Prior transform-based mitigations (e.g., Hadamard rotations) are fixed and data-agnostic, and their optimality for quantization has remained unclear. We derive closed-form optimal linear blockwise transforms for joint weight-activation quantization under standard RTN AbsMax-scaled block quantizers, covering both integer and floating-point formats. The resulting construction, WUSH, combines a Hadamard backbone w"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2512.00956","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/2512.00956/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":"2512.00956","created_at":"2026-06-02T01:03:38.900822+00:00"},{"alias_kind":"arxiv_version","alias_value":"2512.00956v3","created_at":"2026-06-02T01:03:38.900822+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2512.00956","created_at":"2026-06-02T01:03:38.900822+00:00"},{"alias_kind":"pith_short_12","alias_value":"WS4AZZ2D6LOA","created_at":"2026-06-02T01:03:38.900822+00:00"},{"alias_kind":"pith_short_16","alias_value":"WS4AZZ2D6LOA2IIP","created_at":"2026-06-02T01:03:38.900822+00:00"},{"alias_kind":"pith_short_8","alias_value":"WS4AZZ2D","created_at":"2026-06-02T01:03:38.900822+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2601.17187","citing_title":"High-Rate Quantized Matrix Multiplication I","ref_index":25,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WS4AZZ2D6LOA2IIP4AOP5BT4ZE","json":"https://pith.science/pith/WS4AZZ2D6LOA2IIP4AOP5BT4ZE.json","graph_json":"https://pith.science/api/pith-number/WS4AZZ2D6LOA2IIP4AOP5BT4ZE/graph.json","events_json":"https://pith.science/api/pith-number/WS4AZZ2D6LOA2IIP4AOP5BT4ZE/events.json","paper":"https://pith.science/paper/WS4AZZ2D"},"agent_actions":{"view_html":"https://pith.science/pith/WS4AZZ2D6LOA2IIP4AOP5BT4ZE","download_json":"https://pith.science/pith/WS4AZZ2D6LOA2IIP4AOP5BT4ZE.json","view_paper":"https://pith.science/paper/WS4AZZ2D","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2512.00956&json=true","fetch_graph":"https://pith.science/api/pith-number/WS4AZZ2D6LOA2IIP4AOP5BT4ZE/graph.json","fetch_events":"https://pith.science/api/pith-number/WS4AZZ2D6LOA2IIP4AOP5BT4ZE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WS4AZZ2D6LOA2IIP4AOP5BT4ZE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WS4AZZ2D6LOA2IIP4AOP5BT4ZE/action/storage_attestation","attest_author":"https://pith.science/pith/WS4AZZ2D6LOA2IIP4AOP5BT4ZE/action/author_attestation","sign_citation":"https://pith.science/pith/WS4AZZ2D6LOA2IIP4AOP5BT4ZE/action/citation_signature","submit_replication":"https://pith.science/pith/WS4AZZ2D6LOA2IIP4AOP5BT4ZE/action/replication_record"}},"created_at":"2026-06-02T01:03:38.900822+00:00","updated_at":"2026-06-02T01:03:38.900822+00:00"}