{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:RVMXILUTMK7CN6VNE44QCZ2K7O","short_pith_number":"pith:RVMXILUT","schema_version":"1.0","canonical_sha256":"8d59742e9362be26faad273901674afba21a4c01d30c43f44c742ddd97973728","source":{"kind":"arxiv","id":"2505.06653","version":1},"attestation_state":"computed","paper":{"title":"Improving Block-Wise LLM Quantization by 4-bit Block-Wise Optimal Float (BOF4): Analysis and Variations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Patrick Blumenberg, Thomas Graave, Tim Fingscheidt","submitted_at":"2025-05-10T14:00:15Z","abstract_excerpt":"Large language models (LLMs) demand extensive memory capacity during both fine-tuning and inference. To enable memory-efficient fine-tuning, existing methods apply block-wise quantization techniques, such as NF4 and AF4, to the network weights. We show that these quantization techniques incur suboptimal quantization errors. Therefore, as a first novelty, we propose an optimization approach for block-wise quantization. Using this method, we design a family of quantizers named 4-bit block-wise optimal float (BOF4), which consistently reduces the quantization error compared to both baseline metho"},"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":"2505.06653","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-10T14:00:15Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"9950eb31b599b135e69fa156a68a5f5fd82f847ce91fd8401eac751493889e87","abstract_canon_sha256":"9cc87f6bba7dd3ecc24575664ee13cfa8426180b723839511f32af31b03a5f2a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:01:13.760983Z","signature_b64":"S5oj+bQ1hrv2H4mzWHxlTv9odAzIzxkPwMzrKw+E/qzdhcyNzqrrf2vuwQ39MhaDlpAdbYvzqSWJEgRiC6KpCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8d59742e9362be26faad273901674afba21a4c01d30c43f44c742ddd97973728","last_reissued_at":"2026-07-05T11:01:13.760538Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:01:13.760538Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Improving Block-Wise LLM Quantization by 4-bit Block-Wise Optimal Float (BOF4): Analysis and Variations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Patrick Blumenberg, Thomas Graave, Tim Fingscheidt","submitted_at":"2025-05-10T14:00:15Z","abstract_excerpt":"Large language models (LLMs) demand extensive memory capacity during both fine-tuning and inference. To enable memory-efficient fine-tuning, existing methods apply block-wise quantization techniques, such as NF4 and AF4, to the network weights. We show that these quantization techniques incur suboptimal quantization errors. Therefore, as a first novelty, we propose an optimization approach for block-wise quantization. Using this method, we design a family of quantizers named 4-bit block-wise optimal float (BOF4), which consistently reduces the quantization error compared to both baseline metho"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.06653","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/2505.06653/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":"2505.06653","created_at":"2026-07-05T11:01:13.760609+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.06653v1","created_at":"2026-07-05T11:01:13.760609+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.06653","created_at":"2026-07-05T11:01:13.760609+00:00"},{"alias_kind":"pith_short_12","alias_value":"RVMXILUTMK7C","created_at":"2026-07-05T11:01:13.760609+00:00"},{"alias_kind":"pith_short_16","alias_value":"RVMXILUTMK7CN6VN","created_at":"2026-07-05T11:01:13.760609+00:00"},{"alias_kind":"pith_short_8","alias_value":"RVMXILUT","created_at":"2026-07-05T11:01:13.760609+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.14929","citing_title":"A Hardware-Aware, Per-Layer Methodology for Post-Training Quantization of Large Language Models","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12327","citing_title":"Grid Games: The Power of Multiple Grids for Quantizing Large Language Models","ref_index":3,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RVMXILUTMK7CN6VNE44QCZ2K7O","json":"https://pith.science/pith/RVMXILUTMK7CN6VNE44QCZ2K7O.json","graph_json":"https://pith.science/api/pith-number/RVMXILUTMK7CN6VNE44QCZ2K7O/graph.json","events_json":"https://pith.science/api/pith-number/RVMXILUTMK7CN6VNE44QCZ2K7O/events.json","paper":"https://pith.science/paper/RVMXILUT"},"agent_actions":{"view_html":"https://pith.science/pith/RVMXILUTMK7CN6VNE44QCZ2K7O","download_json":"https://pith.science/pith/RVMXILUTMK7CN6VNE44QCZ2K7O.json","view_paper":"https://pith.science/paper/RVMXILUT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.06653&json=true","fetch_graph":"https://pith.science/api/pith-number/RVMXILUTMK7CN6VNE44QCZ2K7O/graph.json","fetch_events":"https://pith.science/api/pith-number/RVMXILUTMK7CN6VNE44QCZ2K7O/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RVMXILUTMK7CN6VNE44QCZ2K7O/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RVMXILUTMK7CN6VNE44QCZ2K7O/action/storage_attestation","attest_author":"https://pith.science/pith/RVMXILUTMK7CN6VNE44QCZ2K7O/action/author_attestation","sign_citation":"https://pith.science/pith/RVMXILUTMK7CN6VNE44QCZ2K7O/action/citation_signature","submit_replication":"https://pith.science/pith/RVMXILUTMK7CN6VNE44QCZ2K7O/action/replication_record"}},"created_at":"2026-07-05T11:01:13.760609+00:00","updated_at":"2026-07-05T11:01:13.760609+00:00"}