{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:6BBITQMBZFXIS5BUMBL2H22W5T","short_pith_number":"pith:6BBITQMB","schema_version":"1.0","canonical_sha256":"f04289c181c96e8974346057a3eb56ececd38686d9bd0cfca475676ebba66ee6","source":{"kind":"arxiv","id":"2501.17116","version":2},"attestation_state":"computed","paper":{"title":"Optimizing Large Language Model Training Using FP4 Quantization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Baining Guo, Guoshuai Zhao, Peng Cheng, Ruizhe Wang, Xiao Liu, Yeyun Gong, Zhengjun Zha, Ziyue Yang","submitted_at":"2025-01-28T18:04:50Z","abstract_excerpt":"The growing computational demands of training large language models (LLMs) necessitate more efficient methods. Quantized training presents a promising solution by enabling low-bit arithmetic operations to reduce these costs. While FP8 precision has demonstrated feasibility, leveraging FP4 remains a challenge due to significant quantization errors and limited representational capacity. This work introduces the first FP4 training framework for LLMs, addressing these challenges with two key innovations: a differentiable quantization estimator for precise weight updates and an outlier clamping and"},"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":"2501.17116","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-01-28T18:04:50Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"5fa5ccf226a058f442e77cc37a69241ad3c7a00c95ca276cb50805274ba60dde","abstract_canon_sha256":"531b8ff0db66be5054d0fe89ba71a312d154718432ab36a2317d0afd930526af"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:00:17.404678Z","signature_b64":"Lvp8FgJcmQprNUYHE5MpcNjG5S7SyeGV1evX4v4h9qemogJLu5lhlsSYB6u/k9n26qU+bOVm/lAII25cfXJ2AA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f04289c181c96e8974346057a3eb56ececd38686d9bd0cfca475676ebba66ee6","last_reissued_at":"2026-05-20T00:00:17.404117Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:00:17.404117Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Optimizing Large Language Model Training Using FP4 Quantization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Baining Guo, Guoshuai Zhao, Peng Cheng, Ruizhe Wang, Xiao Liu, Yeyun Gong, Zhengjun Zha, Ziyue Yang","submitted_at":"2025-01-28T18:04:50Z","abstract_excerpt":"The growing computational demands of training large language models (LLMs) necessitate more efficient methods. Quantized training presents a promising solution by enabling low-bit arithmetic operations to reduce these costs. While FP8 precision has demonstrated feasibility, leveraging FP4 remains a challenge due to significant quantization errors and limited representational capacity. This work introduces the first FP4 training framework for LLMs, addressing these challenges with two key innovations: a differentiable quantization estimator for precise weight updates and an outlier clamping and"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.17116","kind":"arxiv","version":2},"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/2501.17116/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":"2501.17116","created_at":"2026-05-20T00:00:17.404214+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.17116v2","created_at":"2026-05-20T00:00:17.404214+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.17116","created_at":"2026-05-20T00:00:17.404214+00:00"},{"alias_kind":"pith_short_12","alias_value":"6BBITQMBZFXI","created_at":"2026-05-20T00:00:17.404214+00:00"},{"alias_kind":"pith_short_16","alias_value":"6BBITQMBZFXIS5BU","created_at":"2026-05-20T00:00:17.404214+00:00"},{"alias_kind":"pith_short_8","alias_value":"6BBITQMB","created_at":"2026-05-20T00:00:17.404214+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":10,"internal_anchor_count":10,"sample":[{"citing_arxiv_id":"2510.08008","citing_title":"Beyond Sunk Costs: Boosting LLM Pre-training Efficiency via Orthogonal Growth of Mixture-of-Experts","ref_index":23,"is_internal_anchor":true},{"citing_arxiv_id":"2510.04212","citing_title":"Why Low-Precision Transformer Training Fails: An Analysis on Flash Attention","ref_index":29,"is_internal_anchor":true},{"citing_arxiv_id":"2512.02010","citing_title":"Four Over Six: More Accurate NVFP4 Quantization with Adaptive Block Scaling","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2605.09825","citing_title":"Pretraining large language models with MXFP4 on Native FP4 Hardware","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10886","citing_title":"LoKA: Low-precision Kernel Applications for Recommendation Models At Scale","ref_index":73,"is_internal_anchor":true},{"citing_arxiv_id":"2605.09825","citing_title":"Pretraining large language models with MXFP4 on Native FP4 Hardware","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10886","citing_title":"LoKA: Low-precision Kernel Applications for Recommendation Models At Scale","ref_index":73,"is_internal_anchor":true},{"citing_arxiv_id":"2605.09825","citing_title":"Pretraining large language models with MXFP4 on Native FP4 Hardware","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06067","citing_title":"Normalized Architectures are Natively 4-Bit","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2604.08826","citing_title":"HiFloat4 Format for Language Model Pre-training on Ascend NPUs","ref_index":14,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6BBITQMBZFXIS5BUMBL2H22W5T","json":"https://pith.science/pith/6BBITQMBZFXIS5BUMBL2H22W5T.json","graph_json":"https://pith.science/api/pith-number/6BBITQMBZFXIS5BUMBL2H22W5T/graph.json","events_json":"https://pith.science/api/pith-number/6BBITQMBZFXIS5BUMBL2H22W5T/events.json","paper":"https://pith.science/paper/6BBITQMB"},"agent_actions":{"view_html":"https://pith.science/pith/6BBITQMBZFXIS5BUMBL2H22W5T","download_json":"https://pith.science/pith/6BBITQMBZFXIS5BUMBL2H22W5T.json","view_paper":"https://pith.science/paper/6BBITQMB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.17116&json=true","fetch_graph":"https://pith.science/api/pith-number/6BBITQMBZFXIS5BUMBL2H22W5T/graph.json","fetch_events":"https://pith.science/api/pith-number/6BBITQMBZFXIS5BUMBL2H22W5T/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6BBITQMBZFXIS5BUMBL2H22W5T/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6BBITQMBZFXIS5BUMBL2H22W5T/action/storage_attestation","attest_author":"https://pith.science/pith/6BBITQMBZFXIS5BUMBL2H22W5T/action/author_attestation","sign_citation":"https://pith.science/pith/6BBITQMBZFXIS5BUMBL2H22W5T/action/citation_signature","submit_replication":"https://pith.science/pith/6BBITQMBZFXIS5BUMBL2H22W5T/action/replication_record"}},"created_at":"2026-05-20T00:00:17.404214+00:00","updated_at":"2026-05-20T00:00:17.404214+00:00"}