{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:CIMM4MMPMW4LJLHM25E2TE4UXC","short_pith_number":"pith:CIMM4MMP","schema_version":"1.0","canonical_sha256":"1218ce318f65b8b4acecd749a99394b89af8be6786311733f48421a84d52a120","source":{"kind":"arxiv","id":"2502.20853","version":2},"attestation_state":"computed","paper":{"title":"Oscillation-Reduced MXFP4 Training for Vision Transformers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Haocheng Xi, Jianfei Chen, Jun Zhu, Yuxiang Chen","submitted_at":"2025-02-28T08:51:55Z","abstract_excerpt":"Pre-training Transformers in FP4 precision is becoming a promising approach to gain substantial speedup, but it comes with a considerable loss of accuracy. Microscaling (MX) data format provides a fine-grained per-group quantization method to improve the representation ability of the FP4 format and is supported by the next-generation Blackwell GPU architecture. However, training with MXFP4 data format still results in significant degradation and there is a lack of systematic research on the reason.\n  In this work, we propose a novel training method TetraJet for a more accurate FP4 training. We"},"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":"2502.20853","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-28T08:51:55Z","cross_cats_sorted":["cs.AI","cs.CV"],"title_canon_sha256":"bfb4950549fe967b7ff566d5a06970944ac02bdf4f71509081ef4e43a1e83531","abstract_canon_sha256":"5f15dd4918498bd292ac8364dc7201542049a266af826671367cebc6bca1fbe2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:34:05.036747Z","signature_b64":"C51MLDUeEjqjaExqTwbKDsV6bpmT8BTtQ6NGwhQqSLiYTt2JCE3/y4b4cPhCeOom3EqHgyAUT+t0rmSVeC+KAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1218ce318f65b8b4acecd749a99394b89af8be6786311733f48421a84d52a120","last_reissued_at":"2026-07-05T11:34:05.036281Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:34:05.036281Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Oscillation-Reduced MXFP4 Training for Vision Transformers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Haocheng Xi, Jianfei Chen, Jun Zhu, Yuxiang Chen","submitted_at":"2025-02-28T08:51:55Z","abstract_excerpt":"Pre-training Transformers in FP4 precision is becoming a promising approach to gain substantial speedup, but it comes with a considerable loss of accuracy. Microscaling (MX) data format provides a fine-grained per-group quantization method to improve the representation ability of the FP4 format and is supported by the next-generation Blackwell GPU architecture. However, training with MXFP4 data format still results in significant degradation and there is a lack of systematic research on the reason.\n  In this work, we propose a novel training method TetraJet for a more accurate FP4 training. We"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.20853","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/2502.20853/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":"2502.20853","created_at":"2026-07-05T11:34:05.036354+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.20853v2","created_at":"2026-07-05T11:34:05.036354+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.20853","created_at":"2026-07-05T11:34:05.036354+00:00"},{"alias_kind":"pith_short_12","alias_value":"CIMM4MMPMW4L","created_at":"2026-07-05T11:34:05.036354+00:00"},{"alias_kind":"pith_short_16","alias_value":"CIMM4MMPMW4LJLHM","created_at":"2026-07-05T11:34:05.036354+00:00"},{"alias_kind":"pith_short_8","alias_value":"CIMM4MMP","created_at":"2026-07-05T11:34:05.036354+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26587","citing_title":"SharQ: Bridging Activation Sparsity and FP4 Quantization for LLM Inference","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.27616","citing_title":"Not All NVFP4 QAT Recipes Are Equal: How Architecture and Scale Shape Model Quality for Anomaly Segmentation","ref_index":26,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CIMM4MMPMW4LJLHM25E2TE4UXC","json":"https://pith.science/pith/CIMM4MMPMW4LJLHM25E2TE4UXC.json","graph_json":"https://pith.science/api/pith-number/CIMM4MMPMW4LJLHM25E2TE4UXC/graph.json","events_json":"https://pith.science/api/pith-number/CIMM4MMPMW4LJLHM25E2TE4UXC/events.json","paper":"https://pith.science/paper/CIMM4MMP"},"agent_actions":{"view_html":"https://pith.science/pith/CIMM4MMPMW4LJLHM25E2TE4UXC","download_json":"https://pith.science/pith/CIMM4MMPMW4LJLHM25E2TE4UXC.json","view_paper":"https://pith.science/paper/CIMM4MMP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.20853&json=true","fetch_graph":"https://pith.science/api/pith-number/CIMM4MMPMW4LJLHM25E2TE4UXC/graph.json","fetch_events":"https://pith.science/api/pith-number/CIMM4MMPMW4LJLHM25E2TE4UXC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CIMM4MMPMW4LJLHM25E2TE4UXC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CIMM4MMPMW4LJLHM25E2TE4UXC/action/storage_attestation","attest_author":"https://pith.science/pith/CIMM4MMPMW4LJLHM25E2TE4UXC/action/author_attestation","sign_citation":"https://pith.science/pith/CIMM4MMPMW4LJLHM25E2TE4UXC/action/citation_signature","submit_replication":"https://pith.science/pith/CIMM4MMPMW4LJLHM25E2TE4UXC/action/replication_record"}},"created_at":"2026-07-05T11:34:05.036354+00:00","updated_at":"2026-07-05T11:34:05.036354+00:00"}