{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:4OQXDRKL6VJJ5TORWEZGWEN6C7","short_pith_number":"pith:4OQXDRKL","schema_version":"1.0","canonical_sha256":"e3a171c54bf5529ecdd1b1326b11be17e0f963a1457ddc8d6f2fa12c78efbe66","source":{"kind":"arxiv","id":"2403.12422","version":2},"attestation_state":"computed","paper":{"title":"Jetfire: Efficient and Accurate Transformer Pretraining with INT8 Data Flow and Per-Block Quantization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Haocheng Xi, Jianfei Chen, Jun Zhu, Kai Jun Teh, Kang Zhao, Yuxiang Chen","submitted_at":"2024-03-19T04:09:11Z","abstract_excerpt":"Pretraining transformers are generally time-consuming. Fully quantized training (FQT) is a promising approach to speed up pretraining. However, most FQT methods adopt a quantize-compute-dequantize procedure, which often leads to suboptimal speedup and significant performance degradation when used in transformers due to the high memory access overheads and low-precision computations. In this work, we propose Jetfire, an efficient and accurate INT8 training method specific to transformers. Our method features an INT8 data flow to optimize memory access and a per-block quantization method to main"},"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":"2403.12422","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-03-19T04:09:11Z","cross_cats_sorted":[],"title_canon_sha256":"da9fac09d780fd134e8c86391af53b3042026519df61857417aae3a01ade72d7","abstract_canon_sha256":"b7ec8bdd4e548b830985c2e7763a85c3009016801d3bb5aaec0e65df081c297d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:46:23.119769Z","signature_b64":"QOI6z+lTItq6GOKHzNBw749fc3CuX86XjxXIwnTEQCmYS3wwjVONtw1VoSgqNDgzm6SrsbAh0P97EKwerODOCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e3a171c54bf5529ecdd1b1326b11be17e0f963a1457ddc8d6f2fa12c78efbe66","last_reissued_at":"2026-07-05T08:46:23.119292Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:46:23.119292Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Jetfire: Efficient and Accurate Transformer Pretraining with INT8 Data Flow and Per-Block Quantization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Haocheng Xi, Jianfei Chen, Jun Zhu, Kai Jun Teh, Kang Zhao, Yuxiang Chen","submitted_at":"2024-03-19T04:09:11Z","abstract_excerpt":"Pretraining transformers are generally time-consuming. Fully quantized training (FQT) is a promising approach to speed up pretraining. However, most FQT methods adopt a quantize-compute-dequantize procedure, which often leads to suboptimal speedup and significant performance degradation when used in transformers due to the high memory access overheads and low-precision computations. In this work, we propose Jetfire, an efficient and accurate INT8 training method specific to transformers. Our method features an INT8 data flow to optimize memory access and a per-block quantization method to main"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.12422","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/2403.12422/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":"2403.12422","created_at":"2026-07-05T08:46:23.119358+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.12422v2","created_at":"2026-07-05T08:46:23.119358+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.12422","created_at":"2026-07-05T08:46:23.119358+00:00"},{"alias_kind":"pith_short_12","alias_value":"4OQXDRKL6VJJ","created_at":"2026-07-05T08:46:23.119358+00:00"},{"alias_kind":"pith_short_16","alias_value":"4OQXDRKL6VJJ5TOR","created_at":"2026-07-05T08:46:23.119358+00:00"},{"alias_kind":"pith_short_8","alias_value":"4OQXDRKL","created_at":"2026-07-05T08:46:23.119358+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.06836","citing_title":"STQuant: Spatio-Temporal Adaptive Framework for Optimizer Quantization in Large Multimodal Model Training","ref_index":29,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4OQXDRKL6VJJ5TORWEZGWEN6C7","json":"https://pith.science/pith/4OQXDRKL6VJJ5TORWEZGWEN6C7.json","graph_json":"https://pith.science/api/pith-number/4OQXDRKL6VJJ5TORWEZGWEN6C7/graph.json","events_json":"https://pith.science/api/pith-number/4OQXDRKL6VJJ5TORWEZGWEN6C7/events.json","paper":"https://pith.science/paper/4OQXDRKL"},"agent_actions":{"view_html":"https://pith.science/pith/4OQXDRKL6VJJ5TORWEZGWEN6C7","download_json":"https://pith.science/pith/4OQXDRKL6VJJ5TORWEZGWEN6C7.json","view_paper":"https://pith.science/paper/4OQXDRKL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.12422&json=true","fetch_graph":"https://pith.science/api/pith-number/4OQXDRKL6VJJ5TORWEZGWEN6C7/graph.json","fetch_events":"https://pith.science/api/pith-number/4OQXDRKL6VJJ5TORWEZGWEN6C7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4OQXDRKL6VJJ5TORWEZGWEN6C7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4OQXDRKL6VJJ5TORWEZGWEN6C7/action/storage_attestation","attest_author":"https://pith.science/pith/4OQXDRKL6VJJ5TORWEZGWEN6C7/action/author_attestation","sign_citation":"https://pith.science/pith/4OQXDRKL6VJJ5TORWEZGWEN6C7/action/citation_signature","submit_replication":"https://pith.science/pith/4OQXDRKL6VJJ5TORWEZGWEN6C7/action/replication_record"}},"created_at":"2026-07-05T08:46:23.119358+00:00","updated_at":"2026-07-05T08:46:23.119358+00:00"}