{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2022:KZGL6CQJWFLXZEDBB2SHHCPXD5","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"c520978422b7a453110ed090b9d0c9b9786712d389b5a932b2e5184ded23748b","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2022-09-12T17:39:55Z","title_canon_sha256":"a787abcfcac8027bbbf57bf19b5a9d2496b30a55d097c56c0d787ecb293565af"},"schema_version":"1.0","source":{"id":"2209.05433","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2209.05433","created_at":"2026-05-17T23:38:52Z"},{"alias_kind":"arxiv_version","alias_value":"2209.05433v2","created_at":"2026-05-17T23:38:52Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2209.05433","created_at":"2026-05-17T23:38:52Z"},{"alias_kind":"pith_short_12","alias_value":"KZGL6CQJWFLX","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_16","alias_value":"KZGL6CQJWFLXZEDB","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_8","alias_value":"KZGL6CQJ","created_at":"2026-05-18T12:33:33Z"}],"graph_snapshots":[{"event_id":"sha256:89e991b968c1a88a29cbad42af406fdf81fa2b05a8d23441bac24fe6d100ba6b","target":"graph","created_at":"2026-05-17T23:38:52Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"We demonstrate the efficacy of the FP8 format on a variety of image and language tasks, effectively matching the result quality achieved by 16-bit training sessions. Our study covers the main modern neural network architectures - CNNs, RNNs, and Transformer-based models, leaving all the hyperparameters unchanged from the 16-bit baseline training sessions. Our training experiments include large, up to 175B parameter, language models."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the chosen E4M3 and E5M2 encodings will preserve accuracy across all tasks and model scales without any hyperparameter retuning or task-specific adjustments."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"FP8 formats E4M3 and E5M2 match 16-bit training accuracy on CNNs, RNNs, and Transformers up to 175B parameters without hyperparameter changes."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"FP8 with E4M3 and E5M2 encodings matches 16-bit training accuracy on large language and image models without hyperparameter changes."}],"snapshot_sha256":"7bbd391a1fd0fbfb0c7a1b5b1343c74e6846133c3c379d3cf94113bd8804893f"},"formal_canon":{"evidence_count":1,"snapshot_sha256":"ce77c6a6004888e65152fb50e6bad0336fd153b773deac150eca9e7e05f3da7d"},"paper":{"abstract_excerpt":"FP8 is a natural progression for accelerating deep learning training inference beyond the 16-bit formats common in modern processors. In this paper we propose an 8-bit floating point (FP8) binary interchange format consisting of two encodings - E4M3 (4-bit exponent and 3-bit mantissa) and E5M2 (5-bit exponent and 2-bit mantissa). While E5M2 follows IEEE 754 conventions for representatio of special values, E4M3's dynamic range is extended by not representing infinities and having only one mantissa bit-pattern for NaNs. We demonstrate the efficacy of the FP8 format on a variety of image and lang","authors_text":"Alexander Heinecke, Dusan Stosic, Hao Wu, John Kamalu, Marius Cornea, Michael Siu, Mohammad Shoeybi, Naveen Mellempudi, Neil Burgess, Patrick Judd, Paulius Micikevicius, Pradeep Dubey, Richard Grisenthwaite, Sangwon Ha, Stuart Oberman","cross_cats":[],"headline":"FP8 with E4M3 and E5M2 encodings matches 16-bit training accuracy on large language and image models without hyperparameter changes.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2022-09-12T17:39:55Z","title":"FP8 Formats for Deep Learning"},"references":{"count":26,"internal_anchors":7,"resolved_work":26,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Michael J. Anderson, Benny Chen, Stephen Chen, Summer Deng, Jordan Fix, Michael Gschwind, Aravind Kalaiah, Changkyu Kim, Jaewon Lee, Jason Liang, Haixin Liu, Yinghai Lu, Jack Montgomery, Arun Moorthy,","work_id":"f1a0c3ad-b21b-473f-a72e-8f0ebd320c2f","year":2021},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Language models are few-shot learners","work_id":"ba7f3302-4a4e-4bc1-85c2-b07ce5a54ccb","year":1901},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Bfloat16 processing for neural networks","work_id":"b2b017b2-573e-4349-9022-7eb14d1228b7","year":2019},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, Parker Schuh, Kensen Shi, Sasha Tsvyashc","work_id":"81cf3668-3ff3-4308-a45a-5415987c08ac","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Binaryconnect: Training deep neural networks with binary weights during propagations","work_id":"07fa38a3-f9ab-416c-bfde-b731f877260c","year":2015}],"snapshot_sha256":"a1347b5bf9598f44b8373eaf98db0aff74d092b150a5f78d5dcdf82406aedd1c"},"source":{"id":"2209.05433","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-15T09:42:41.400605Z","id":"65622646-54db-4a02-a96e-a76631551664","model_set":{"reader":"grok-4.3"},"one_line_summary":"FP8 formats E4M3 and E5M2 match 16-bit training accuracy on CNNs, RNNs, and Transformers up to 175B parameters without hyperparameter changes.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"FP8 with E4M3 and E5M2 encodings matches 16-bit training accuracy on large language and image models without hyperparameter changes.","strongest_claim":"We demonstrate the efficacy of the FP8 format on a variety of image and language tasks, effectively matching the result quality achieved by 16-bit training sessions. Our study covers the main modern neural network architectures - CNNs, RNNs, and Transformer-based models, leaving all the hyperparameters unchanged from the 16-bit baseline training sessions. Our training experiments include large, up to 175B parameter, language models.","weakest_assumption":"That the chosen E4M3 and E5M2 encodings will preserve accuracy across all tasks and model scales without any hyperparameter retuning or task-specific adjustments."}},"verdict_id":"65622646-54db-4a02-a96e-a76631551664"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:d6ef12a8d32e031e2e88e76574957cf00708bcb9856305e977144338a84b2fb1","target":"record","created_at":"2026-05-17T23:38:52Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"c520978422b7a453110ed090b9d0c9b9786712d389b5a932b2e5184ded23748b","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2022-09-12T17:39:55Z","title_canon_sha256":"a787abcfcac8027bbbf57bf19b5a9d2496b30a55d097c56c0d787ecb293565af"},"schema_version":"1.0","source":{"id":"2209.05433","kind":"arxiv","version":2}},"canonical_sha256":"564cbf0a09b1577c90610ea47389f71f657f9815920159d0397f4be07f509302","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"564cbf0a09b1577c90610ea47389f71f657f9815920159d0397f4be07f509302","first_computed_at":"2026-05-17T23:38:52.855387Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:52.855387Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"4/CAvVbeeNXN3W4sPtyS2faUr8KF/6U2g7+BKxLwCn598CSbrjdZoE7pxFPajAfdR+K7h54qN70HErW9AVNLCg==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:52.855929Z","signed_message":"canonical_sha256_bytes"},"source_id":"2209.05433","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:d6ef12a8d32e031e2e88e76574957cf00708bcb9856305e977144338a84b2fb1","sha256:89e991b968c1a88a29cbad42af406fdf81fa2b05a8d23441bac24fe6d100ba6b"],"state_sha256":"eb4390d010f73f1de6285d6a1ee380a45df8be6d37176afa438d95eaf2b4c36d"}