{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:KZGL6CQJWFLXZEDBB2SHHCPXD5","short_pith_number":"pith:KZGL6CQJ","schema_version":"1.0","canonical_sha256":"564cbf0a09b1577c90610ea47389f71f657f9815920159d0397f4be07f509302","source":{"kind":"arxiv","id":"2209.05433","version":2},"attestation_state":"computed","paper":{"title":"FP8 Formats for Deep Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"FP8 with E4M3 and E5M2 encodings matches 16-bit training accuracy on large language and image models without hyperparameter changes.","cross_cats":[],"primary_cat":"cs.LG","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","submitted_at":"2022-09-12T17:39:55Z","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"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2209.05433","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2022-09-12T17:39:55Z","cross_cats_sorted":[],"title_canon_sha256":"a787abcfcac8027bbbf57bf19b5a9d2496b30a55d097c56c0d787ecb293565af","abstract_canon_sha256":"c520978422b7a453110ed090b9d0c9b9786712d389b5a932b2e5184ded23748b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:52.855929Z","signature_b64":"4/CAvVbeeNXN3W4sPtyS2faUr8KF/6U2g7+BKxLwCn598CSbrjdZoE7pxFPajAfdR+K7h54qN70HErW9AVNLCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"564cbf0a09b1577c90610ea47389f71f657f9815920159d0397f4be07f509302","last_reissued_at":"2026-05-17T23:38:52.855387Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:52.855387Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"FP8 Formats for Deep Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"FP8 with E4M3 and E5M2 encodings matches 16-bit training accuracy on large language and image models without hyperparameter changes.","cross_cats":[],"primary_cat":"cs.LG","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","submitted_at":"2022-09-12T17:39:55Z","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"},"claims":{"count":4,"items":[{"kind":"strongest_claim","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.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","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.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"FP8 formats E4M3 and E5M2 match 16-bit training accuracy on CNNs, RNNs, and Transformers up to 175B parameters without hyperparameter changes.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"FP8 with E4M3 and E5M2 encodings matches 16-bit training accuracy on large language and image models without hyperparameter changes.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"7bbd391a1fd0fbfb0c7a1b5b1343c74e6846133c3c379d3cf94113bd8804893f"},"source":{"id":"2209.05433","kind":"arxiv","version":2},"verdict":{"id":"65622646-54db-4a02-a96e-a76631551664","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T09:42:41.400605Z","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.","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","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.","pith_extraction_headline":"FP8 with E4M3 and E5M2 encodings matches 16-bit training accuracy on large language and image models without hyperparameter changes."},"references":{"count":26,"sample":[{"doi":"","year":2021,"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","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1901,"title":"Language models are few-shot learners","work_id":"ba7f3302-4a4e-4bc1-85c2-b07ce5a54ccb","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Bfloat16 processing for neural networks","work_id":"b2b017b2-573e-4349-9022-7eb14d1228b7","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"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","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2015,"title":"Binaryconnect: Training deep neural networks with binary weights during propagations","work_id":"07fa38a3-f9ab-416c-bfde-b731f877260c","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":26,"snapshot_sha256":"a1347b5bf9598f44b8373eaf98db0aff74d092b150a5f78d5dcdf82406aedd1c","internal_anchors":7},"formal_canon":{"evidence_count":1,"snapshot_sha256":"ce77c6a6004888e65152fb50e6bad0336fd153b773deac150eca9e7e05f3da7d"},"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":"2209.05433","created_at":"2026-05-17T23:38:52.855477+00:00"},{"alias_kind":"arxiv_version","alias_value":"2209.05433v2","created_at":"2026-05-17T23:38:52.855477+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2209.05433","created_at":"2026-05-17T23:38:52.855477+00:00"},{"alias_kind":"pith_short_12","alias_value":"KZGL6CQJWFLX","created_at":"2026-05-18T12:33:33.725879+00:00"},{"alias_kind":"pith_short_16","alias_value":"KZGL6CQJWFLXZEDB","created_at":"2026-05-18T12:33:33.725879+00:00"},{"alias_kind":"pith_short_8","alias_value":"KZGL6CQJ","created_at":"2026-05-18T12:33:33.725879+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":39,"internal_anchor_count":39,"sample":[{"citing_arxiv_id":"2605.20402","citing_title":"Decomposing MXFP4 quantization error for LLM reinforcement learning: reducible bias, recoverable deadzone, and an irreducible floor","ref_index":28,"is_internal_anchor":true},{"citing_arxiv_id":"2605.23656","citing_title":"Recursive Block-Diagonal Coupling for Resource-Efficient Training of Vision Models","ref_index":23,"is_internal_anchor":true},{"citing_arxiv_id":"2412.04468","citing_title":"NVILA: Efficient Frontier Visual Language Models","ref_index":87,"is_internal_anchor":true},{"citing_arxiv_id":"2602.05743","citing_title":"Balancing FP8 Computation Accuracy and Efficiency on Digital CIM via Shift-Aware On-the-fly Aligned-Mantissa Bitwidth Prediction","ref_index":4,"is_internal_anchor":true},{"citing_arxiv_id":"2605.20402","citing_title":"Decomposing MXFP4 quantization error for LLM reinforcement learning: reducible bias, recoverable deadzone, and an irreducible floor","ref_index":28,"is_internal_anchor":true},{"citing_arxiv_id":"2407.08608","citing_title":"FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision","ref_index":37,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18733","citing_title":"Advancing Narrative Long Video Generation via Training-Free Identity-Aware Memory","ref_index":29,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18739","citing_title":"LongLive-2.0: An NVFP4 Parallel Infrastructure for Long Video Generation","ref_index":51,"is_internal_anchor":true},{"citing_arxiv_id":"2605.17745","citing_title":"StatQAT: Statistical Quantizer Optimization for Deep Networks","ref_index":28,"is_internal_anchor":true},{"citing_arxiv_id":"2506.11277","citing_title":"Analysis of Floating-Point Matrix Multiplication Computed via Integer Arithmetic","ref_index":32,"is_internal_anchor":true},{"citing_arxiv_id":"2508.08822","citing_title":"OISMA: On-the-fly In-memory Stochastic Multiplication Architecture for Matrix-Multiplication Workloads","ref_index":26,"is_internal_anchor":true},{"citing_arxiv_id":"2509.03472","citing_title":"DPQuant: Efficient and Differentially-Private Model Training via Dynamic Quantization Scheduling","ref_index":32,"is_internal_anchor":true},{"citing_arxiv_id":"2510.04212","citing_title":"Why Low-Precision Transformer Training Fails: An Analysis on Flash Attention","ref_index":16,"is_internal_anchor":true},{"citing_arxiv_id":"2512.20856","citing_title":"NVIDIA Nemotron 3: Efficient and Open Intelligence","ref_index":31,"is_internal_anchor":true},{"citing_arxiv_id":"2511.06838","citing_title":"P3-LLM: An Integrated NPU-PIM Accelerator for Edge LLM Inference Using Hybrid Numerical Formats","ref_index":57,"is_internal_anchor":true},{"citing_arxiv_id":"2511.10909","citing_title":"Bit-Accurate Modeling of GPU Matrix Multiply-Accumulate Units: Demystifying Numerical Discrepancy and Accuracy","ref_index":20,"is_internal_anchor":true},{"citing_arxiv_id":"2512.12131","citing_title":"BOOST: BOttleneck-Optimized Scalable Training Framework for Low-Rank Large Language Models","ref_index":16,"is_internal_anchor":true},{"citing_arxiv_id":"2602.10718","citing_title":"SnapMLA: Efficient Long-Context MLA Decoding via Hardware-Aware FP8 Quantized Pipelining","ref_index":30,"is_internal_anchor":true},{"citing_arxiv_id":"2605.13907","citing_title":"AIS: Adaptive Importance Sampling for Quantized RL","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2604.03279","citing_title":"Rewriting TTS Inference Economics: Lightning V2 on Tenstorrent Achieves 4x Lower Cost Than NVIDIA L40S","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2604.03298","citing_title":"ENEC: A Lossless AI Model Compression Method Enabling Fast Inference on Ascend NPUs","ref_index":44,"is_internal_anchor":true},{"citing_arxiv_id":"2604.02525","citing_title":"AdaHOP: Fast and Accurate Low-Precision Training via Outlier-Pattern-Aware Rotation","ref_index":27,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06057","citing_title":"FalconGEMM: Surpassing Hardware Peaks with Lower-Complexity Matrix Multiplication","ref_index":29,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12464","citing_title":"Search Your Block Floating Point Scales!","ref_index":144,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11999","citing_title":"The Illusion of Power Capping in LLM Decode: A Phase-Aware Energy Characterisation Across Attention Architectures","ref_index":13,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":1,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KZGL6CQJWFLXZEDBB2SHHCPXD5","json":"https://pith.science/pith/KZGL6CQJWFLXZEDBB2SHHCPXD5.json","graph_json":"https://pith.science/api/pith-number/KZGL6CQJWFLXZEDBB2SHHCPXD5/graph.json","events_json":"https://pith.science/api/pith-number/KZGL6CQJWFLXZEDBB2SHHCPXD5/events.json","paper":"https://pith.science/paper/KZGL6CQJ"},"agent_actions":{"view_html":"https://pith.science/pith/KZGL6CQJWFLXZEDBB2SHHCPXD5","download_json":"https://pith.science/pith/KZGL6CQJWFLXZEDBB2SHHCPXD5.json","view_paper":"https://pith.science/paper/KZGL6CQJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2209.05433&json=true","fetch_graph":"https://pith.science/api/pith-number/KZGL6CQJWFLXZEDBB2SHHCPXD5/graph.json","fetch_events":"https://pith.science/api/pith-number/KZGL6CQJWFLXZEDBB2SHHCPXD5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KZGL6CQJWFLXZEDBB2SHHCPXD5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KZGL6CQJWFLXZEDBB2SHHCPXD5/action/storage_attestation","attest_author":"https://pith.science/pith/KZGL6CQJWFLXZEDBB2SHHCPXD5/action/author_attestation","sign_citation":"https://pith.science/pith/KZGL6CQJWFLXZEDBB2SHHCPXD5/action/citation_signature","submit_replication":"https://pith.science/pith/KZGL6CQJWFLXZEDBB2SHHCPXD5/action/replication_record"}},"created_at":"2026-05-17T23:38:52.855477+00:00","updated_at":"2026-05-17T23:38:52.855477+00:00"}