{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:TIEJU5MXWZLTXUIPPM2SSKSNYE","short_pith_number":"pith:TIEJU5MX","schema_version":"1.0","canonical_sha256":"9a089a7597b6573bd10f7b35292a4dc13b6f21e759eb046c535a80357864f32e","source":{"kind":"arxiv","id":"1905.12322","version":3},"attestation_state":"computed","paper":{"title":"A Study of BFLOAT16 for Deep Learning Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Abhisek Kundu, Alexander Heinecke, Bharat Kaul, Dharma Teja Vooturi, Dheevatsa Mudigere, Dhiraj Kalamkar, Dipankar Das, Evangelos Georganas, Hector Yuen, Jianyu Huang, Jiyan Yang, Jongsoo Park, Kunal Banerjee, Misha Smelyanskiy, Nataraj Jammalamadaka, Naveen Mellempudi, Pradeep Dubey, Sasikanth Avancha, Sudarshan Srinivasan","submitted_at":"2019-05-29T10:50:32Z","abstract_excerpt":"This paper presents the first comprehensive empirical study demonstrating the efficacy of the Brain Floating Point (BFLOAT16) half-precision format for Deep Learning training across image classification, speech recognition, language modeling, generative networks and industrial recommendation systems. BFLOAT16 is attractive for Deep Learning training for two reasons: the range of values it can represent is the same as that of IEEE 754 floating-point format (FP32) and conversion to/from FP32 is simple. Maintaining the same range as FP32 is important to ensure that no hyper-parameter tuning is re"},"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":"1905.12322","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2019-05-29T10:50:32Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"fad78c35dc4141f5775940c5d608e21e17db6356fbd6f84596749156b2d76943","abstract_canon_sha256":"8ada7b6619e1ae3f6707d5d618f0f2d9c5f8ea70a80a702e1288234798a960f4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:43:27.573650Z","signature_b64":"xgGqG1sHTLxsP8GQHr3dqWS+OmFfCEeaXB5sYIqIg5haczNStK++HNzF/ixYB6zXfmly/olrGz8hS/5CmdghCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9a089a7597b6573bd10f7b35292a4dc13b6f21e759eb046c535a80357864f32e","last_reissued_at":"2026-05-17T23:43:27.573214Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:43:27.573214Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Study of BFLOAT16 for Deep Learning Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Abhisek Kundu, Alexander Heinecke, Bharat Kaul, Dharma Teja Vooturi, Dheevatsa Mudigere, Dhiraj Kalamkar, Dipankar Das, Evangelos Georganas, Hector Yuen, Jianyu Huang, Jiyan Yang, Jongsoo Park, Kunal Banerjee, Misha Smelyanskiy, Nataraj Jammalamadaka, Naveen Mellempudi, Pradeep Dubey, Sasikanth Avancha, Sudarshan Srinivasan","submitted_at":"2019-05-29T10:50:32Z","abstract_excerpt":"This paper presents the first comprehensive empirical study demonstrating the efficacy of the Brain Floating Point (BFLOAT16) half-precision format for Deep Learning training across image classification, speech recognition, language modeling, generative networks and industrial recommendation systems. BFLOAT16 is attractive for Deep Learning training for two reasons: the range of values it can represent is the same as that of IEEE 754 floating-point format (FP32) and conversion to/from FP32 is simple. Maintaining the same range as FP32 is important to ensure that no hyper-parameter tuning is re"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1905.12322","kind":"arxiv","version":3},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"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":"1905.12322","created_at":"2026-05-17T23:43:27.573281+00:00"},{"alias_kind":"arxiv_version","alias_value":"1905.12322v3","created_at":"2026-05-17T23:43:27.573281+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1905.12322","created_at":"2026-05-17T23:43:27.573281+00:00"},{"alias_kind":"pith_short_12","alias_value":"TIEJU5MXWZLT","created_at":"2026-05-18T12:33:27.125529+00:00"},{"alias_kind":"pith_short_16","alias_value":"TIEJU5MXWZLTXUIP","created_at":"2026-05-18T12:33:27.125529+00:00"},{"alias_kind":"pith_short_8","alias_value":"TIEJU5MX","created_at":"2026-05-18T12:33:27.125529+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":18,"internal_anchor_count":11,"sample":[{"citing_arxiv_id":"2601.21198","citing_title":"ZipMoE: Efficient On-Device MoE Serving via Lossless Compression and Cache-Affinity Scheduling","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2412.04468","citing_title":"NVILA: Efficient Frontier Visual Language Models","ref_index":32,"is_internal_anchor":true},{"citing_arxiv_id":"2605.19893","citing_title":"SSV: Sparse Speculative Verification for Efficient LLM Inference","ref_index":17,"is_internal_anchor":true},{"citing_arxiv_id":"2604.22282","citing_title":"STEM: Structure-Tracing Evidence Mining for Knowledge Graphs-Driven Retrieval-Augmented Generation","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2605.17064","citing_title":"Towards Human-Level Book-Writing Capability","ref_index":22,"is_internal_anchor":true},{"citing_arxiv_id":"2510.04212","citing_title":"Why Low-Precision Transformer Training Fails: An Analysis on Flash Attention","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2512.02010","citing_title":"Four Over Six: More Accurate NVFP4 Quantization with Adaptive Block Scaling","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2512.17738","citing_title":"When the Gold Standard Isn't Necessarily Standard: Challenges of Evaluating the Translation of User-Generated Content","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2603.14360","citing_title":"M$^2$RNN: Non-Linear RNNs with Matrix-Valued States for Scalable Language Modeling","ref_index":16,"is_internal_anchor":true},{"citing_arxiv_id":"2209.05433","citing_title":"FP8 Formats for Deep Learning","ref_index":9,"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":32,"is_internal_anchor":true},{"citing_arxiv_id":"2604.02651","citing_title":"Communication-free Sampling and 4D Hybrid Parallelism for Scalable Mini-batch GNN Training","ref_index":55,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11111","citing_title":"ShardTensor: Domain Parallelism for Scientific Machine Learning","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2604.27844","citing_title":"ZipCCL: Efficient Lossless Data Compression of Communication Collectives for Accelerating LLM Training","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2604.22282","citing_title":"STEM: Structure-Tracing Evidence Mining for Knowledge Graphs-Driven Retrieval-Augmented Generation","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00479","citing_title":"Reduced-Precision Stochastic Simulation for Mathematical Biology","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10494","citing_title":"From Characterization to Microarchitecture: Designing an Elegant and Reliable BFP-Based NPU","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10390","citing_title":"LLM-PRISM: Characterizing Silent Data Corruption from Permanent GPU Faults in LLM Training","ref_index":24,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TIEJU5MXWZLTXUIPPM2SSKSNYE","json":"https://pith.science/pith/TIEJU5MXWZLTXUIPPM2SSKSNYE.json","graph_json":"https://pith.science/api/pith-number/TIEJU5MXWZLTXUIPPM2SSKSNYE/graph.json","events_json":"https://pith.science/api/pith-number/TIEJU5MXWZLTXUIPPM2SSKSNYE/events.json","paper":"https://pith.science/paper/TIEJU5MX"},"agent_actions":{"view_html":"https://pith.science/pith/TIEJU5MXWZLTXUIPPM2SSKSNYE","download_json":"https://pith.science/pith/TIEJU5MXWZLTXUIPPM2SSKSNYE.json","view_paper":"https://pith.science/paper/TIEJU5MX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1905.12322&json=true","fetch_graph":"https://pith.science/api/pith-number/TIEJU5MXWZLTXUIPPM2SSKSNYE/graph.json","fetch_events":"https://pith.science/api/pith-number/TIEJU5MXWZLTXUIPPM2SSKSNYE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TIEJU5MXWZLTXUIPPM2SSKSNYE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TIEJU5MXWZLTXUIPPM2SSKSNYE/action/storage_attestation","attest_author":"https://pith.science/pith/TIEJU5MXWZLTXUIPPM2SSKSNYE/action/author_attestation","sign_citation":"https://pith.science/pith/TIEJU5MXWZLTXUIPPM2SSKSNYE/action/citation_signature","submit_replication":"https://pith.science/pith/TIEJU5MXWZLTXUIPPM2SSKSNYE/action/replication_record"}},"created_at":"2026-05-17T23:43:27.573281+00:00","updated_at":"2026-05-17T23:43:27.573281+00:00"}