{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:WV555LEBCM4LJXSSMJXAWTLOKI","short_pith_number":"pith:WV555LEB","schema_version":"1.0","canonical_sha256":"b57bdeac811338b4de52626e0b4d6e5216831a95de01bbc0c362c2bacacc0257","source":{"kind":"arxiv","id":"2303.17951","version":2},"attestation_state":"computed","paper":{"title":"FP8 versus INT8 for efficient deep learning inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Andrey Kuzmin, Chirag Patel, Eric Mahurin, Joseph Soriaga, Markus Nagel, Mart van Baalen, Sanghyuk Lee, Sundar Subramanian, Suparna S Nair, Tijmen Blankevoort, Yuwei Ren","submitted_at":"2023-03-31T10:29:17Z","abstract_excerpt":"Recently, the idea of using FP8 as a number format for neural network training has been floating around the deep learning world. Given that most training is currently conducted with entire networks in FP32, or sometimes FP16 with mixed-precision, the step to having some parts of a network run in FP8 with 8-bit weights is an appealing potential speed-up for the generally costly and time-intensive training procedures in deep learning. A natural question arises regarding what this development means for efficient inference on edge devices. In the efficient inference device world, workloads are fre"},"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":"2303.17951","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-03-31T10:29:17Z","cross_cats_sorted":[],"title_canon_sha256":"6b3e8d76a688e73e23609a82fef5bf12b3e9099a2d17058530eef36ac96dcc98","abstract_canon_sha256":"5f42dd16192463ec1dbb7b563b8ffd4a68c657b9ac63169d27c0e30c55f07996"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:20:42.335454Z","signature_b64":"zKqf+i/XDksTd4ee0+blKe3FzC+NDwYvBWADrCllQSqCDv7bJI+hMj5MQofZIQOU3k+npAnr3JUJQx9AIhhDCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b57bdeac811338b4de52626e0b4d6e5216831a95de01bbc0c362c2bacacc0257","last_reissued_at":"2026-07-05T06:20:42.334971Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:20:42.334971Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"FP8 versus INT8 for efficient deep learning inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Andrey Kuzmin, Chirag Patel, Eric Mahurin, Joseph Soriaga, Markus Nagel, Mart van Baalen, Sanghyuk Lee, Sundar Subramanian, Suparna S Nair, Tijmen Blankevoort, Yuwei Ren","submitted_at":"2023-03-31T10:29:17Z","abstract_excerpt":"Recently, the idea of using FP8 as a number format for neural network training has been floating around the deep learning world. Given that most training is currently conducted with entire networks in FP32, or sometimes FP16 with mixed-precision, the step to having some parts of a network run in FP8 with 8-bit weights is an appealing potential speed-up for the generally costly and time-intensive training procedures in deep learning. A natural question arises regarding what this development means for efficient inference on edge devices. In the efficient inference device world, workloads are fre"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2303.17951","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/2303.17951/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":"2303.17951","created_at":"2026-07-05T06:20:42.335028+00:00"},{"alias_kind":"arxiv_version","alias_value":"2303.17951v2","created_at":"2026-07-05T06:20:42.335028+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2303.17951","created_at":"2026-07-05T06:20:42.335028+00:00"},{"alias_kind":"pith_short_12","alias_value":"WV555LEBCM4L","created_at":"2026-07-05T06:20:42.335028+00:00"},{"alias_kind":"pith_short_16","alias_value":"WV555LEBCM4LJXSS","created_at":"2026-07-05T06:20:42.335028+00:00"},{"alias_kind":"pith_short_8","alias_value":"WV555LEB","created_at":"2026-07-05T06:20:42.335028+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.04115","citing_title":"dMX: Differentiable Mixed-Precision Assignment for Low-Precision Floating-Point Formats","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2506.11277","citing_title":"Analysis of Floating-Point Matrix Multiplication Computed via Integer Arithmetic","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2511.06838","citing_title":"P3-LLM: An Integrated NPU-PIM Accelerator for Edge LLM Inference Using Hybrid Numerical Formats","ref_index":70,"is_internal_anchor":false},{"citing_arxiv_id":"2602.10718","citing_title":"SnapMLA: Efficient Long-Context MLA Decoding via Hardware-Aware FP8 Quantized Pipelining","ref_index":39,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WV555LEBCM4LJXSSMJXAWTLOKI","json":"https://pith.science/pith/WV555LEBCM4LJXSSMJXAWTLOKI.json","graph_json":"https://pith.science/api/pith-number/WV555LEBCM4LJXSSMJXAWTLOKI/graph.json","events_json":"https://pith.science/api/pith-number/WV555LEBCM4LJXSSMJXAWTLOKI/events.json","paper":"https://pith.science/paper/WV555LEB"},"agent_actions":{"view_html":"https://pith.science/pith/WV555LEBCM4LJXSSMJXAWTLOKI","download_json":"https://pith.science/pith/WV555LEBCM4LJXSSMJXAWTLOKI.json","view_paper":"https://pith.science/paper/WV555LEB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2303.17951&json=true","fetch_graph":"https://pith.science/api/pith-number/WV555LEBCM4LJXSSMJXAWTLOKI/graph.json","fetch_events":"https://pith.science/api/pith-number/WV555LEBCM4LJXSSMJXAWTLOKI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WV555LEBCM4LJXSSMJXAWTLOKI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WV555LEBCM4LJXSSMJXAWTLOKI/action/storage_attestation","attest_author":"https://pith.science/pith/WV555LEBCM4LJXSSMJXAWTLOKI/action/author_attestation","sign_citation":"https://pith.science/pith/WV555LEBCM4LJXSSMJXAWTLOKI/action/citation_signature","submit_replication":"https://pith.science/pith/WV555LEBCM4LJXSSMJXAWTLOKI/action/replication_record"}},"created_at":"2026-07-05T06:20:42.335028+00:00","updated_at":"2026-07-05T06:20:42.335028+00:00"}