{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:RL65NZWJHR4ABNW6F46LT3SRQN","short_pith_number":"pith:RL65NZWJ","schema_version":"1.0","canonical_sha256":"8afdd6e6c93c7800b6de2f3cb9ee518374d0ff2d5802b482fb97178cfb10888d","source":{"kind":"arxiv","id":"2502.05967","version":3},"attestation_state":"computed","paper":{"title":"$\\mu$nit Scaling: Simple and Scalable FP8 LLM Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Abhay Gupta, Davis Blalock, Mansheej Paul, Saaketh Narayan","submitted_at":"2025-02-09T17:31:09Z","abstract_excerpt":"Large Language Model training with 8-bit floating point (FP8) formats promises significant efficiency improvements, but reduced numerical precision makes training challenging. It is currently possible to train in FP8 only if one is willing to tune various hyperparameters, reduce model scale, or accept the overhead of computing dynamic scale factors. We demonstrate simple, scalable FP8 training that requires no dynamic scaling factors or special hyperparameters, even at large model sizes. Our method, $\\mu$nit Scaling ($\\mu$S), also enables simple hyperparameter transfer across model widths, mat"},"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":"2502.05967","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-09T17:31:09Z","cross_cats_sorted":[],"title_canon_sha256":"b9478ef27866e375475f698c343831c5f4dd3e41000183c57047d7345d48d5b5","abstract_canon_sha256":"2cdf0630959265057947340756351c3d11866b85ff124ecfa609ee92922363ad"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:16:05.997923Z","signature_b64":"NnxXvxO+M61qv2fe94Z3fIszUBagQwuO/AodBx3YbumE3qYWrM1+T8UUI55E/CY3ToSJ8RjmvEK2aeyQgVVIAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8afdd6e6c93c7800b6de2f3cb9ee518374d0ff2d5802b482fb97178cfb10888d","last_reissued_at":"2026-07-05T11:16:05.997419Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:16:05.997419Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"$\\mu$nit Scaling: Simple and Scalable FP8 LLM Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Abhay Gupta, Davis Blalock, Mansheej Paul, Saaketh Narayan","submitted_at":"2025-02-09T17:31:09Z","abstract_excerpt":"Large Language Model training with 8-bit floating point (FP8) formats promises significant efficiency improvements, but reduced numerical precision makes training challenging. It is currently possible to train in FP8 only if one is willing to tune various hyperparameters, reduce model scale, or accept the overhead of computing dynamic scale factors. We demonstrate simple, scalable FP8 training that requires no dynamic scaling factors or special hyperparameters, even at large model sizes. Our method, $\\mu$nit Scaling ($\\mu$S), also enables simple hyperparameter transfer across model widths, mat"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.05967","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":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2502.05967/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":"2502.05967","created_at":"2026-07-05T11:16:05.997474+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.05967v3","created_at":"2026-07-05T11:16:05.997474+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.05967","created_at":"2026-07-05T11:16:05.997474+00:00"},{"alias_kind":"pith_short_12","alias_value":"RL65NZWJHR4A","created_at":"2026-07-05T11:16:05.997474+00:00"},{"alias_kind":"pith_short_16","alias_value":"RL65NZWJHR4ABNW6","created_at":"2026-07-05T11:16:05.997474+00:00"},{"alias_kind":"pith_short_8","alias_value":"RL65NZWJ","created_at":"2026-07-05T11:16:05.997474+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.15290","citing_title":"GQA-{\\mu}P: The maximal parameterization update for grouped query attention","ref_index":14,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RL65NZWJHR4ABNW6F46LT3SRQN","json":"https://pith.science/pith/RL65NZWJHR4ABNW6F46LT3SRQN.json","graph_json":"https://pith.science/api/pith-number/RL65NZWJHR4ABNW6F46LT3SRQN/graph.json","events_json":"https://pith.science/api/pith-number/RL65NZWJHR4ABNW6F46LT3SRQN/events.json","paper":"https://pith.science/paper/RL65NZWJ"},"agent_actions":{"view_html":"https://pith.science/pith/RL65NZWJHR4ABNW6F46LT3SRQN","download_json":"https://pith.science/pith/RL65NZWJHR4ABNW6F46LT3SRQN.json","view_paper":"https://pith.science/paper/RL65NZWJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.05967&json=true","fetch_graph":"https://pith.science/api/pith-number/RL65NZWJHR4ABNW6F46LT3SRQN/graph.json","fetch_events":"https://pith.science/api/pith-number/RL65NZWJHR4ABNW6F46LT3SRQN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RL65NZWJHR4ABNW6F46LT3SRQN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RL65NZWJHR4ABNW6F46LT3SRQN/action/storage_attestation","attest_author":"https://pith.science/pith/RL65NZWJHR4ABNW6F46LT3SRQN/action/author_attestation","sign_citation":"https://pith.science/pith/RL65NZWJHR4ABNW6F46LT3SRQN/action/citation_signature","submit_replication":"https://pith.science/pith/RL65NZWJHR4ABNW6F46LT3SRQN/action/replication_record"}},"created_at":"2026-07-05T11:16:05.997474+00:00","updated_at":"2026-07-05T11:16:05.997474+00:00"}