{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:VQLUMH5CHINZYDROH5UJQCPR4K","short_pith_number":"pith:VQLUMH5C","schema_version":"1.0","canonical_sha256":"ac17461fa23a1b9c0e2e3f689809f1e2a17c80139d4aaccfd2438d0fea655b97","source":{"kind":"arxiv","id":"2405.02803","version":1},"attestation_state":"computed","paper":{"title":"Is Flash Attention Stable?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.DC"],"primary_cat":"cs.LG","authors_text":"Alicia Golden, Basil Hosmer, Bilge Acun, Carole-Jean Wu, David Brooks, Fei Sun, Gu-Yeon Wei, Jeff Johnson, Samuel Hsia, Yejin Lee, Zachary DeVito","submitted_at":"2024-05-05T03:25:25Z","abstract_excerpt":"Training large-scale machine learning models poses distinct system challenges, given both the size and complexity of today's workloads. Recently, many organizations training state-of-the-art Generative AI models have reported cases of instability during training, often taking the form of loss spikes. Numeric deviation has emerged as a potential cause of this training instability, although quantifying this is especially challenging given the costly nature of training runs. In this work, we develop a principled approach to understanding the effects of numeric deviation, and construct proxies to "},"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":"2405.02803","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-05-05T03:25:25Z","cross_cats_sorted":["cs.DC"],"title_canon_sha256":"dd347804ee438b44eaca5af1afce1dfff00a074bbff3d33b74820afdd4942343","abstract_canon_sha256":"2180d9772b3587805344c589ed7bee34757565dbc6c52e8987b689eb3a43d8b1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:15:54.059428Z","signature_b64":"KrD6QzxnCWXHcvcYJ9NKFPLKgLfuq68/uuhnRIidKoEA9zig8+Ydejnriuz9yCdPf8lM1WFHVNcNICJaRfERAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ac17461fa23a1b9c0e2e3f689809f1e2a17c80139d4aaccfd2438d0fea655b97","last_reissued_at":"2026-07-05T08:15:54.059018Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:15:54.059018Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Is Flash Attention Stable?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.DC"],"primary_cat":"cs.LG","authors_text":"Alicia Golden, Basil Hosmer, Bilge Acun, Carole-Jean Wu, David Brooks, Fei Sun, Gu-Yeon Wei, Jeff Johnson, Samuel Hsia, Yejin Lee, Zachary DeVito","submitted_at":"2024-05-05T03:25:25Z","abstract_excerpt":"Training large-scale machine learning models poses distinct system challenges, given both the size and complexity of today's workloads. Recently, many organizations training state-of-the-art Generative AI models have reported cases of instability during training, often taking the form of loss spikes. Numeric deviation has emerged as a potential cause of this training instability, although quantifying this is especially challenging given the costly nature of training runs. In this work, we develop a principled approach to understanding the effects of numeric deviation, and construct proxies to "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.02803","kind":"arxiv","version":1},"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/2405.02803/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":"2405.02803","created_at":"2026-07-05T08:15:54.059063+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.02803v1","created_at":"2026-07-05T08:15:54.059063+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.02803","created_at":"2026-07-05T08:15:54.059063+00:00"},{"alias_kind":"pith_short_12","alias_value":"VQLUMH5CHINZ","created_at":"2026-07-05T08:15:54.059063+00:00"},{"alias_kind":"pith_short_16","alias_value":"VQLUMH5CHINZYDRO","created_at":"2026-07-05T08:15:54.059063+00:00"},{"alias_kind":"pith_short_8","alias_value":"VQLUMH5C","created_at":"2026-07-05T08:15:54.059063+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.02119","citing_title":"An Efficient vLLM-Based Inference Pipeline for Unified Audio Understanding and Generation","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29979","citing_title":"Fingerprinting Inference Systems of Large Language Models","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2511.21016","citing_title":"Gated KalmaNet: A Fading Memory Layer Through Test-Time Ridge Regression","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2407.08608","citing_title":"FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2510.04212","citing_title":"Why Low-Precision Transformer Training Fails: An Analysis on Flash Attention","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2510.15596","citing_title":"PRISM: Probabilistic Runtime Insights and Scalable Performance Modeling for Large-Scale Distributed Training","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2404.14294","citing_title":"A Survey on Efficient Inference for Large Language Models","ref_index":28,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VQLUMH5CHINZYDROH5UJQCPR4K","json":"https://pith.science/pith/VQLUMH5CHINZYDROH5UJQCPR4K.json","graph_json":"https://pith.science/api/pith-number/VQLUMH5CHINZYDROH5UJQCPR4K/graph.json","events_json":"https://pith.science/api/pith-number/VQLUMH5CHINZYDROH5UJQCPR4K/events.json","paper":"https://pith.science/paper/VQLUMH5C"},"agent_actions":{"view_html":"https://pith.science/pith/VQLUMH5CHINZYDROH5UJQCPR4K","download_json":"https://pith.science/pith/VQLUMH5CHINZYDROH5UJQCPR4K.json","view_paper":"https://pith.science/paper/VQLUMH5C","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.02803&json=true","fetch_graph":"https://pith.science/api/pith-number/VQLUMH5CHINZYDROH5UJQCPR4K/graph.json","fetch_events":"https://pith.science/api/pith-number/VQLUMH5CHINZYDROH5UJQCPR4K/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VQLUMH5CHINZYDROH5UJQCPR4K/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VQLUMH5CHINZYDROH5UJQCPR4K/action/storage_attestation","attest_author":"https://pith.science/pith/VQLUMH5CHINZYDROH5UJQCPR4K/action/author_attestation","sign_citation":"https://pith.science/pith/VQLUMH5CHINZYDROH5UJQCPR4K/action/citation_signature","submit_replication":"https://pith.science/pith/VQLUMH5CHINZYDROH5UJQCPR4K/action/replication_record"}},"created_at":"2026-07-05T08:15:54.059063+00:00","updated_at":"2026-07-05T08:15:54.059063+00:00"}