{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:FVEV65XEOH3E4ERP6AEGSX2F6S","short_pith_number":"pith:FVEV65XE","schema_version":"1.0","canonical_sha256":"2d495f76e471f64e122ff008695f45f491d7b793e1adf5a73d6cb420515c4b05","source":{"kind":"arxiv","id":"2407.09105","version":6},"attestation_state":"computed","paper":{"title":"Enhancing Training Efficiency Using Packing with Flash Attention","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Achintya Kundu, Laura Wynter, Mayank Mishra, Raghu Kiran Ganti, Rhui Dih Lee","submitted_at":"2024-07-12T09:10:37Z","abstract_excerpt":"Padding is often used in tuning LLM models by adding special tokens to shorter training examples to match the length of the longest sequence in each batch. While this ensures uniformity for batch processing, it introduces inefficiencies by including irrelevant padding tokens in the computation and wastes GPU resources. Hugging Face SFT trainer has always offered the option to use packing to combine multiple training examples, allowing for maximal utilization of GPU resources. However, up till now, it did not offer proper masking of each packed training example. This capability has been added t"},"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":"2407.09105","kind":"arxiv","version":6},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2024-07-12T09:10:37Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"a4532839004e06e4e251a629a671136bdba0a730ef00cffbc4e5457b2bab4228","abstract_canon_sha256":"191de8e3b646d50cd820a5ed3ff5e28104152b52d463ce3e7556a1a0d7c09418"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:01:36.147846Z","signature_b64":"SgwIvNIp5CcawLX1BsbmwBA8jpYQf28u6wAc296s7dK5c5Mjv5Ot9ub4ZPKFTycIUKrWGEYh2xFzeqnZwd92Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2d495f76e471f64e122ff008695f45f491d7b793e1adf5a73d6cb420515c4b05","last_reissued_at":"2026-07-05T09:01:36.147396Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:01:36.147396Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Enhancing Training Efficiency Using Packing with Flash Attention","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Achintya Kundu, Laura Wynter, Mayank Mishra, Raghu Kiran Ganti, Rhui Dih Lee","submitted_at":"2024-07-12T09:10:37Z","abstract_excerpt":"Padding is often used in tuning LLM models by adding special tokens to shorter training examples to match the length of the longest sequence in each batch. While this ensures uniformity for batch processing, it introduces inefficiencies by including irrelevant padding tokens in the computation and wastes GPU resources. Hugging Face SFT trainer has always offered the option to use packing to combine multiple training examples, allowing for maximal utilization of GPU resources. However, up till now, it did not offer proper masking of each packed training example. This capability has been added t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.09105","kind":"arxiv","version":6},"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/2407.09105/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":"2407.09105","created_at":"2026-07-05T09:01:36.147449+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.09105v6","created_at":"2026-07-05T09:01:36.147449+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.09105","created_at":"2026-07-05T09:01:36.147449+00:00"},{"alias_kind":"pith_short_12","alias_value":"FVEV65XEOH3E","created_at":"2026-07-05T09:01:36.147449+00:00"},{"alias_kind":"pith_short_16","alias_value":"FVEV65XEOH3E4ERP","created_at":"2026-07-05T09:01:36.147449+00:00"},{"alias_kind":"pith_short_8","alias_value":"FVEV65XE","created_at":"2026-07-05T09:01:36.147449+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.08476","citing_title":"FlashCP: Load-Balanced Communication-Efficient Context Parallelism for LLM Training","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2505.13211","citing_title":"MAGI-1: Autoregressive Video Generation at Scale","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FVEV65XEOH3E4ERP6AEGSX2F6S","json":"https://pith.science/pith/FVEV65XEOH3E4ERP6AEGSX2F6S.json","graph_json":"https://pith.science/api/pith-number/FVEV65XEOH3E4ERP6AEGSX2F6S/graph.json","events_json":"https://pith.science/api/pith-number/FVEV65XEOH3E4ERP6AEGSX2F6S/events.json","paper":"https://pith.science/paper/FVEV65XE"},"agent_actions":{"view_html":"https://pith.science/pith/FVEV65XEOH3E4ERP6AEGSX2F6S","download_json":"https://pith.science/pith/FVEV65XEOH3E4ERP6AEGSX2F6S.json","view_paper":"https://pith.science/paper/FVEV65XE","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.09105&json=true","fetch_graph":"https://pith.science/api/pith-number/FVEV65XEOH3E4ERP6AEGSX2F6S/graph.json","fetch_events":"https://pith.science/api/pith-number/FVEV65XEOH3E4ERP6AEGSX2F6S/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FVEV65XEOH3E4ERP6AEGSX2F6S/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FVEV65XEOH3E4ERP6AEGSX2F6S/action/storage_attestation","attest_author":"https://pith.science/pith/FVEV65XEOH3E4ERP6AEGSX2F6S/action/author_attestation","sign_citation":"https://pith.science/pith/FVEV65XEOH3E4ERP6AEGSX2F6S/action/citation_signature","submit_replication":"https://pith.science/pith/FVEV65XEOH3E4ERP6AEGSX2F6S/action/replication_record"}},"created_at":"2026-07-05T09:01:36.147449+00:00","updated_at":"2026-07-05T09:01:36.147449+00:00"}