{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2023:MGKPCIXJYYEXICVREX4KUQJQBZ","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"84953e0a13173d207eb3658876e04a86f84bea9bbd978969ec905a88c7e6e302","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-05-22T17:16:38Z","title_canon_sha256":"7929c6e9d522dca9ceba7ad6e7fc9a8b3efad9d631da45c97403d77cf800d192"},"schema_version":"1.0","source":{"id":"2305.13245","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2305.13245","created_at":"2026-07-05T07:27:25Z"},{"alias_kind":"arxiv_version","alias_value":"2305.13245v3","created_at":"2026-07-05T07:27:25Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.13245","created_at":"2026-07-05T07:27:25Z"},{"alias_kind":"pith_short_12","alias_value":"MGKPCIXJYYEX","created_at":"2026-07-05T07:27:25Z"},{"alias_kind":"pith_short_16","alias_value":"MGKPCIXJYYEXICVR","created_at":"2026-07-05T07:27:25Z"},{"alias_kind":"pith_short_8","alias_value":"MGKPCIXJ","created_at":"2026-07-05T07:27:25Z"}],"graph_snapshots":[{"event_id":"sha256:265984860f1afb23c88a2bab95f5ef6221d6d535a26fdf3fa29bb4d7d83b2855","target":"graph","created_at":"2026-07-05T07:27:25Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"We show that uptrained GQA achieves quality close to multi-head attention with comparable speed to MQA."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The uptraining recipe with 5% compute is sufficient to recover near-original quality without task-specific degradation or architecture-dependent failures."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Uptraining multi-head transformer checkpoints to grouped-query attention models achieves near multi-head quality at multi-query inference speeds using 5% additional compute."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Uptraining multi-head attention checkpoints to grouped-query attention recovers near-original quality with only 5% additional compute and achieves multi-query inference speeds."}],"snapshot_sha256":"083e9c9899dea7d57941e9ab9c3d62927a7ba7c1505b4c5d2ac0c8c264197600"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"934aee5a5c19d1e9498cbc8f71a6f6069daa79f5e1b5ace090291877a0775c6f"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2305.13245/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Multi-query attention (MQA), which only uses a single key-value head, drastically speeds up decoder inference. However, MQA can lead to quality degradation, and moreover it may not be desirable to train a separate model just for faster inference. We (1) propose a recipe for uptraining existing multi-head language model checkpoints into models with MQA using 5% of original pre-training compute, and (2) introduce grouped-query attention (GQA), a generalization of multi-query attention which uses an intermediate (more than one, less than number of query heads) number of key-value heads. We show t","authors_text":"Federico Lebr\\'on, James Lee-Thorp, Joshua Ainslie, Michiel de Jong, Sumit Sanghai, Yury Zemlyanskiy","cross_cats":["cs.LG"],"headline":"Uptraining multi-head attention checkpoints to grouped-query attention recovers near-original quality with only 5% additional compute and achieves multi-query inference speeds.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-05-22T17:16:38Z","title":"GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints"},"references":{"count":40,"internal_anchors":11,"resolved_work":40,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"James Bradbury and Roy Frostig and Peter Hawkins and Matthew James Johnson and Chris Leary and Dougal Maclaurin and George Necula and Adam Paszke and Jake Vander","work_id":"f722cabc-fda6-4889-82c5-313113edc5fb","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Jonathan Heek and Anselm Levskaya and Avital Oliver and Marvin Ritter and Bertrand Rondepierre and Andreas Steiner and Marc van","work_id":"099845dd-6d9a-4e0c-a9f2-0f6843660d5f","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Alexey Romanov and Chaitanya Shivade","work_id":"b4ce3d71-9be2-4401-8618-528a87b188bd","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Kingma and Jimmy Ba , editor =","work_id":"03bfe487-67cd-4832-be0e-f98027fded15","year":2015},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Adafactor: Adaptive Learning Rates with Sublinear Memory Cost , booktitle =","work_id":"2835d2d4-5323-4fa8-a5df-250f93b86aee","year":2018}],"snapshot_sha256":"c71236cb8c84b6325f2f3c7c9ee93b91ca4873d7a633cd472e5edd15e77c805d"},"source":{"id":"2305.13245","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-11T06:48:00.303359Z","id":"0a685e45-ceed-4ec1-98b4-cb5802fe461b","model_set":{"reader":"grok-4.3"},"one_line_summary":"Uptraining multi-head transformer checkpoints to grouped-query attention models achieves near multi-head quality at multi-query inference speeds using 5% additional compute.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Uptraining multi-head attention checkpoints to grouped-query attention recovers near-original quality with only 5% additional compute and achieves multi-query inference speeds.","strongest_claim":"We show that uptrained GQA achieves quality close to multi-head attention with comparable speed to MQA.","weakest_assumption":"The uptraining recipe with 5% compute is sufficient to recover near-original quality without task-specific degradation or architecture-dependent failures."}},"verdict_id":"0a685e45-ceed-4ec1-98b4-cb5802fe461b"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:cfb70d005df740c855307bf156c56b71af813198624727a4310f35ec6af143a6","target":"record","created_at":"2026-07-05T07:27:25Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"84953e0a13173d207eb3658876e04a86f84bea9bbd978969ec905a88c7e6e302","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-05-22T17:16:38Z","title_canon_sha256":"7929c6e9d522dca9ceba7ad6e7fc9a8b3efad9d631da45c97403d77cf800d192"},"schema_version":"1.0","source":{"id":"2305.13245","kind":"arxiv","version":3}},"canonical_sha256":"6194f122e9c609740ab125f8aa41300e5f84bb0ea6d601c0a535b0844aae309b","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"6194f122e9c609740ab125f8aa41300e5f84bb0ea6d601c0a535b0844aae309b","first_computed_at":"2026-07-05T07:27:25.130641Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T07:27:25.130641Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"gUs4BA8LogM4x/sDMSJvDn3HWFLToltHCRXUO7iEMP1E94TfIJ90CehIhOxha/54WW1aRYw4nVZOyTDpuQnsCQ==","signature_status":"signed_v1","signed_at":"2026-07-05T07:27:25.131183Z","signed_message":"canonical_sha256_bytes"},"source_id":"2305.13245","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:cfb70d005df740c855307bf156c56b71af813198624727a4310f35ec6af143a6","sha256:265984860f1afb23c88a2bab95f5ef6221d6d535a26fdf3fa29bb4d7d83b2855"],"state_sha256":"ed778b7f4eb17d561f213a289227225703f681e225f28dbb076cf36ae5b1097b"}