{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:MGKPCIXJYYEXICVREX4KUQJQBZ","short_pith_number":"pith:MGKPCIXJ","schema_version":"1.0","canonical_sha256":"6194f122e9c609740ab125f8aa41300e5f84bb0ea6d601c0a535b0844aae309b","source":{"kind":"arxiv","id":"2305.13245","version":3},"attestation_state":"computed","paper":{"title":"GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints","license":"http://creativecommons.org/licenses/by/4.0/","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.","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Federico Lebr\\'on, James Lee-Thorp, Joshua Ainslie, Michiel de Jong, Sumit Sanghai, Yury Zemlyanskiy","submitted_at":"2023-05-22T17:16:38Z","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"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2305.13245","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-05-22T17:16:38Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"7929c6e9d522dca9ceba7ad6e7fc9a8b3efad9d631da45c97403d77cf800d192","abstract_canon_sha256":"84953e0a13173d207eb3658876e04a86f84bea9bbd978969ec905a88c7e6e302"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:27:25.131183Z","signature_b64":"gUs4BA8LogM4x/sDMSJvDn3HWFLToltHCRXUO7iEMP1E94TfIJ90CehIhOxha/54WW1aRYw4nVZOyTDpuQnsCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6194f122e9c609740ab125f8aa41300e5f84bb0ea6d601c0a535b0844aae309b","last_reissued_at":"2026-07-05T07:27:25.130641Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:27:25.130641Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints","license":"http://creativecommons.org/licenses/by/4.0/","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.","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Federico Lebr\\'on, James Lee-Thorp, Joshua Ainslie, Michiel de Jong, Sumit Sanghai, Yury Zemlyanskiy","submitted_at":"2023-05-22T17:16:38Z","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"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We show that uptrained GQA achieves quality close to multi-head attention with comparable speed to MQA.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The uptraining recipe with 5% compute is sufficient to recover near-original quality without task-specific degradation or architecture-dependent failures.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","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.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","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.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"083e9c9899dea7d57941e9ab9c3d62927a7ba7c1505b4c5d2ac0c8c264197600"},"source":{"id":"2305.13245","kind":"arxiv","version":3},"verdict":{"id":"0a685e45-ceed-4ec1-98b4-cb5802fe461b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-11T06:48:00.303359Z","strongest_claim":"We show that uptrained GQA achieves quality close to multi-head attention with comparable speed to MQA.","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","weakest_assumption":"The uptraining recipe with 5% compute is sufficient to recover near-original quality without task-specific degradation or architecture-dependent failures.","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."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2305.13245/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":40,"sample":[{"doi":"","year":null,"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","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"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","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Alexey Romanov and Chaitanya Shivade","work_id":"b4ce3d71-9be2-4401-8618-528a87b188bd","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2015,"title":"Kingma and Jimmy Ba , editor =","work_id":"03bfe487-67cd-4832-be0e-f98027fded15","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Adafactor: Adaptive Learning Rates with Sublinear Memory Cost , booktitle =","work_id":"2835d2d4-5323-4fa8-a5df-250f93b86aee","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":40,"snapshot_sha256":"c71236cb8c84b6325f2f3c7c9ee93b91ca4873d7a633cd472e5edd15e77c805d","internal_anchors":11},"formal_canon":{"evidence_count":2,"snapshot_sha256":"934aee5a5c19d1e9498cbc8f71a6f6069daa79f5e1b5ace090291877a0775c6f"},"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":"2305.13245","created_at":"2026-07-05T07:27:25.130705+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.13245v3","created_at":"2026-07-05T07:27:25.130705+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.13245","created_at":"2026-07-05T07:27:25.130705+00:00"},{"alias_kind":"pith_short_12","alias_value":"MGKPCIXJYYEX","created_at":"2026-07-05T07:27:25.130705+00:00"},{"alias_kind":"pith_short_16","alias_value":"MGKPCIXJYYEXICVR","created_at":"2026-07-05T07:27:25.130705+00:00"},{"alias_kind":"pith_short_8","alias_value":"MGKPCIXJ","created_at":"2026-07-05T07:27:25.130705+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":136,"internal_anchor_count":136,"sample":[{"citing_arxiv_id":"2606.25971","citing_title":"Improving Neural Network Training by Decoupling the Magnitude and Direction of Weight Vectors","ref_index":169,"is_internal_anchor":true},{"citing_arxiv_id":"2606.25342","citing_title":"Lifelong In-Context Learning with Transformers Requires Parametric Forms of Attention","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2606.24320","citing_title":"ZONOS2 Technical Report","ref_index":172,"is_internal_anchor":true},{"citing_arxiv_id":"2606.24747","citing_title":"Scaling Laws for Task-Specific LLM Distillation","ref_index":34,"is_internal_anchor":true},{"citing_arxiv_id":"2606.20945","citing_title":"Grouped Query Experts: Mixture-of-Experts on GQA Self-Attention","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2605.24956","citing_title":"NITP: Next Implicit Token Prediction for LLM Pre-training","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2606.28565","citing_title":"KernelSight-LM: A Kernel-Level LLM Inference Simulator","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2606.14122","citing_title":"Beyond Perplexity: UTF-8 Validity in Byte-aware Language Models","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2606.11634","citing_title":"Architecture-Aware Reinforcement Learning Makes Sliding-Window Attention Competitive in Math Reasoning","ref_index":35,"is_internal_anchor":true},{"citing_arxiv_id":"2606.08382","citing_title":"STAR-KV: Low-Rank KV Cache Compression via Soft Thresholding for Adaptive Rank Control","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2606.15079","citing_title":"Ling and Ring 2.6 Technical Report: Efficient and Instant Agentic Intelligence at Trillion-Parameter Scale","ref_index":136,"is_internal_anchor":true},{"citing_arxiv_id":"2606.07703","citing_title":"How Much Dense Attention is Necessary? Oracle-Guided Sparse Prefill for Full/GQA Layers in Hybrid Long-Context Models","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2606.03979","citing_title":"Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories","ref_index":52,"is_internal_anchor":true},{"citing_arxiv_id":"2606.02964","citing_title":"Multi-Segment Attention: Enabling Efficient KV-Cache Management for Faster Large Language Model Serving","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2606.02780","citing_title":"Do Value Vectors in Deep Layers Need Context from the Residual Stream?","ref_index":61,"is_internal_anchor":true},{"citing_arxiv_id":"2606.01666","citing_title":"DOT-MoE: Differentiable Optimal Transport for MoEfication","ref_index":49,"is_internal_anchor":true},{"citing_arxiv_id":"2606.27449","citing_title":"Prism Transformer: Progressive Head Schedules for Hierarchical Attention Processing","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2606.24320","citing_title":"ZONOS2 Technical Report","ref_index":172,"is_internal_anchor":true},{"citing_arxiv_id":"2606.28565","citing_title":"KernelSight-LM: A Kernel-Level LLM Inference Simulator","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18359","citing_title":"RAVE: Re-Allocating Visual Attention in Large Multimodal Models","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2606.30709","citing_title":"Hierarchical Global Attention (HGA)","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2605.24956","citing_title":"NITP: Next Implicit Token Prediction for LLM Pre-training","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2606.28831","citing_title":"HARD-KV: Head-Adaptive Regularization for Decoding-time KV Compression","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2606.30634","citing_title":"One-Step Gradient Delay is Not a Barrier for Large-Scale Asynchronous Pipeline Parallel LLM Pretraining","ref_index":1,"is_internal_anchor":true},{"citing_arxiv_id":"2605.25966","citing_title":"Mapping the Schedule x Bit-Width Boundary in Sub-100M Quantisation-Aware Training","ref_index":1,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MGKPCIXJYYEXICVREX4KUQJQBZ","json":"https://pith.science/pith/MGKPCIXJYYEXICVREX4KUQJQBZ.json","graph_json":"https://pith.science/api/pith-number/MGKPCIXJYYEXICVREX4KUQJQBZ/graph.json","events_json":"https://pith.science/api/pith-number/MGKPCIXJYYEXICVREX4KUQJQBZ/events.json","paper":"https://pith.science/paper/MGKPCIXJ"},"agent_actions":{"view_html":"https://pith.science/pith/MGKPCIXJYYEXICVREX4KUQJQBZ","download_json":"https://pith.science/pith/MGKPCIXJYYEXICVREX4KUQJQBZ.json","view_paper":"https://pith.science/paper/MGKPCIXJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.13245&json=true","fetch_graph":"https://pith.science/api/pith-number/MGKPCIXJYYEXICVREX4KUQJQBZ/graph.json","fetch_events":"https://pith.science/api/pith-number/MGKPCIXJYYEXICVREX4KUQJQBZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MGKPCIXJYYEXICVREX4KUQJQBZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MGKPCIXJYYEXICVREX4KUQJQBZ/action/storage_attestation","attest_author":"https://pith.science/pith/MGKPCIXJYYEXICVREX4KUQJQBZ/action/author_attestation","sign_citation":"https://pith.science/pith/MGKPCIXJYYEXICVREX4KUQJQBZ/action/citation_signature","submit_replication":"https://pith.science/pith/MGKPCIXJYYEXICVREX4KUQJQBZ/action/replication_record"}},"created_at":"2026-07-05T07:27:25.130705+00:00","updated_at":"2026-07-05T07:27:25.130705+00:00"}