{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:C4EW5CF5ETZB7Y2CQ4JOCWDXXG","short_pith_number":"pith:C4EW5CF5","schema_version":"1.0","canonical_sha256":"17096e88bd24f21fe3428712e15877b9989ea67ba7b88a5a35cb99e74b6d2990","source":{"kind":"arxiv","id":"2403.03507","version":2},"attestation_state":"computed","paper":{"title":"GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection","license":"http://creativecommons.org/licenses/by/4.0/","headline":"GaLore projects full gradients onto low-rank subspaces periodically, cutting optimizer memory by 65.5% while training every parameter of large language models.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Anima Anandkumar, Beidi Chen, Jiawei Zhao, Yuandong Tian, Zhangyang Wang, Zhenyu Zhang","submitted_at":"2024-03-06T07:29:57Z","abstract_excerpt":"Training Large Language Models (LLMs) presents significant memory challenges, predominantly due to the growing size of weights and optimizer states. Common memory-reduction approaches, such as low-rank adaptation (LoRA), add a trainable low-rank matrix to the frozen pre-trained weight in each layer, reducing trainable parameters and optimizer states. However, such approaches typically underperform training with full-rank weights in both pre-training and fine-tuning stages since they limit the parameter search to a low-rank subspace and alter the training dynamics, and further, may require full"},"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":"2403.03507","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-03-06T07:29:57Z","cross_cats_sorted":[],"title_canon_sha256":"b3d8e896d144add070f5ec4b0c898932d6e3a058b1cd57804eba277c6a3cfaa2","abstract_canon_sha256":"7b1eac0584b3aab5267eb2f6d6033547186f3b3a34a6f003bdeae70ede5b6275"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:46.187232Z","signature_b64":"mK00rJkKhh0MOWufNa/2sGbH5xL3iYsVTptZYaMVu14JSSD5ykvnpy0iXmGh/CJtLr91QFQMUSV1mEUVhkUBDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"17096e88bd24f21fe3428712e15877b9989ea67ba7b88a5a35cb99e74b6d2990","last_reissued_at":"2026-05-17T23:38:46.186714Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:46.186714Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection","license":"http://creativecommons.org/licenses/by/4.0/","headline":"GaLore projects full gradients onto low-rank subspaces periodically, cutting optimizer memory by 65.5% while training every parameter of large language models.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Anima Anandkumar, Beidi Chen, Jiawei Zhao, Yuandong Tian, Zhangyang Wang, Zhenyu Zhang","submitted_at":"2024-03-06T07:29:57Z","abstract_excerpt":"Training Large Language Models (LLMs) presents significant memory challenges, predominantly due to the growing size of weights and optimizer states. Common memory-reduction approaches, such as low-rank adaptation (LoRA), add a trainable low-rank matrix to the frozen pre-trained weight in each layer, reducing trainable parameters and optimizer states. However, such approaches typically underperform training with full-rank weights in both pre-training and fine-tuning stages since they limit the parameter search to a low-rank subspace and alter the training dynamics, and further, may require full"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We demonstrate, for the first time, the feasibility of pre-training a 7B model on consumer GPUs with 24GB memory (e.g., NVIDIA RTX 4090) without model parallel, checkpointing, or offloading strategies.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That periodically recomputed low-rank bases for gradient projection preserve optimization dynamics close enough to full-rank gradients that final model quality remains comparable across pre-training and fine-tuning regimes.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GaLore performs full-parameter LLM training with up to 65.5% less optimizer memory by projecting gradients onto a low-rank subspace at each step, matching full-rank performance on LLaMA pre-training and RoBERTa fine-tuning.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"GaLore projects full gradients onto low-rank subspaces periodically, cutting optimizer memory by 65.5% while training every parameter of large language models.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"716137a95d8b5a93ac44d1d24eb53cd4af93ab812ef2053f7d65aaa71f64fa98"},"source":{"id":"2403.03507","kind":"arxiv","version":2},"verdict":{"id":"8a22c18f-eb03-4180-bb8b-a3ccd8e03981","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T23:47:37.157229Z","strongest_claim":"We demonstrate, for the first time, the feasibility of pre-training a 7B model on consumer GPUs with 24GB memory (e.g., NVIDIA RTX 4090) without model parallel, checkpointing, or offloading strategies.","one_line_summary":"GaLore performs full-parameter LLM training with up to 65.5% less optimizer memory by projecting gradients onto a low-rank subspace at each step, matching full-rank performance on LLaMA pre-training and RoBERTa fine-tuning.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That periodically recomputed low-rank bases for gradient projection preserve optimization dynamics close enough to full-rank gradients that final model quality remains comparable across pre-training and fine-tuning regimes.","pith_extraction_headline":"GaLore projects full gradients onto low-rank subspaces periodically, cutting optimizer memory by 65.5% while training every parameter of large language models."},"references":{"count":55,"sample":[{"doi":"","year":2019,"title":"Memory efficient adaptive optimization","work_id":"852fa34d-c73a-4d07-8012-d983fe698557","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Belle: Be everyone's large language model engine","work_id":"84a8bc20-e93c-4601-8413-3f9c43212cae","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Continual learning in low-rank orthogonal subspaces","work_id":"c83f5d43-b1ee-4cf1-8c48-0c547920edd0","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Non- Convex Projected Gradient Descent for Generalized Low-Rank Tensor Regression","work_id":"ab8e2819-94e0-4e6d-bcfd-75ab25d5680b","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"Training Deep Nets with Sublinear Memory Cost","work_id":"f2c5c287-a500-40e4-a136-e7e3172db1d7","ref_index":5,"cited_arxiv_id":"1604.06174","is_internal_anchor":true}],"resolved_work":55,"snapshot_sha256":"fb16556210ca767e58199a9a7a212f2ab5f7dace7778461b526f01ce8ca68a16","internal_anchors":9},"formal_canon":{"evidence_count":2,"snapshot_sha256":"27962be8dc6346f405f988acb1bfa29e625c6b7e11ae694deef8035f87cf45c8"},"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":"2403.03507","created_at":"2026-05-17T23:38:46.186806+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.03507v2","created_at":"2026-05-17T23:38:46.186806+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.03507","created_at":"2026-05-17T23:38:46.186806+00:00"},{"alias_kind":"pith_short_12","alias_value":"C4EW5CF5ETZB","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"C4EW5CF5ETZB7Y2C","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"C4EW5CF5","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":29,"internal_anchor_count":29,"sample":[{"citing_arxiv_id":"2603.10067","citing_title":"HTMuon: Improving Muon via Heavy-Tailed Spectral Correction","ref_index":34,"is_internal_anchor":true},{"citing_arxiv_id":"2404.09005","citing_title":"Proof-of-Learning with Incentive Security","ref_index":70,"is_internal_anchor":true},{"citing_arxiv_id":"2501.07237","citing_title":"GWT: Scalable Optimizer State Compression for Large Language Model Training","ref_index":54,"is_internal_anchor":true},{"citing_arxiv_id":"2501.05465","citing_title":"Small Language Models (SLMs) Can Still Pack a Punch: A survey (updated 2026)","ref_index":152,"is_internal_anchor":true},{"citing_arxiv_id":"2506.15588","citing_title":"Memory-Efficient Differentially Private Training with Gradient Random Projection","ref_index":20,"is_internal_anchor":true},{"citing_arxiv_id":"2505.23737","citing_title":"On the Convergence Analysis of Muon","ref_index":28,"is_internal_anchor":true},{"citing_arxiv_id":"2506.13674","citing_title":"PrefixMemory-Tuning: Modernizing Prefix-Tuning by Decoupling the Prefix from Attention","ref_index":41,"is_internal_anchor":true},{"citing_arxiv_id":"2509.15113","citing_title":"Low-rank surrogate modeling and stochastic zero-order optimization for training of neural networks with black-box layers","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2509.18993","citing_title":"CR-Net: Scaling Parameter-Efficient Training with Cross-Layer Low-Rank Structure","ref_index":64,"is_internal_anchor":true},{"citing_arxiv_id":"2512.12131","citing_title":"BOOST: BOttleneck-Optimized Scalable Training Framework for Low-Rank Large Language Models","ref_index":28,"is_internal_anchor":true},{"citing_arxiv_id":"2601.11568","citing_title":"AdaFRUGAL: Adaptive Memory-Efficient Training with Dynamic Control","ref_index":4,"is_internal_anchor":true},{"citing_arxiv_id":"2602.21545","citing_title":"MUON+: Towards More Effective Muon via One Additional Normalization Step for LLM Pre-training","ref_index":40,"is_internal_anchor":true},{"citing_arxiv_id":"2604.00733","citing_title":"Spectral Compact Training: Pre-Training Large Language Models via Permanent Truncated SVD and Stiefel QR Retraction","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2604.03388","citing_title":"Scalable Variational Bayesian Fine-Tuning of LLMs via Orthogonalized Low-Rank Adapters","ref_index":18,"is_internal_anchor":true},{"citing_arxiv_id":"2403.14608","citing_title":"Parameter-Efficient Fine-Tuning for Large Models: A Comprehensive Survey","ref_index":139,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11838","citing_title":"Gradient Clipping Beyond Vector Norms: A Spectral Approach for Matrix-Valued Parameters","ref_index":70,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10288","citing_title":"BROS: Bias-Corrected Randomized Subspaces for Memory-Efficient Single-Loop Bilevel Optimization","ref_index":67,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12492","citing_title":"Pion: A Spectrum-Preserving Optimizer via Orthogonal Equivalence Transformation","ref_index":92,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10288","citing_title":"BROS: Bias-Corrected Randomized Subspaces for Memory-Efficient Single-Loop Bilevel Optimization","ref_index":67,"is_internal_anchor":true},{"citing_arxiv_id":"2605.09238","citing_title":"Intrinsic Muon: Spectral Optimization on Riemannian Matrix Manifolds","ref_index":71,"is_internal_anchor":true},{"citing_arxiv_id":"2605.03667","citing_title":"ELAS: Efficient Pre-Training of Low-Rank Large Language Models via 2:4 Activation Sparsity","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06316","citing_title":"Pro-KLShampoo: Projected KL-Shampoo with Whitening Recovered by Orthogonalization","ref_index":20,"is_internal_anchor":true},{"citing_arxiv_id":"2605.04913","citing_title":"Rethinking Local Learning: A Cheaper and Faster Recipe for LLM Post-Training","ref_index":33,"is_internal_anchor":true},{"citing_arxiv_id":"2605.01968","citing_title":"AdamO: A Collapse-Suppressed Optimizer for Offline RL","ref_index":89,"is_internal_anchor":true},{"citing_arxiv_id":"2604.11705","citing_title":"Agentic Driving Coach: Robustness and Determinism of Agentic AI-Powered Human-in-the-Loop Cyber-Physical Systems","ref_index":42,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/C4EW5CF5ETZB7Y2CQ4JOCWDXXG","json":"https://pith.science/pith/C4EW5CF5ETZB7Y2CQ4JOCWDXXG.json","graph_json":"https://pith.science/api/pith-number/C4EW5CF5ETZB7Y2CQ4JOCWDXXG/graph.json","events_json":"https://pith.science/api/pith-number/C4EW5CF5ETZB7Y2CQ4JOCWDXXG/events.json","paper":"https://pith.science/paper/C4EW5CF5"},"agent_actions":{"view_html":"https://pith.science/pith/C4EW5CF5ETZB7Y2CQ4JOCWDXXG","download_json":"https://pith.science/pith/C4EW5CF5ETZB7Y2CQ4JOCWDXXG.json","view_paper":"https://pith.science/paper/C4EW5CF5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.03507&json=true","fetch_graph":"https://pith.science/api/pith-number/C4EW5CF5ETZB7Y2CQ4JOCWDXXG/graph.json","fetch_events":"https://pith.science/api/pith-number/C4EW5CF5ETZB7Y2CQ4JOCWDXXG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/C4EW5CF5ETZB7Y2CQ4JOCWDXXG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/C4EW5CF5ETZB7Y2CQ4JOCWDXXG/action/storage_attestation","attest_author":"https://pith.science/pith/C4EW5CF5ETZB7Y2CQ4JOCWDXXG/action/author_attestation","sign_citation":"https://pith.science/pith/C4EW5CF5ETZB7Y2CQ4JOCWDXXG/action/citation_signature","submit_replication":"https://pith.science/pith/C4EW5CF5ETZB7Y2CQ4JOCWDXXG/action/replication_record"}},"created_at":"2026-05-17T23:38:46.186806+00:00","updated_at":"2026-05-17T23:38:46.186806+00:00"}