{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:C4EW5CF5ETZB7Y2CQ4JOCWDXXG","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":"7b1eac0584b3aab5267eb2f6d6033547186f3b3a34a6f003bdeae70ede5b6275","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-03-06T07:29:57Z","title_canon_sha256":"b3d8e896d144add070f5ec4b0c898932d6e3a058b1cd57804eba277c6a3cfaa2"},"schema_version":"1.0","source":{"id":"2403.03507","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2403.03507","created_at":"2026-05-17T23:38:46Z"},{"alias_kind":"arxiv_version","alias_value":"2403.03507v2","created_at":"2026-05-17T23:38:46Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.03507","created_at":"2026-05-17T23:38:46Z"},{"alias_kind":"pith_short_12","alias_value":"C4EW5CF5ETZB","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"C4EW5CF5ETZB7Y2C","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"C4EW5CF5","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:98815ee06178dbf16685f4bb39f7cbfa69502e96ded62a4d4042aede7fc2325d","target":"graph","created_at":"2026-05-17T23:38:46Z","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 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."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","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."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","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."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"GaLore projects full gradients onto low-rank subspaces periodically, cutting optimizer memory by 65.5% while training every parameter of large language models."}],"snapshot_sha256":"716137a95d8b5a93ac44d1d24eb53cd4af93ab812ef2053f7d65aaa71f64fa98"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"27962be8dc6346f405f988acb1bfa29e625c6b7e11ae694deef8035f87cf45c8"},"paper":{"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","authors_text":"Anima Anandkumar, Beidi Chen, Jiawei Zhao, Yuandong Tian, Zhangyang Wang, Zhenyu Zhang","cross_cats":[],"headline":"GaLore projects full gradients onto low-rank subspaces periodically, cutting optimizer memory by 65.5% while training every parameter of large language models.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-03-06T07:29:57Z","title":"GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection"},"references":{"count":55,"internal_anchors":9,"resolved_work":55,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Memory efficient adaptive optimization","work_id":"852fa34d-c73a-4d07-8012-d983fe698557","year":2019},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Belle: Be everyone's large language model engine","work_id":"84a8bc20-e93c-4601-8413-3f9c43212cae","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Continual learning in low-rank orthogonal subspaces","work_id":"c83f5d43-b1ee-4cf1-8c48-0c547920edd0","year":2020},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Non- Convex Projected Gradient Descent for Generalized Low-Rank Tensor Regression","work_id":"ab8e2819-94e0-4e6d-bcfd-75ab25d5680b","year":2019},{"cited_arxiv_id":"1604.06174","doi":"","is_internal_anchor":true,"ref_index":5,"title":"Training Deep Nets with Sublinear Memory Cost","work_id":"f2c5c287-a500-40e4-a136-e7e3172db1d7","year":2016}],"snapshot_sha256":"fb16556210ca767e58199a9a7a212f2ab5f7dace7778461b526f01ce8ca68a16"},"source":{"id":"2403.03507","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-16T23:47:37.157229Z","id":"8a22c18f-eb03-4180-bb8b-a3ccd8e03981","model_set":{"reader":"grok-4.3"},"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","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.","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.","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."}},"verdict_id":"8a22c18f-eb03-4180-bb8b-a3ccd8e03981"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:84e456eabb979e57bf8e2f6307adc095701beb250d359126d26bc5cda1acc522","target":"record","created_at":"2026-05-17T23:38:46Z","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":"7b1eac0584b3aab5267eb2f6d6033547186f3b3a34a6f003bdeae70ede5b6275","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-03-06T07:29:57Z","title_canon_sha256":"b3d8e896d144add070f5ec4b0c898932d6e3a058b1cd57804eba277c6a3cfaa2"},"schema_version":"1.0","source":{"id":"2403.03507","kind":"arxiv","version":2}},"canonical_sha256":"17096e88bd24f21fe3428712e15877b9989ea67ba7b88a5a35cb99e74b6d2990","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"17096e88bd24f21fe3428712e15877b9989ea67ba7b88a5a35cb99e74b6d2990","first_computed_at":"2026-05-17T23:38:46.186714Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:46.186714Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"mK00rJkKhh0MOWufNa/2sGbH5xL3iYsVTptZYaMVu14JSSD5ykvnpy0iXmGh/CJtLr91QFQMUSV1mEUVhkUBDA==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:46.187232Z","signed_message":"canonical_sha256_bytes"},"source_id":"2403.03507","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:84e456eabb979e57bf8e2f6307adc095701beb250d359126d26bc5cda1acc522","sha256:98815ee06178dbf16685f4bb39f7cbfa69502e96ded62a4d4042aede7fc2325d"],"state_sha256":"bef87e8290f42429fc01902964f7ae3ae57d809c9b39cb30e4e64a8a7de5721b"}