{"paper":{"arxiv_id":"2106.09685","title":"LoRA: Low-Rank Adaptation of Large Language Models","abstract":"An important paradigm of natural language processing consists of large-scale pre-training on general domain data and adaptation to particular tasks or domains. As we pre-train larger models, full fine-tuning, which retrains all model parameters, becomes less feasible. We propose Low-Rank Adaptation, or LoRA, which freezes the pre-trained model weights and injects trainable rank decomposition matrices into each layer of the Transformer architecture, greatly reducing the number of trainable parameters for downstream tasks.","primary_category":"cs.CL","venue":"ICLR 2022","published_at":null,"latest_version":1,"withdrawn":false},"latest_version":{"id":"a4b78855-cf89-4faf-a766-d5bc7c32ea6c","version":1,"source_url":"https://arxiv.org/abs/2106.09685","rendered_html_url":null,"rendering_engine":null},"verdict":{"id":"80aeb720-3ade-4c1e-b669-bb2b45e377d6","kind":"POST","status":"reproduced","score":0.89,"confidence":0.8,"agent_version":"v0.1.0-lora-mrpc-microslice","computed_at":"2026-05-14T23:40:53.397Z","is_current":true,"claim_citation":{"paper_arxiv_id":"2106.09685","section":"Table 2","row":"RoBERTa-base + LoRA (r=8)","column":"MRPC dev accuracy","reported_value":89.7,"reported_metric":"accuracy","quoted_text":"RoBERTa base 89.7","pdf_page":6,"notes":"Table 2 of arXiv:2106.09685 reports RoBERTa-base + LoRA (r=8) MRPC dev = 89.7. Driver fine-tunes `FacebookAI/roberta-base` with the paper's recipe (r=8, alpha=16, target query+value per §4.1) on a MRPC dev micro-slice. PROTOCOL_MATCH is `proxy` (dataset-size)."},"protocol_match":"proxy"},"verdicts":{"post":{"id":"80aeb720-3ade-4c1e-b669-bb2b45e377d6","kind":"POST","status":"reproduced","score":0.89,"confidence":0.8,"agent_version":"v0.1.0-lora-mrpc-microslice","computed_at":"2026-05-14T23:40:53.397Z","is_current":true,"claim_citation":{"paper_arxiv_id":"2106.09685","section":"Table 2","row":"RoBERTa-base + LoRA (r=8)","column":"MRPC dev accuracy","reported_value":89.7,"reported_metric":"accuracy","quoted_text":"RoBERTa base 89.7","pdf_page":6,"notes":"Table 2 of arXiv:2106.09685 reports RoBERTa-base + LoRA (r=8) MRPC dev = 89.7. Driver fine-tunes `FacebookAI/roberta-base` with the paper's recipe (r=8, alpha=16, target query+value per §4.1) on a MRPC dev micro-slice. PROTOCOL_MATCH is `proxy` (dataset-size)."},"protocol_match":"proxy"},"pre":null}}