{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,7,9]],"date-time":"2026-07-09T16:09:43Z","timestamp":1783613383657,"version":"3.55.0"},"reference-count":33,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,2,17]],"date-time":"2026-02-17T00:00:00Z","timestamp":1771286400000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Artif. Intell."],"abstract":"<jats:sec>\n                    <jats:title>Introduction<\/jats:title>\n                    <jats:p>Large Language Models (LLMs) achieve strong performance on many Natural Language Processing tasks, but adapting them to domain-specific applications is resource-intensive due to the cost of curating task-specific datasets and the compute required for fine-tuning. This work proposes an end-to-end strategy for rapidly fine-tuning LLMs for domain-specific tasks when both data and compute are limited.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>The strategy uses Distilling Step-by-Step (DSS) for dataset development and model training, where a teacher model generates task labels and intermediate rationales via Chain-of-Thought prompting for a natural-language-to-Query-DSL structured generation task. Using the resulting supervision, we benchmark three fine-tuning modalities through hyperparameter sweeps: full-precision fine-tuning, Low-Rank Adaptation (LoRA), and Quantized LoRA (QLoRA). To isolate the effect of rationale supervision, we additionally conduct an ablation study comparing DSS training (label + rationale supervision) against a label-only configuration.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>Across the evaluated configurations, DSS combined with full-precision fine-tuning yields the strongest overall performance. Under resource constraints, DSS with LoRA provides an effective performance-efficiency tradeoff, and DSS with QLoRA enables training under tighter GPU memory budgets while maintaining competitive performance. In the parameter-efficient regimes, an alpha-to-rank ratio of 4:1 provides a consistent balance of performance and compute consumption across the explored settings.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Discussion<\/jats:title>\n                    <jats:p>These findings support a practical process for resource-constrained domain adaptation: use DSS to efficiently construct datasets, then select the fine-tuning modality based on available compute (full-precision when feasible; LoRA or QLoRA when memory-limited). The proposed workflow offers a general guide for efficiently fine-tuning LLMs for domain-specific tasks with limited data availability.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/frai.2026.1665992","type":"journal-article","created":{"date-parts":[[2026,2,17]],"date-time":"2026-02-17T06:34:19Z","timestamp":1771310059000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":1,"title":["An efficient strategy for fine-tuning large language models"],"prefix":"10.3389","volume":"9","author":[{"given":"Benjamin","family":"Marsh","sequence":"first","affiliation":[{"name":"Marine Corps Tactical Systems Support Activity, United States Marine Corps","place":["Camp Pendleton, CA, United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Adam","family":"Michaleas","sequence":"additional","affiliation":[{"name":"MIT Lincoln Laboratory, Artificial Intelligence Technology","place":["Lexington, MA, United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Darrell O.","family":"Ricke","sequence":"additional","affiliation":[{"name":"MIT Lincoln Laboratory, Artificial Intelligence Technology","place":["Lexington, MA, United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Shaun","family":"Monera","sequence":"additional","affiliation":[{"name":"Marine Corps Tactical Systems Support Activity, United States Marine Corps","place":["Camp Pendleton, CA, United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Shriya","family":"Zembruski","sequence":"additional","affiliation":[{"name":"Marine Corps Tactical Systems Support Activity, United States Marine Corps","place":["Camp Pendleton, CA, United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,2,17]]},"reference":[{"key":"B1","first-page":"932","article-title":"A neural probabilistic language model","volume":"3","author":"Bengio","year":"2000","journal-title":"J. Mach. Learn. Res"},{"key":"B2","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2305.05176","article-title":"FrugalGPT: how to use large language models while reducing cost and improving performance","author":"Chen","year":"2023","journal-title":"arXiv"},{"key":"B3","unstructured":"Contributors\n              O.\n            \n          \n          Query DSL\n          \n          2024"},{"key":"B4","article-title":"QLoRA: Efficient finetuning of quantized LLMs","author":"Dettmers","year":"2023","journal-title":"arXiv"},{"key":"B5","doi-asserted-by":"crossref","first-page":"8342","DOI":"10.18653\/v1\/2020.acl-main.740","article-title":"\u201cDon't stop pretraining: Adapt language models to domains and tasks,\u201d","volume-title":"Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics","author":"Gururangan","year":"2020"},{"key":"B6","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2403.14608","article-title":"Parameter-efficient fine-tuning for large models: a comprehensive survey","author":"Han","year":"2024","journal-title":"arXiv"},{"key":"B7","article-title":"\u201cTowards a unified view of parameter-efficient transfer learning,\u201d","volume-title":"Proceedings of the 10th International Conference on Learning Representations (ICLR-2022)","author":"He","year":"2022"},{"key":"B8","author":"Hinton","year":"2015","journal-title":"Distilling the knowledge in a neural network."},{"key":"B9","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2023.findings-acl.507","article-title":"Distilling step-by-step! outperforming larger language models with less training data and smaller model sizes","author":"Hsieh","year":"2023","journal-title":"arXiv"},{"key":"B10","article-title":"LORA: low-rank adaptation of large language models","author":"Hu","year":"2021","journal-title":"arXiv"},{"key":"B11","doi-asserted-by":"publisher","DOI":"10.1109\/ICCTIT64404.2024.10928552","article-title":"Optimizing large language models with an enhanced lora fine-tuning algorithm for efficiency and robustness in nlp tasks","author":"Hu","year":"2024","journal-title":"arXiv"},{"key":"B12","unstructured":"bitsandbytes"},{"key":"B13","unstructured":"uppercasePEFT (Parameter-Efficient Fine-Tuning)"},{"key":"B14","doi-asserted-by":"publisher","DOI":"10.1016\/j.knosys.2025.114241","article-title":"Lora dropout as a sparsity regularizer for overfitting control","author":"Lin","year":"2024","journal-title":"arXiv"},{"key":"B15","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2303.13375","article-title":"Capabilities of GPT-4 on medical challenge problems","author":"Nori","year":"2023","journal-title":"arXiv"},{"key":"B16","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2407.13906","article-title":"Crafting efficient fine-tuning strategies for large language models","author":"Oliver","year":"2024","journal-title":"arXiv"},{"key":"B17","unstructured":"Pytorch\n          \n          2025"},{"key":"B18","unstructured":"Quentin\n              A.\n            \n            \n              Stella Biderman\n              H. S.\n            \n          \n          Transformer Math 101\n          \n          2023"},{"key":"B19","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1910.10683","article-title":"Exploring the limits of transfer learning with a unified text-to-text transformer","author":"Raffel","year":"2023","journal-title":"arXiv"},{"key":"B20","first-page":"455","article-title":"\u201cRepeatability of fine-tuning large language models illustrated using QLoRA,\u201d","volume-title":"Proceedings of the IEEE International Conference on Machine Learning and Applications","author":"Rajabzadeh","year":"2024"},{"key":"B21","doi-asserted-by":"crossref","DOI":"10.1109\/SC41405.2020.00024","article-title":"\u201cZeRO: memory optimizations toward training trillion parameter models,\u201d","volume-title":"Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis, SC '20","author":"Rajbhandari","year":"2020"},{"key":"B22","unstructured":"Rancher Community Edition Kubernetes Environment\n          \n          2024"},{"key":"B23","doi-asserted-by":"crossref","first-page":"9895","DOI":"10.18653\/v1\/2021.emnlp-main.779","article-title":"\u201cPICARD: Parsing incrementally for constrained auto-regressive decoding from language models,\u201d","volume-title":"Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing","author":"Scholak","year":"2021"},{"key":"B24","doi-asserted-by":"publisher","first-page":"1929","DOI":"10.5555\/2627435.2670313","article-title":"Dropout: A simple way to prevent neural networks from overfitting","volume":"15","author":"Srivastava","year":"2014","journal-title":"J. Mach. Learn. Res"},{"key":"B25","unstructured":"Team\n              M. A.\n            \n          \n          Cheaper, Better, Faster, Stronger\n          \n          2024"},{"key":"B26","unstructured":"Team\n              T. R.\n            \n          \n          Get Started with Distributed Training Using Hugging Face Accelerate\n          \n          2023"},{"key":"B27","doi-asserted-by":"publisher","first-page":"100729","DOI":"10.1016\/j.patter.2023.100729","article-title":"Fine-tuning large neural language models for biomedical natural language processing","volume":"4","author":"Tinn","year":"2023","journal-title":"Patterns"},{"key":"B28","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1706.03762","article-title":"Attention is all you need","author":"Vaswani","year":"2023","journal-title":"arXiv"},{"key":"B29","article-title":"\u201cFinetuned language models are zero-shot learners,\u201d","volume-title":"International Conference on Learning Representations","author":"Wei","year":"2022"},{"key":"B30","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2201.11903","article-title":"Chain-of-thought prompting elicits reasoning in large language models","author":"Wei","year":"2023","journal-title":"arXiv"},{"key":"B31","doi-asserted-by":"crossref","first-page":"3911","DOI":"10.18653\/v1\/D18-1425","article-title":"\u201cSpider: a large-scale human-labeled dataset for complex and cross-domain semantic parsing and text-to-SQL task,\u201d","volume-title":"Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing","author":"Yu","year":"2018"},{"key":"B32","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2309.01219","article-title":"Siren's song in the ai ocean: a survey on hallucination in large language models","author":"Zhang","year":"2023","journal-title":"arXiv"},{"key":"B33","article-title":"Seq2SQL: generating structured queries from natural language using reinforcement learning","author":"Zhong","year":"2017","journal-title":"arXiv preprint"}],"container-title":["Frontiers in Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1665992\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,2,17]],"date-time":"2026-02-17T06:34:23Z","timestamp":1771310063000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1665992\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,2,17]]},"references-count":33,"alternative-id":["10.3389\/frai.2026.1665992"],"URL":"https:\/\/doi.org\/10.3389\/frai.2026.1665992","relation":{},"ISSN":["2624-8212"],"issn-type":[{"value":"2624-8212","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,2,17]]},"article-number":"1665992"}}