Bookmark · 6 August 2025, 09:34 Fine-tuned Small LLMs Can Beat Large Ones at 5-30x Lower Cost with Programmatic Data Curation · TensorZero ml-localmitp