Type: url
Original: https://developers.googleblog.com/a-guide-to-fine-tuning-functiongemma/
Status: generated
Summary
A Google DeepMind tutorial post on fine-tuning FunctionGemma, a function-calling variant of Gemma 3 270M, so that it resolves tool-selection ambiguity according to an organization’s own policy — in its case study, routing internal-policy questions to a knowledge-base search tool rather than to Google search. It walks through supervised fine-tuning with Hugging Face TRL, warns that an unshuffled train-test split of category-sorted data can leave the model trained on only one tool, and introduces the FunctionGemma Tuning Lab, a no-code Hugging Face Spaces demo for the same process.