Google has introduced a new way to personalize its artificial intelligence models through fine-tuning, available in Google AI Studio and through the Gemini API. This article, updated from an original October 2023 post, explains how to fine-tune the Gemini 1.0 Pro model for specific use cases.
What is refining?
Refinement, or tuning, is a technique that improves the quality of a model's results by training it on numerous task-specific examples.
This method is particularly useful for developers who require high-quality output that the few-example prompting method cannot provide.
Fine refining vs. efficient parameter refinement
Traditional fine-tuning adapts a pre-trained model to specific tasks using a small set of labeled data.
However, with today's large language models, this method can be complex and requires a lot of data and computation.
Conversely, Parameter Efficient Tuning (PET) in Google AI Studio delivers high-quality, personalized models with less data, less latency, and without the additional costs of classic fine refining.
Why use refining?
Fine-tuning allows you to customize Gemini models with your own data for better performance on specialized tasks while reducing prompt context size and response latency.
Use cases include classification, information extraction, generating structured output, and creating critical models.
How to quickly get started with Google AI Studio
Create a refined model.
It's easy to create a fine-tuned model in Google AI Studio, eliminating the need for engineering expertise to develop custom models.
Start by selecting “New tuned model” from the menu bar on the left.
Select the data for refinement.
You can refine your template from an existing structured prompt or import data from Google Sheets or a CSV file.
It is recommended to start with at least 20 examples and use a dataset of at least 100 examples for best performance.
See your refined model
Track the refinement progress in your library, and once the model is ready, view the details by clicking on your model.
You can then run your refined model through a structured or free prompt.
Run your refined model at any time.
Access your refined template by creating a new structured or freeform prompt and selecting your template from the list of available templates.
Refinement with the Gemini API
In addition to Google AI Studio, you can access refinement via the Gemini API by passing the training data in the API request when creating a refined model.
Google encourages users to share their ideas or use cases on X (formerly Twitter) or LinkedIn.