Posted by: Pranay Bhatia – Google Labs Product Manager
AI is also changing the way developers work and enabling more people to build. In Part 1, you learned how to easily prompt her LLM in plain language using MakerSuite. In today’s part 2, we’ll show you tuning in MakerSuite, which allows you to customize your model to your specific needs in minutes.
What is tuning?
In Part 1, I introduced several examples of a technique called few-shot prompting to improve model performance. Tuning improves this technique by training the model on more examples (more examples than can all fit in the prompt).
Fine-tune and adjust parameters efficiently
You may have heard of the classic “tweaking” of models. Here, we adapt a pre-trained model to a specific task by training it on a smaller set of task-specific labeled data. However, with today’s LLM and its huge number of parameters, retraining is complicated. It requires machine learning expertise, large amounts of data, and large amounts of computation.
MakerSuite tuning uses a technique called Parameter Efficient Tuning (PET) to create customized, high-quality models without the added cost and complexity of traditional fine-tuning. Additionally, PET produces high-quality models with just a few hundred data points, reducing the data collection burden for developers.
Adjust models in minutes with MakerSuite
1. Create a calibrated model
Adjusting your model in MakerSuite is easy. Just select “Create new” and select “Adjusted model”.
2. Selecting data to tune
Tune your model from saved data prompts or import data from Google Sheets or CSV files.[調整]We recommend using at least 100 samples for best performance before pressing the button.
3. Display the adjusted model
Check the tuning progress in the library. Once you have finished tuning your model, you can click on the model to view its details.
4. Run the adjusted model
To start using your newly adjusted model, create a new text or data prompt and select your newly adjusted model from the list of available models.
MakerSuite: Powerful and Easy Tuning Tool
Tuning with MakerSuite makes it easy for developers to exploit the full potential of models like PaLM 2. Whether you’ve already used the API to tune models or are just starting to experiment with generative AI, MakerSuite makes your models more relevant and effective for your applications in just minutes. You’ll find that it opens up exciting possibilities.
