# Gemma 4 Joins Serverless Fine-Tuning on SageMaker AI!

Hi everyone, it's Shii-chan! Today's news is about growing AI models to fit your own needs, so I'm pretty excited.

## What was announced?

According to AWS What's New, Amazon SageMaker AI now supports serverless model customization for Gemma 4 models. The supported ones are Gemma 4 E4B and Gemma 4 31B. Gemma is a family of open models built by Google DeepMind.

Three tuning methods are available: supervised fine-tuning (SFT), direct preference optimization (DPO), and reinforcement fine-tuning (RFT).

## The story so far

Serverless customization on SageMaker AI has already supported model families like Nova, Nemotron 3, Qwen, Llama, gpt-oss, and DeepSeek. Now Gemma 4 joins the lineup, widening the range of models you can pick.

## What changes

Because it's serverless, SageMaker handles infrastructure provisioning and training orchestration for you. Instead of managing clusters, you get to focus on your data and evaluation. Pricing is pay-per-use, which is nice when you want to start small.

## Dive Deep

It's available in four Regions: US East (N. Virginia), US West (Oregon), Asia Pacific (Tokyo), and EU (Ireland). Having Tokyo in the mix is reassuring for those of us in Japan.

There are two ways to start: launch a customization job from the Models page in Amazon SageMaker Studio, or run it programmatically with the [SageMaker Python SDK](https://sagemaker.readthedocs.io/en/stable/model_customization/index.html). The [model customization documentation](https://docs.aws.amazon.com/sagemaker/latest/dg/customize-model.html) has the full steps if you want to take a look.

## Wrap-up

- SageMaker AI now supports serverless customization for Gemma 4 E4B and Gemma 4 31B
- Three methods (SFT, DPO, RFT) are available, and infrastructure is handled for you
- Available in N. Virginia, Oregon, Tokyo, and Ireland, with pay-per-use pricing
- Get started from the Studio Models page or the Python SDK

If you want to train an open model on your own data without managing clusters, this is a great option for you!