Get started fine-tuning foundation models in Amazon SageMaker Unified Studio

Amazon SageMaker Unified Studio provides a large collection of state-of-the-art foundation models. These models support use cases such as content writing, code generation, question answering, copywriting, summarization, classification, information retrieval, and more. You can find, customize, and deploy these foundation models in the JumpStart model catalog. You can use the foundation models to build your own generative AI solutions for a wide range of applications.

A foundation model is a large pre-trained model that is adaptable to many downstream tasks and often serves as the starting point for developing more specialized models. Examples of foundation models include Meta Llama 4 Maverick 17B, DeepSeek-R1, or Stable Diffusion 3.5 Large. These models are pre-trained on massive amounts of data.

Model customization

You might need to customize a base foundation model to better align it with your specific use cases. The recommended way to first customize a foundation model is through prompt engineering. Providing your foundation model with well-engineered, context-rich prompts can help achieve desired results without any fine-tuning or changing of model weights. For more information, see Prompt engineering for foundation models in the Amazon SageMaker AI Developer Guide.

If prompt engineering alone is not enough to customize your foundation model to a specific task, you can fine-tune a foundation model on additional domain-specific data. The fine-tuning process involves changing model weights.

To help you learn how to fine-tune foundation models, Amazon SageMaker Unified Studio provides an example training dataset for each model that's eligible for training. You can also choose to fine-tune the model with your own data set. Before you can do that, you must prepare your data set and store it in an Amazon S3 bucket. The required format for the data set varies between models. You can learn about the required format in the model details page in Amazon SageMaker Unified Studio.

Fine-tuning a foundation model

One way to fine-tune a model in Amazon SageMaker Unified Studio is to use JumpStart. First, you choose a foundation model from the catalog. Then, you train the model with a training data set. Follow these steps to learn how to fine-tune with this approach.

1. Sign in to Amazon SageMaker Unified Studio using the link that your administrator gave you.

2. Choose a model to train.

   1. From the main menu, choose Build.

   2. From the drop-down menu, choose Jumpstart Models.

      The JumpStart page lists the model providers.

   3. Choose a model provider. The page displays the models for that provider.

   4. Under Action, choose Trainable. The page displays the trainable models for that provider.

   5. From the provider's list of models, choose the model you want to train.

      Amazon SageMaker Unified Studio shows the model details page, which provides information from the model provider. If you want to prepare a custom fine-tuning data set, use this page to learn the required format.

3. From the model details page, choose Train to create a training job.

4. On the Fine-tune model page, under Data, do one of the following:

   1. Keep the default selection of Example training dataset. This data set is useful when you want to learn how to fine-tune with Amazon SageMaker Unified Studio. However, it won't be effective for customizing the model for your specific needs.

   2. If you've prepared a custom training data set, choose Enter training dataset, and provide the URI that locates it in Amazon S3.

5. (Optional) Under Hyperparameters, update the hyperparameters you want to change.

   The hyperparameters available for each fine-tunable model differ depending on the model. Review the help text and additional information in the model details pages in Amazon SageMaker Unified Studio to learn more about hyperparameters specific to the model of your choice.

   For more information on available hyperparameters, see Commonly supported fine-tuning hyperparameters in the Amazon SageMaker AI Developer Guide.

6. Under Deployment, for Training Instance, specify the training instance type for your training job. You can only choose from instances that are compatible with the model that you chose.

   For Output artifact location (S3 URI), specify where Amazon SageMaker uploads the fine-tuned model. You can choose to use the default bucket, or you can specify a custom location in Amazon S3.

7. (Optional) Under Additional Information, for Training Job Name, you can edit the default name.

8. (Optional) For Tags, you can add and remove tags in the form of key-value pairs to help organize and categorize your fine-tuning training jobs.

9. Enter Submit to submit the training job. You can view the training job from the Training jobs page.