Supervised Fine-Tuning

Amazon Nova Forge data mixing allows you to combine your custom training data with Amazon Nova's proprietary training data during supervised fine-tuning (SFT). This helps preserve the model's general capabilities while specializing it for your target domain. Data mixing is available on two platforms:

SageMaker HyperPod – Use YAML recipe files to configure data mixing with both text and multimodal data. Supports LoRA and full-rank SFT.
SageMaker Training Jobs – Use the serverless CreateTrainingJob API with ServerlessJobConfig for text-only SFT with data mixing. Supports LoRA and full-rank.

Topics

Data mixing on SageMaker HyperPod
Data mixing on SageMaker Training Jobs

Warning Javascript is disabled or is unavailable in your browser.

To use the Amazon Web Services Documentation, Javascript must be enabled. Please refer to your browser's Help pages for instructions.

Document Conventions

Continued Pre-Training and Mid-Training

Data mixing on SageMaker HyperPod