Customizing Amazon Nova models on Amazon SageMaker AI
You can customize Amazon Nova models through recipes and train them on SageMaker AI. These recipes support techniques such as supervised fine-tuning (SFT) and Direct Preference Optimization (DPO), with both full-rank and low-rank adaptation (LoRA) options.
The end-to-end customization workflow involves stages like model training, model evaluation, and deployment for inference. This model customization approach on SageMaker AI provides greater flexibility and control to fine-tune its supported Amazon Nova models, optimize hyperparameters with precision, and implement techniques such as LoRA parameter-efficient fine-tuning (PEFT), full-rank SFT, DPO, Continued Pre-Training (CPT), Proximal Policy Optimization (PPO), etc.
SageMaker AI offers two environments for customizing Amazon Nova models.
-
Amazon SageMaker training jobs provides a fully managed environment for customizing Nova models where you don't need to create or maintain any clusters. The service automatically handles all infrastructure provisioning, scaling, and resource management, allowing you to focus solely on configuring your training parameters and submitting your job. You can customize Nova models on SageMaker training jobs with techniques like Parameter Efficient Fine-tuning (PEFT), Full rank fine tuning as well as Direct Preference Optimization (DPO). For more information, see Amazon Nova customization on SageMaker training jobs.
-
Amazon SageMaker HyperPod offers a specialized environment to train Nova models by requiring you to create and manage SageMaker HyperPod EKS clusters with restricted instance groups (RIGs). This environment gives you flexibility in configuring your training environment with specialized GPU instances and integrated Amazon FSx for Lustre storage, making it particularly well-suited for advanced distributed training scenarios and ongoing model development. For more information, see Amazon Nova customization on Amazon SageMaker HyperPod.
In this chapter
