SageMaker Serverless Model Customization Adds Full Fine-Tuning
Amazon SageMaker now supports full fine-tuning for over 25 open-source models within its serverless model customization feature. This enhancement allows for deeper adaptation by updating all model parameters, which is crucial for specialized tasks and domain knowledge acquisition. The feature is now available in select AWS regions and can be accessed via SageMaker Studio or the Python SDK, benefiting engineers and architects working with LLMs.
- →Full fine-tuning support for open-source models
- →Serverless infrastructure management for fine-tuning
- →Enables deeper model adaptation for specialized use cases
- →Availability and access methods
Features (2) ›
- Full fine-tuning support for open-source models
SageMaker serverless model customization now supports full fine-tuning for over 25 open-source models, including those from gpt-oss, Gemma, Llama, Nemotron, and Qwen families. This allows updating all model parameters for deeper adaptation beyond parameter-efficient methods like LoRA.
- Serverless infrastructure management for fine-tuning
SageMaker manages all infrastructure provisioning and training orchestration for full fine-tuning jobs. This allows users to run these jobs without provisioning or managing infrastructure, with pay-as-you-go pricing.
Enhancements (1) ›
- Enables deeper model adaptation for specialized use cases
Full fine-tuning is valuable for models needing to learn domain-specific patterns, terminology, task structures, specialized reasoning, complex output formats, or internalize knowledge from large datasets.
Notes (1) ›
- Availability and access methods
Serverless full fine-tuning is available in US East (N. Virginia), US West (Oregon), Asia Pacific (Tokyo), and Europe (Ireland). Users can initiate jobs via SageMaker Studio's JumpStart and Models page or the SageMaker Python SDK.
https://aws.amazon.com/about-aws/whats-new/2026/08/amazon-sagemaker-fft
Related releases
- AWS Blog: Unified Access Control for Enterprise Lakehouses AWS Big Data Blog ·
- AWS Glue integrates with SageMaker Unified Studio for one-click data access AWS What's New ·
- NVIDIA and Qwen AI Models Now on Amazon SageMaker JumpStart AWS What's New ·
- NVIDIA Nemotron 3.5 Lightning AI Model Now Available on Amazon SageMaker JumpStart AWS What's New ·
- New GLM-5.2 FP8, Nemotron-Nano-12B-v2, and GLM-OCR models on SageMaker JumpStart AWS What's New ·
- New AI Models from Redis, JetBrains, and LightOn Now on SageMaker JumpStart AWS What's New ·