Amazon SageMaker AI introduces serverless customization for Qwen3.5 models

Amazon SageMaker AI has expanded its capabilities to include serverless model customization for the Qwen3.5 model series. This enhancement allows users to fine-tune Qwen3.5 models, which include 4B, 9B, and 27B parameter configurations, using both supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT) techniques. The Qwen3.5 models are a well-regarded open-weight model family developed by Alibaba Cloud.

Previously, users could deploy these base models on SageMaker AI, but the new update offers the ability to adapt them to specific domains and workflows. This model customization feature enables users to infuse the foundation models with proprietary data, ensuring they better reflect domain-specific knowledge, terminology, and quality standards. Instead of creating models from scratch, fine-tuning allows users to start with a robust base model and refine it for specific applications, such as enhancing accuracy on specialized tasks, aligning outcomes with organizational tone, or boosting performance on new tasks with labeled data.

The serverless customization feature on SageMaker AI simplifies the process by managing all infrastructure provisioning and training orchestration. This allows users to concentrate on data and evaluation without the need for cluster management, and costs are incurred only for the resources used.

Currently, serverless model customization for Qwen3.5 is available in several regions: US East (N. Virginia), US West (Oregon), Asia Pacific (Tokyo), and EU (Ireland). To begin utilizing this feature, users can go to the Models page in Amazon SageMaker Studio to initiate a customization job or employ the SageMaker Python SDK for programmatic access. For further details, users are encouraged to consult the Amazon SageMaker AI model customization documentation.