Amazon SageMaker AI now supports custom sequence lengths for serverless model customization jobs

This article announces that Amazon SageMaker AI now supports custom sequence lengths for serverless model customization jobs, enabling fine-tuning of open-weight models with flexible parameters.

• Configure sequence length parameters for SFT, DPO, and RFT fine-tuning jobs
• Set dataset_max_len for SFT/DPO or max_prompt_length and max_response_length for RFT jobs
• Support for models including gpt-oss, Gemma, Llama, Nemotron, and Qwen
• Sequence lengths up to 16K tokens depending on model and technique
• Available in US East (N. Virginia), US West (Oregon), Asia Pacific (Tokyo), and Europe (Ireland)
• Access via SageMaker Studio, Python SDK, or Agent skills

This enhancement gives customers greater control over model customization to match their specific training data characteristics and use case requirements.