Roboflow + Amazon SageMaker
Pull Roboflow dataset versions into SageMaker notebooks with the Python package, train there, then upload the weights to Roboflow or serve the model on a SageMaker endpoint.
How to connect Amazon SageMaker to Roboflow
5 stepsGenerate a dataset version
In Roboflow, generate a version of your project with the preprocessing and augmentation you want. Downloads always come from a version.
Install the Roboflow package
In a SageMaker Studio notebook, run pip install roboflow and set your Roboflow API key as an environment variable.
Download the version
Call project.version(n).download("yolov8") (or "coco", "voc" and so on). The SDK fetches and unzips the export into the notebook's storage.
Train on SageMaker compute
Run your training code against the downloaded folder on the instance type you choose.
Upload the weights or deploy on SageMaker
For a supported architecture, call version.deploy() with the model type and output folder to host the model on Roboflow. To keep it in AWS, follow our guide to serve it as ONNX on a SageMaker endpoint.
What teams build with Amazon SageMaker
Train inside your AWS account
Keep training jobs and data on SageMaker while Roboflow manages labeling and dataset versions.
Host a SageMaker-trained model on Roboflow
Upload YOLO or RF-DETR weights trained on SageMaker and call them through the Serverless Hosted API or Workflows.
Send hard images back for labeling
Upload images your SageMaker endpoint is unsure about to a Roboflow project with project.upload() for review.
Questions
No. The integration is the roboflow Python package running in your SageMaker notebook, which downloads dataset versions and uploads weights.
Architectures listed for weight upload in our docs, including RF-DETR, YOLO11, YOLO26, YOLOv8 and YOLOv5. Each has its own package version requirements, such as ultralytics==8.0.196 for YOLOv8.
Yes. Our docs cover deploying a Roboflow Inference server on an EC2 instance with the Inference CLI.
Why are companies choosing Roboflow?
The end-to-end platform for vision AI: create datasets, train models and deploy to production.
Flexible deployment modes
Move from a scalable cloud API to edge hardware like NVIDIA Jetson without managing complex dependencies.
Integrated workflow builder
Combine multiple AI models, custom business logic and third-party integrations into one production-ready pipeline in a low-code editor.
Model monitoring and alerts
Dashboards track confidence, inference time and request volume for your deployed models, and email alerts tell your team when confidence suddenly drops.
