Models: Local Weights

You can upload supported weights to Roboflow and deploy them to your device.

This is ideal if you have already trained a model outside of Roboflow that you want to deploy with Inference.

To upload weights to Roboflow, you will need:

  1. A Roboflow account
  2. A project with your dataset (that does not have a trained model)

To learn how to create a project and a dataset, refer to these guides:

Once you have a project with a dataset, you can upload your weights.

Install the Roboflow Python package:

pip install roboflow

Then, create a new Python file and add the following code:

import roboflow

roboflow.login()

rf = roboflow.Roboflow()
project = rf.project("your-project-id")
version = project.version(1)
version.deploy("model-type", "path/to/training/results/")

The following model types are supported:

Model Architecture Task Model Type ID
YOLOv5 Object Detection yolov5
YOLOv5 Segmentation yolov5-seg
YOLOv7 Object Detection yolov7-seg
YOLOv8 Object Detection yolov8
YOLOv8 Segmentation yolov8-seg
YOLOv8 Classification yolov8-cls
YOLOv8 Pose Estimation yolov8-pose
YOLOv9 Object Detection yolov9
YOLOv9 Segmentation yolov9
YOLO-NAS Object Detection yolonas
YOLOv10 Object Detection yolov10
PaliGemma Multimodal paligemma-3b-pt-224
PaliGemma Multimodal paligemma-3b-pt-448
PaliGemma Multimodal paligemma-3b-pt-896
Florence-2 Multimodal florence-2-large
Florence-2 Multimodal florence-2-base

In the code above, replace:

  1. your-project-id with the ID of your project. Learn how to retrieve your Roboflow project ID.
  2. 1 with the version number of your project.
  3. model-type with the model type you want to deploy.
  4. path/to/training/results/ with the path to the weights you want to upload. This path will vary depending on what model architecture you are using.

Your model weights will be uploaded to Roboflow. It may take a few minutes for your weights to be processed. Once your weights have been processed, your dataset version page will be updated to say that a model is available with your weights.

You can then use the model with Inference following our Run a Private, Fine-Tuned Model model.