Train a "clean" Yolov12 model (not pre-trained) on a custom dataset

Support YOLO
, , ,
1

Hi,
Been strugling with finding the “clean”, not-pre-trained Yolov12 .yaml files in order to train a yolo model from scratch.
ChatGPT and google, etc didnt provided working solution.
the files are not on the Yolov12 seperated github as well.
please help with understand where can i get the different yolov12 flavors .yaml files for the un-trained models.
ref to files or Simple python code with example will be much appriciated,
Thanks,
Mac_J

2

You just provide YAML filename. Ultralytics automatically uses the correct file. You don’t need to download it manually.

model = YOLO("yolo12n.yaml")

1 Like
3

Thank you!

4

You’re welcome!

We generally advise against using YOLO12 due to training instability, high memory consumption, and significantly slower CPU speeds. For all use cases, we recommend using YOLO11, which provides state-of-the-art performance and stability.

To train a new model from scratch on a custom dataset, you can simply use the model’s .yaml file. Here is a short example using YOLO11:

from ultralytics import YOLO

# Build a new YOLO11n model from its configuration file
model = YOLO("yolo11n.yaml")

# Train the model on your custom dataset
results = model.train(data="path/to/your_dataset.yaml", epochs=100, imgsz=640)

You can find more information in our documentation on how to train a model. Hope this helps