- Switch to the model training interface.
- In the dropdown selection boxes under “Provider”, choose “Torchvision”, and under “Architecture”, select “EfficientNet”.
- For the dataset, select the one you have just uploaded and annotated on the platform.
- For time reasons, the training should not produce a perfect model at this stage but only demonstrate the process. Therefore, reduce the number of training epochs and set the batch size to 2 — either in the form fields or in the editable JSON section on the right-hand side.
- Set the Train/Test split to 50% (0.5) each.
- Next, define the ONNX export of the model. To do this, click the blue button next to the “Provider” dropdown.
- Enable the export configuration using the toggle next to the modal’s title.
- In the JSON field, insert the following configuration:
{
“model_type”: “pytorch”,
“input_names”: [],
“input_shapes”: [[1, 3, 300, 300]],
“input_types”: [ “float32” ],
“output_names”: [],
“verbose”: false,
“training”: “EVAL”,
“operator_export_type”: “ONNX”,
“opset_version”: 14
}
- Once the training container has been created, you can start the training directly in the interface. To do this, click the corresponding button in the list of created training containers.
- After the training is complete, view the generated log files. These contain all information that was recorded during the training process.
- Transfer the trained model from the training container to the platform’s model management by clicking the designated button in the list.