Creating weight when exporting from Roboflow
We don’t yet support exporting model weights from Roboflow as it is a valuable feature of our platform. We sell it as an add-on for Enterprise plans.
We highly recommend our hosted inference service as our full time ML team will handle all the infrastructure and hosting for you. If you are interested in running inference on your own machine, we just recently open-sourced our inference server for custom deployment.
If you specifically need the .pt file, you can use our training notebooks to train on your own compute (which will provide the .pt file). As an extra upside, models trained this way can be uploaded back to Roboflow for use in deployment, label assist or model evaluation.