Which model checkpoint does the Deci platform support?
The model’s checkpoint file should meet the following requirements
Framework | Checkpoint File Requirements | How to generate? |
|---|---|---|
ONNX 1.8.0 (.onnx) | The checkpoint should be saved as a .onnx file, and must support inference with dynamic batch size. We currently support Opsets 9-11, (Opset 12 is in beta) | See link for instructions. |
TensorFlow 2.3 (zip with saved_model.pb) | The checkpoint should be a TF2 saved-model format. Make sure to zip the file's directory before uploading. | See link for instructions. |
Keras (.h5) | Save your Keras model in .h5 format. Make sure to save both the architecture and the weights. | See link for instructions. |
PyTorch (pth) | To enable the upload of a PyTorch model into Deci, convert it to an ONNX 1.8.0 format ( .onnx file) that supports inferencing with dynamic batch size. | See link for instructions. |
TorchScript what we want to remember what we want | Save your traced TorchScript module as a .pth file. |
Updated about 1 year ago