TFJS support model.json to ONNX conversion

[JohnRSim]

Feature request

Currently using node to create an image-classifier model.json with tfjs

• I don't think Optimum support this format to convert to onnx?

It would be nice to just use optimum and point to model.json.

Motivation

Currently I'm creating the model converting it to graph and then converting to onnx like this -

tensorflowjs_converter --input_format=tfjs_layers_model \  --output_format=tfjs_graph_model \  ./saved-model/layers-model/model.json \  ./saved-model/graph-model

python3 -m tf2onnx.convert --tfjs ./saved-model/graph-model/model.json --output ./saved-model/model.onnx

I'm not sure how to switch to use optimum - do I need to convert model.json to .h5 and then run?

• if I try this I run into huggingface_hub.errors.HFValidationError: Repo id must be in the form 'repo_name' or 'namespace/repo_name': './path_to_save/model.h5'. Use repo_type argument if needed

Your contribution

N/A