Hello! I wanted to ask if there was a way to bundle all the dataclasses needed to generate a model explorer graph (from graph_builder.py and types.py) into a pure-python package in order to make it easier to generate Model Explorer graphs that can be loaded into the frontend.
This will make it a lot easier to use [https://www.npmjs.com/package/ai-edge-model-explorer-visualizer], since it would be possible to create these graphs in a generic fashion without depending on ai-edge-model-explorer-adapter (which bundles cpp extensions and extra dependencies). Essentially one could write a method that looks like Adapter.convert without actually importing Adapter, and pre-process their data to create a model explorer graph that can be loaded by the frontend. This would give users the choice of i) writing a lightweight conversion function and using the default support in the frontend component, or ii) importing Adapter, and getting the full functionality of a custom adapter.
Hello! I wanted to ask if there was a way to bundle all the dataclasses needed to generate a model explorer graph (from
graph_builder.pyandtypes.py) into a pure-python package in order to make it easier to generate Model Explorer graphs that can be loaded into the frontend.This will make it a lot easier to use [https://www.npmjs.com/package/ai-edge-model-explorer-visualizer], since it would be possible to create these graphs in a generic fashion without depending on
ai-edge-model-explorer-adapter(which bundles cpp extensions and extra dependencies). Essentially one could write a method that looks likeAdapter.convertwithout actually importingAdapter, and pre-process their data to create a model explorer graph that can be loaded by the frontend. This would give users the choice of i) writing a lightweight conversion function and using the default support in the frontend component, or ii) importingAdapter, and getting the full functionality of a custom adapter.