Can the original Tensorflow Model architechture and code be provided?

[aamod-wick]

Is your feature request related to a problem? Please describe. There are a few things hindering modernisation of fetch to pytorch due to lack of code for the original models .Can this code be added ? It would make it extremely easy to load fetch appropriately rather than loading from onnx backend. Describe the solution you'd like Hopefully a clear file containing original model architechture written in tensorflow for atleast some of the models originally used in fetch ,

Describe alternatives you've considered An alternative is loading the conversion of the tensorflow json to onnx files ,These onnx files now can be converted using custom converters in onnx2torch to a complete pytorch model. However this model cannot be stored as oading it again would require a clear definition of its architecture in pytorch .Hence we must convert the model repeatedly from onnx to torch during inference.This causes a performance issue .

Additional context I am just hoping for the tensorflow code to create original model . It is not feasible to recreate the model by looking at the keras summary of the models due to the vast number of custom layers as shown below for one of the fetch models . Any help from the original author or from the internet community giving either a workaround or the model code is welcome.