Dear All,
I would like to ask, Is there a way to convert the PyTorch model to TensorFlow 1.x / 2.x without passing through ONNX? Because the generated graph by ONNX is not optimized, it usually adds cast
, reshape
, squeeze
, and unsqueeze
operations for the graph.
My main concern is about converting nested models that have nn.sequential(*[layers])
inside their architecture!
Best,
Samir