How is TorchScript support in FastAI?
We’re using PyTorch Lightning, and so far it’s suitable for longer research projects. However, I want to speed up making PoCs (Proof of Concepts), which I think FastAI is more suitable for.
The deployment infrastructure however relies on TorchScript (because there is also C++ involved, and we just have to pass this file to deployment team to get the model running).
I know that FastAIs standpoint is that Julia (or was it Swift?) is better suited for this, instead of limiting Python’s structure to remove the GIL, but that would be a too big project to undertake, and I’m not familiar with Julia/Swift.
In PyTorch Lightning, you can get a TorchScript version of the model in 1 line of code:
What would I need to do in FastAI?
p.s. I searched this forum, but only a few topics touched on TorchScript, but none discussed it in detail.