Change handler without reloading model

[sd3ntato]

🚀 The feature

When developing custom handlers, every time I change a piece of code I need to unload the model and load it back on; this proces can take quite some time.

[agunapal]

@sd3ntato you can make sure your handler works as expected by following this example . Once you have finalized your changes, you can load it in torchserve

https://github.com/pytorch/serve/blob/master/examples/image_classifier/resnet_18/debugging_backend/test_handler.py

[msaroufim]

I believe @sd3ntato is suggesting something closer to a torchserve --watch command which would be very useful honestly

It should work with a workflow when the package is just a directory