Hello guys, I was going through the related PR that was attached with the TypeTransformer plugin for

tf.tensor

issue. I'd a doubt, what does it mean when it was said that "The plugin could specify an unpinned

torch

as a dependency". I have the same doubt that was raised by @tall-lock-23197. How do we prevent re-installing

torch

? Can someone explain me?

Hi, @billowy-continent-33554! To use the Flyte PyTorch type, PyTorch needs to be installed beforehand. If not, the plugin wouldn’t work. That discussion was about having a standalone plugins vs. having it in the core library. Since it’s now a part of the core library, you needn’t worry about it.

Thanks for clarifying. I guess

tensorflow

too is a part of the core library now? I've not checked it manually

No, it isn’t. I’ve created some issues to add support for TensorFlow: https://github.com/flyteorg/flyte/issues/2917 (see the flytekit section).

Yeah, my bad. So as a general rule, if someone wants to write a plugin, would it be good to include the required libraries first in the core library?

This plugin is in the core library, but not all of them are. Also, we haven’t included the dependencies in the core library. If you install the dependency, this plugin gets activated. In the PyTorch type case, it’s the

torch

library.