Hello, I am trying to launch a Flyte workflow to run remotely on a specific image (

pyflyte run

using the flags

--remote

and

--image <IMAGE>

), but I always get an error saying that I am missing packages, for example scikit-learn, which I indeed am missing locally. How can I get around this error, since the workflows will not be running locally? Thanks

I think you will have to create a custom docker image using dockerfile. You will have to specify your requirements.

Do you mean the image i pass with the

--image

flag? The one I am passing has all the requirements

Yeah, that's how I did it last time.

For me it is not working, it always complains that I am missing those dependencies locally, which I am since I don't intend to run them locally, only on the target docker image

pyflyte run still “compiles” your workflow and tasks locally before registering it to flyte, so i think for pyflyte run you need the requirements locally as well

Is there a workaround for not compiling the workflow locally?

Hm i never tried but maybe switching to a dynamic workflow helps? Dynamic workflows are getting compiled during run time!

Don't think that helps, still get the same error

Can you try importing dependencies within the relevant tasks, if that's a possibility?

If I move the imports inside the tasks then yes, I am able to do it. But I'd rather avoid it since importing other files from my project from inside a task would lead them not to be available at runtime when using "pyflyte run"

In that case, you can use

pyflyte register

to register the whole directory that contains the relevant files.