Because all my config will be in one place (my training script). I'd like the zipped runs to be at a specific place on the shared disk. Otherwise I'll have to manually copy/paste them so they are visible from the other node (the one running the server)

Related, how to I specify in my code the cache_dir where the zip is saved?

Thanks, Martin.

Seems to work. Although if I import the same run at two different stages (because I want to see how it's doing while training, not only at the end). it doesn't recognise it's the same task/run just with more iterations. It just creates a new task...

Just to clarify, where do I run the second command?

Thanks. On the second one, can I specify it my code? (I'd prefer to avoid a separate trains.conf file if possible)

Is there a way to automatically deduplicate? I'm guessing tasks have some unique id in the zip (maybe even the file name)

E.g. this one https://github.com/allegroai/trains/blob/master/docs/trains.conf

So copy train.conf from elsewhere to ~/train.conf?

Thanks. let me try it...

I see that to create a trains.conf file I need to run trains-init which requires a browser. What if I can't open a browser on that machine?