Hey folks! Potentially silly question: Suppose I ...
# flyte-support
i
Hey folks! Potentially silly question: Suppose I have an extremely capable headless desktop with a bunch of GPUs and memory. What would be the best way to execute a flyte pipeline while developing from a remote laptop? In other words, does flyte include a native way to set up my headless desktop to receive a workflow, or should I focus on remote execution via SSH?
f
There has to be connectivity
That’s it
Flyte sandbox is all you need with gpu drivers
Some folks do that
i
Thank Ketan! Yes, Sandbox caught my eye but I wasn’t sure if that was the correct approach.
f
It stores state on a local disk so decent
You will Have to scale it up
i
No worries! I’ll be sure to grab some extra SSDs. While working with other workflow orchestrators in the past, I found it useful to set up an automated job that deletes files older than the cache staleness requirement.
g
i
Thank you @glamorous-carpet-83516 , this is extremely insightful. Yes, I would want to be able to have select tasks utilize my GPUs for model training within Flyte Sandbox. It seems that this is not yet possible, but is currently a WIP?
I would be happy to test out this feature if it is still being worked on and the functionality is not yet available in Flyte.
f
@damp-lion-88352 / @average-secretary-61436 were working on it, it supposedly works, but this is purely a community effort
d
it is possible now
you have to build the image by yourself
❤️ 1
i
This is awesome, thank you! I might reach out with some questions later.
🙏 1
d
and please remeber to rebase with latest branch