CTGAN - cuda=TRUE; multiple GPU training

[meaneych]

I am wondering whether the CTGAN model allows for training/sampling using multiple GPUs on a node.

I am training a CTGAN model (doing hyper-parameter optimization) on the following HPC/GPU cluster.
https://docs.scinet.utoronto.ca/index.php/Mist

I have access to up to 4 GPUs per node. If possible I wonder if I can exploit the multiple GPU availability in CTGAN??
I am using PyTorch backend.
I have moved from using CPU training (cuda=False), to single GPU training (cuda=True) arguments.
With cuda=True, will sdv/CTGAN simply recognize the multiple GPU available, if I request them on the cluster?
Or is there another argument (or additional syntax) that needs to be used to enable multi-GPU training/sampling?

Thanks in advance for the advice!
Chris

Environment details

If you are already running SDV, please indicate the following details about the environment in
which you are running it:

• SDV version: 0.18
• Python version: 3.9
• Operating System: RedHat Enterprise Linux 8.2

Problem description

<Replace this with a description of the problem that you are trying to solve using SDV. If
possible, describe the data that you are using, or consider attaching some example data
that others can use to propose a working solution for your problem.>

What I already tried

<Replace with a description of what you already tried and what is the behavior that you observe.
If possible, also add below the exact code that you are running.>

Paste the command(s) you ran and the output.
If there was a crash, please include the traceback here.

[npatki]

Hi @meaneych, nice to meet you. At the moment, the CTGANSynthesizer supports training on a single GPU though we have an outstanding feature request for supporting multi GPU training.

We prioritize new features based on user demand and importance to you. So if you are able to provide more information about your use case (what kind of data are you using, how are you planning to use and evaluate the synthetic data once it's created, etc.), that would be very helpful. Thanks!

[npatki]

Hello, I'm closing off this issue as a duplicate in favor of the original CTGAN issue: sdv-dev/CTGAN#290

If there is more to discuss, please feel free to reply and we can always reopen the issue.