[Q&A] How to progress from simulation to POC

[louis-van-der-stam]

Python version (python3 -V)

3.9

NVFlare version (python3 -m pip list | grep "nvflare")

2.6.x

NVFlare branch (if running examples, please use the branch that corresponds to the NVFlare version, git branch)

No response

Operating system

MacOS 15.7

Have you successfully run any of the following examples?

• hello-numpy-sag with simulator
• hello-pt with simulator
• hello-numpy-sag with POC
• hello-pt with POC

Please describe your question

I run into the following issue and don't know what to do ...
We are trying to evolve a simulated training to run in a POC environment/setup. But need some guidance on the steps that are involved. Is there a tutorial that takes us through these steps and/or what part of the documentation provides information on this?

[louis-van-der-stam]

We are following the tutorial from https://nvflare.readthedocs.io/en/2.4/examples/hello_pt.html and have successfully run the simulation with nvflare simulator -w simulator-example/workspace -n 2 -t 2 simulator-example/hello-pt/jobs/hello-pt after copying the hello-pt folder into the simulator-example folder (as documented). But for the POC part, it says that after doing the nvflare poc prepare and nvflare poc start the command >submit_job hello-pt should start the job. But we get: /tmp/nvflare/poc/example_project/prod_00/admin@nvidia.com/startup/../transfer/hello-pt' is not a valid folder. So it appears some more copying needs to be done or ....

[YuanTingHsieh]

@louis-van-der-stam thanks for the question!

So you have a "job" already in "simulator-example/hello-pt/jobs/hello-pt"

To use in POC, yes we need to copy it over, in your case please do it like:

cp -r simulator-example/hello-pt/jobs/hello-pt /tmp/nvflare/poc/example_project/prod_00/admin@nvidia.com/transfer/

Afterwards you can use:

/tmp/nvflare/poc/example_project/prod_00/admin@nvidia.com/startup/fl_admin.sh

To launch the admin console, then you can do >submit_job hello-pt

So the gist is this job needs to be inside the "Admin Console"'s transfer folder so it can see it and submit to a system.

Note: In a real production environment, the server and clients will be distributed across different machines. Data scientists or researchers can simply use this "admin" startup kit on their local laptops to submit jobs to the system.