[WIP] Support AmSC IRI API - #439
Conversation
Co-authored-by: Edoardo Zoni <59625522+EZoni@users.noreply.github.com>
|
I confirm that, as of caefcc7, as expected I encounter the following error when I try to train an ML model with the AmSC IRI API backend: |
408006d to
a1283b7
Compare
|
For easier debugging later on, a1283b7 removed the installation of the AmSC Python client from the local source expected to contain the original |
After some debugging, I think the |
When I try to bypass the package issue described in #439 (comment) by installing from source the latest version of |
|
Here's the latest version of the script I use to authenticate through Globus: from amsc_client import Client
# Create client instance
client = Client(auth_method="globus")
# Check facility resources
nersc = client.facility("nersc")
for r in nersc.resources():
print(r.name, r.status)
# Submit a job
perlmutter = nersc.resource("compute")
run_dir = "/pscratch/sd/e/ezoni/runs/iri" # FIXME
job = perlmutter.submit(
executable="/bin/echo",
arguments=["Hello World!"],
directory=run_dir,
name="hello",
queue="debug",
account="m558",
duration=300,
nodes=1,
)
job.wait()
# Read job output via the filesystem API
home = nersc.resource("homes")
task = home.fs.head(f"{run_dir}/hello.stdout", lines=10)
task.wait()
print(task.result)I think this should save a |
Overview
Replaces #428 with further fixes and features.
Merge first
To do
Notes
Here's a list of related repository, issues, tutorials, etc.: