Merge pull request #40 from apoorvkh/argument-class

Change `launch` to `Launcher` dataclass

Author: pmcurtin

.github/workflows/docs-preview.yml

@@ -1,8 +1,6 @@
 name: Readthedocs preview
 on:
   pull_request_target:
-    types:
-      - opened
     paths:
       - "docs/**"
 

docs/source/api.rst

@@ -1,5 +1,5 @@
-torchrunx API
+API
 =============
 
 .. automodule:: torchrunx
-    :members:
+    :members:

src/torchrunx/__init__.py

@@ -1,45 +1,4 @@
-from __future__ import annotations
+from .launcher import Launcher, launch
+from .slurm import slurm_hosts, slurm_workers
 
-from .launcher import launch
-
-
-def slurm_hosts() -> list[str]:
-    """Retrieves hostnames of Slurm-allocated nodes.
-
-    :return: Hostnames of nodes in current Slurm allocation
-    :rtype: list[str]
-    """
-    import os
-    import subprocess
-
-    # TODO: sanity check SLURM variables, commands
-    assert "SLURM_JOB_ID" in os.environ
-    return (
-        subprocess.check_output(["scontrol", "show", "hostnames", os.environ["SLURM_JOB_NODELIST"]])
-        .decode()
-        .strip()
-        .split("\n")
-    )
-
-
-def slurm_workers() -> int:
-    """
-    |  Determines number of workers per node in current Slurm allocation using
-    |  the ``SLURM_JOB_GPUS`` or ``SLURM_CPUS_ON_NODE`` environmental variables.
-
-    :return: The implied number of workers per node
-    :rtype: int
-    """
-    import os
-
-    # TODO: sanity check SLURM variables, commands
-    assert "SLURM_JOB_ID" in os.environ
-    if "SLURM_JOB_GPUS" in os.environ:
-        # TODO: is it possible to allocate uneven GPUs across nodes?
-        return len(os.environ["SLURM_JOB_GPUS"].split(","))
-    else:
-        # TODO: should we assume that we plan to do one worker per CPU?
-        return int(os.environ["SLURM_CPUS_ON_NODE"])
-
-
-__all__ = ["launch", "slurm_hosts", "slurm_workers"]
+__all__ = ["Launcher", "launch", "slurm_hosts", "slurm_workers"]