def run(self, wait=True, capture_error=False):
"""Run the process.
Args:
wait (bool): A boolean indicating whether to wait and check for errors.
Defaults to True.
capture_error (bool): A boolean indicating whether to direct stderr to a stream
that can later be read. Defaults to False.
Returns:
process (subprocess.Popen): The spawned process.
"""
self._setup()
cmd = self._create_command()
cmd.extend(super(SMDataParallelRunner, self)._create_command())
logging_config.log_script_invocation(cmd, self._env_vars)
if wait:
process_spawned = process.check_error(
cmd,
errors.ExecuteUserScriptError,
capture_error=capture_error,
cwd=environment.code_dir,
)
else:
process_spawned = process.create(
cmd,
errors.ExecuteUserScriptError,
capture_error=capture_error,
cwd=environment.code_dir,
)
self._tear_down()
return process_spawned
def run(self, wait=True, capture_error=False):
"""Run the process.
Args:
wait (bool): A boolean indicating whether to wait and check for errors.
Defaults to True.
capture_error (bool): A boolean indicating whether to direct stderr to a stream
that can later be read. Defaults to False.
Returns:
process (subprocess.Popen): The spawned process.
"""
self._setup()
cmd = self._create_command()
logging_config.log_script_invocation(cmd, self._env_vars)
training_env = environment.Environment()
exception_classes = get_modelparallel_exception_classes()
if wait:
process_spawned = process.check_error(
cmd,
exception_classes
if training_env.is_modelparallel_enabled
else errors.ExecuteUserScriptError,
self._processes_per_host,
capture_error=capture_error,
cwd=environment.code_dir,
)
else:
_, _, process_spawned = process.create(
cmd,
exception_classes
if training_env.is_modelparallel_enabled
else errors.ExecuteUserScriptError,
self._processes_per_host,
capture_error=capture_error,
cwd=environment.code_dir,
)
self._tear_down()
return process_spawned