def _run(args):
# If LSF is used, use default values from job config
if lsf.LSFUtils.using_lsf():
if not args.np:
args.np = lsf.LSFUtils.get_num_processes()
if not args.hosts and not args.hostfile and not args.host_discovery_script:
args.hosts = ','.join('{host}:{np}'.format(host=host, np=lsf.LSFUtils.get_num_gpus())
for host in lsf.LSFUtils.get_compute_hosts())
# if hosts are not specified, either parse from hostfile, or default as
# localhost
if not args.hosts and not args.host_discovery_script:
if args.hostfile:
args.hosts = hosts.parse_host_files(args.hostfile)
else:
# Set hosts to localhost if not specified
args.hosts = 'localhost:{np}'.format(np=args.np)
# Convert nics into set
args.nics = set(args.nics.split(',')) if args.nics else None
if _is_elastic(args):
return _run_elastic(args)
else:
return _run_static(args)
def test_horovodrun_hostfile(self):
with temppath() as host_filename:
with open(host_filename, 'w+') as fp:
fp.write('172.31.32.7 slots=8\n')
fp.write('172.31.33.9 slots=8\n')
hostnames = hosts.parse_host_files(host_filename)
self.assertEqual(hostnames, '172.31.32.7:8,172.31.33.9:8')
def _run(args):
# If LSF is used, use default values from job config
if lsf.LSFUtils.using_lsf():
if not args.num_proc:
args.num_proc = lsf.LSFUtils.get_num_processes()
if not args.hosts and not args.hostfile and not args.host_discovery_script:
args.hosts = ','.join(f'{host}:{lsf.LSFUtils.get_num_gpus()}'
for host in lsf.LSFUtils.get_compute_hosts())
# if hosts are not specified, either parse from hostfile, or default as
# localhost
if not args.hosts and not args.host_discovery_script:
if args.hostfile:
args.hosts = hosts.parse_host_files(args.hostfile)
else:
# Set hosts to localhost if not specified
args.hosts = f'localhost:{args.num_proc}'
if _is_elastic(args):
return _run_elastic(args)
else:
return _run_static(args)