You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

81 lines
3.1 KiB

#!/usr/bin/python3
# Copyright (c) ByteDance, Inc. and its affiliates.
# All rights reserved.
#
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
import argparse
import functools
import os
import socket
import subprocess
import sys
from typing import List
os_system = functools.partial(subprocess.call, shell=True)
echo = lambda info: os_system(f'echo "[$(date "+%m-%d-%H:%M:%S")] ({os.path.basename(sys._getframe().f_back.f_code.co_filename)}, line{sys._getframe().f_back.f_lineno})=> {info}"')
def __find_free_port():
sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
sock.bind(("", 0))
port = sock.getsockname()[1]
sock.close()
return port
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='PyTorch Distributed Launcher')
parser.add_argument('--main_py_relpath', type=str, default='main.py',
help='specify launcher script.')
# distributed environment
parser.add_argument('--num_nodes', type=int, default=1)
parser.add_argument('--ngpu_per_node', type=int, default=1)
parser.add_argument('--node_rank', type=int, default=0,
help='node rank, ranged from 0 to [dist_num_nodes]-1')
parser.add_argument('--master_address', type=str, default='128.0.0.0',
help='master address for distributed communication')
parser.add_argument('--master_port', type=int, default=30001,
help='master port for distributed communication')
args_for_this, args_for_python = parser.parse_known_args()
args_for_python: List[str]
echo(f'[initial args_for_python]: {args_for_python}')
# auto-complete: update args like `--sbn` to `--sbn=1`
kwargs = args_for_python[-1]
kwargs = '='.join(map(str.strip, kwargs.split('=')))
kwargs = kwargs.split(' ')
for i, a in enumerate(kwargs):
if len(a) and '=' not in a:
kwargs[i] = f'{a}=1'
args_for_python[-1] = ' '.join(kwargs)
echo(f'[final args_for_python]: {args_for_python}')
if args_for_this.num_nodes > 1: # distributed
os.environ['NPROC_PER_NODE'] = str(args_for_this.ngpu_per_node)
cmd = (
f'python3 -m torch.distributed.launch'
f' --nnodes={args_for_this.num_nodes}'
f' --nproc_per_node={args_for_this.ngpu_per_node}'
f' --node_rank={args_for_this.node_rank}'
f' --master_addr={args_for_this.master_address}'
f' --master_port={args_for_this.master_port}'
f' {args_for_this.main_py_relpath}'
f' {" ".join(args_for_python)}'
)
else: # single machine with multiple GPUs
cmd = (
f'python3 -m torch.distributed.launch'
f' --nproc_per_node={args_for_this.ngpu_per_node}'
f' --master_port={__find_free_port()}'
f' {args_for_this.main_py_relpath}'
f' {" ".join(args_for_python)}'
)
exit_code = subprocess.call(cmd, shell=True)
sys.exit(exit_code)