未验证 提交 cf180c1f 编写于 作者: K Kaipeng Deng 提交者: GitHub

fix training hang in PaddleCloud single machine. (#1233)

上级 1cc363f6
...@@ -53,7 +53,9 @@ logger = logging.getLogger(__name__) ...@@ -53,7 +53,9 @@ logger = logging.getLogger(__name__)
def main(): def main():
env = os.environ env = os.environ
FLAGS.dist = 'PADDLE_TRAINER_ID' in env and 'PADDLE_TRAINERS_NUM' in env FLAGS.dist = 'PADDLE_TRAINER_ID' in env \
and 'PADDLE_TRAINERS_NUM' in env \
and int(env['PADDLE_TRAINERS_NUM']) > 1
if FLAGS.dist: if FLAGS.dist:
trainer_id = int(env['PADDLE_TRAINER_ID']) trainer_id = int(env['PADDLE_TRAINER_ID'])
local_seed = (99 + trainer_id) local_seed = (99 + trainer_id)
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册