mirror of
https://github.com/wassname/pytorch-lightning.git
synced 2026-09-20 13:10:42 +08:00
fixed nccl init
This commit is contained in:
@@ -375,19 +375,23 @@ class Trainer(TrainerIO):
|
||||
:param tries:
|
||||
:return:
|
||||
"""
|
||||
if tries > 20:
|
||||
raise RuntimeError('Failed to connect using 20 different ip addresses')
|
||||
|
||||
try:
|
||||
root_node = os.environ['SLURM_NODELIST'].split(' ')[0]
|
||||
os.environ['MASTER_ADDR'] = root_node
|
||||
os.environ['MASTER_PORT'] = f'{port}'
|
||||
dist.init_process_group("nccl", rank=self.proc_rank, world_size=self.world_size)
|
||||
|
||||
except RuntimeError as e:
|
||||
# port taken
|
||||
warnings.warn(f'port {port} taken, trying port {port}...')
|
||||
self.__init_tcp_connection(port + 1, tries + 1)
|
||||
root_node = os.environ['SLURM_NODELIST'].split(' ')[0]
|
||||
os.environ['MASTER_ADDR'] = root_node
|
||||
os.environ['MASTER_PORT'] = f'{port}'
|
||||
dist.init_process_group("nccl", rank=self.proc_rank, world_size=self.world_size)
|
||||
# if tries > 20:
|
||||
# raise RuntimeError('Failed to connect using 20 different ip addresses')
|
||||
#
|
||||
# try:
|
||||
# root_node = os.environ['SLURM_NODELIST'].split(' ')[0]
|
||||
# os.environ['MASTER_ADDR'] = root_node
|
||||
# os.environ['MASTER_PORT'] = f'{port}'
|
||||
# dist.init_process_group("nccl", rank=self.proc_rank, world_size=self.world_size)
|
||||
#
|
||||
# except RuntimeError as e:
|
||||
# # port taken
|
||||
# warnings.warn(f'port {port} taken, trying port {port}...')
|
||||
# self.__init_tcp_connection(port + 1, tries + 1)
|
||||
|
||||
def __run_pretrain_routine(self, model):
|
||||
"""
|
||||
|
||||
Reference in New Issue
Block a user