update dags

This commit is contained in:
2026-04-14 23:58:12 +03:00
parent a9b3696815
commit 8f3e78c16a

View File

@@ -84,6 +84,7 @@ def run_training_node_func(rank, world_size):
os.environ['WORLD_SIZE'] = str(world_size) os.environ['WORLD_SIZE'] = str(world_size)
os.environ['RANK'] = str(rank) os.environ['RANK'] = str(rank)
os.environ['NCCL_SOCKET_IFNAME'] = 'eth0' os.environ['NCCL_SOCKET_IFNAME'] = 'eth0'
os.environ['NCCL_PORT_RANGE'] = '30000-30100'
os.environ['NCCL_DEBUG'] = 'INFO' os.environ['NCCL_DEBUG'] = 'INFO'
os.environ['NCCL_TIMEOUT'] = str(NCCL_TIMEOUT) os.environ['NCCL_TIMEOUT'] = str(NCCL_TIMEOUT)
os.environ['NCCL_BLOCKING_WAIT'] = '1' os.environ['NCCL_BLOCKING_WAIT'] = '1'