@@ -9,7 +9,6 @@ def ParseArguments(parseScriptArgs = None):
99 parser .add_argument ("--slurm" , action = 'store_true' , default = False ,
1010 help = "Set number of nodes, master node and "
1111 "nproc per node from slurm environment variables" )
12-
1312 # Communication
1413 parser .add_argument ("--nnodes" , type = int , default = 1 ,
1514 help = "The number of nodes to use for distributed "
@@ -18,21 +17,13 @@ def ParseArguments(parseScriptArgs = None):
1817 help = "The rank of the node for multi-node distributed "
1918 "training" )
2019 parser .add_argument ("--nproc_per_node" , type = int , default = 1 ,
21- help = "The number of processes to launch on each node, "
22- "for GPU training, this is recommended to be set "
23- "to the number of GPUs in your system so that "
24- "each process can be bound to a single GPU." )
20+ help = "The number of processes to use per node" )
2521 parser .add_argument ("--master_addr" , default = "127.0.0.1" , type = str ,
26- help = "Master node (rank 0)'s address, should be either "
27- "the IP address or the hostname of node 0, for "
28- "single node multi-proc training, the "
29- "--master_addr can simply be 127.0.0.1" )
22+ help = "ip-Address of node with rank 0, default is fine for single node" )
3023
31- # Set the port (if desired, else just keep the default)
24+ # Set the port
3225 parser .add_argument ("--master_port" , default = 29500 , type = int ,
33- help = "Master node (rank 0)'s free port that needs to "
34- "be used for communication during distributed "
35- "training" )
26+ help = "port of node with rank 0 for communication, default is fine for single node" )
3627
3728 # Additional Arguments to be parsed!
3829 parseScriptArgs (parser )
@@ -59,12 +50,6 @@ def PrepareMultiprocessing(args, func):
5950 processes = []
6051 if 'OMP_NUM_THREADS' not in os .environ and args .nproc_per_node > 1 :
6152 os .environ ["OMP_NUM_THREADS" ] = str (1 )
62- print ("*****************************************\n "
63- "Setting OMP_NUM_THREADS environment variable for each process "
64- "to be {} in default, to avoid your system being overloaded, "
65- "please further tune the variable for optimal performance in "
66- "your application as needed. \n "
67- "*****************************************" .format (os .environ ["OMP_NUM_THREADS" ]))
6853
6954 #Spawny
7055 if (world_size > 1 ):
0 commit comments