*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
[2021-05-26 04:37:41,622] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:41,622] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:41,633] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:41,636] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:41,975] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:41,989] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:41,993] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:41,995] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:41,995] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:41,997] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,005] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,005] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,008] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,011] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,011] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,013] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,013] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,018] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,021] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,025] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,025] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,027] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,028] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,032] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,033] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,033] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,040] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,041] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,041] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,042] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,042] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,044] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,048] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,049] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,049] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,050] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,052] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,054] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,056] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,060] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,060] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,067] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,068] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,072] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,080] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,093] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,094] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,096] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,099] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,102] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,107] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,108] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,144] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,144] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,151] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,157] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,353] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,353] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,353] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:42,353] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:43,820] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:43,820] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:43,820] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:37:43,820] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
nn.functional.linear has been overridden with a more memory efficient version. This will persist unless manually reset.
[2021-05-26 04:37:51,493] [INFO] [logging.py:60:log_dist] [Rank 0] DeepSpeed info: version=0.3.17+unknown, git-hash=unknown, git-branch=unknown
[2021-05-26 04:37:51,549] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,550] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,550] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,550] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,551] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,552] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,552] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,552] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,552] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,552] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,553] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,553] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,553] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,553] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,554] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,554] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,554] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,554] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,555] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,555] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,555] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,556] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,557] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,557] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,557] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,557] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,557] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,558] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,558] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,560] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,560] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,560] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,560] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,561] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,561] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,561] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,561] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,561] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,561] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,562] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,565] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,565] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,566] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,566] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,567] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,567] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,567] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,568] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,568] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,569] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,569] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,570] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,571] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,572] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,572] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,573] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,573] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,573] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,574] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,575] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,683] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,684] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,684] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:51,687] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:37:52,017] [INFO] [engine.py:164:__init__] DeepSpeed Flops Profiler Enabled: False
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
[2021-05-26 04:37:53,035] [INFO] [engine.py:636:_configure_optimizer] Using DeepSpeed Optimizer param name adamw as basic optimizer
[2021-05-26 04:37:53,035] [INFO] [engine.py:641:_configure_optimizer] DeepSpeed Basic Optimizer = DeepSpeedCPUAdam
Checking ZeRO support for optimizer=DeepSpeedCPUAdam type=<class 'deepspeed.ops.adam.cpu_adam.DeepSpeedCPUAdam'>
[2021-05-26 04:37:53,035] [INFO] [logging.py:60:log_dist] [Rank 0] Creating fp16 ZeRO stage 3 optimizer
Initializing ZeRO Stage 3
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
[2021-05-26 04:37:53,113] [INFO] [utils.py:588:see_memory_usage] Stage 3 initialize beginning
[2021-05-26 04:37:53,114] [INFO] [utils.py:589:see_memory_usage] MA 2.34 GB         Max_MA 3.81 GB         CA 5.4 GB         Max_CA 5 GB 
[2021-05-26 04:37:53,115] [INFO] [utils.py:597:see_memory_usage] CPU Virtual Memory:  used = 40.02 GB, percent = 21.4%
[2021-05-26 04:37:53,115] [INFO] [stage3.py:624:__init__] Reduce bucket size 67108864
[2021-05-26 04:37:53,115] [INFO] [stage3.py:625:__init__] Allgather bucket size 60397977.6
[2021-05-26 04:37:53,128] [INFO] [stage3.py:39:print_rank_0] FP16 params swapping is False, Max params in CPU is 1000000000.0
[2021-05-26 04:37:53,191] [INFO] [utils.py:588:see_memory_usage] Before creating fp16 partitions
[2021-05-26 04:37:53,192] [INFO] [utils.py:589:see_memory_usage] MA 2.34 GB         Max_MA 2.34 GB         CA 5.4 GB         Max_CA 5 GB 
[2021-05-26 04:37:53,192] [INFO] [utils.py:597:see_memory_usage] CPU Virtual Memory:  used = 40.02 GB, percent = 21.4%
[2021-05-26 04:37:54,059] [INFO] [stage3.py:39:print_rank_0] fp16 group 0 has 1 subgroups
[2021-05-26 04:37:55,731] [INFO] [stage3.py:39:print_rank_0] Swappable FP32 Partitions: count=0 size= 0.00 GB
[2021-05-26 04:37:55,731] [INFO] [stage3.py:39:print_rank_0] In-Memory FP32 Partitions: count=1 size= 3.02 GB
[2021-05-26 04:38:02,821] [INFO] [stage3.py:819:__init__] optimizer state initialized
[2021-05-26 04:38:02,822] [INFO] [stage3.py:39:print_rank_0] Largest partitioned param numel = 811977088
[2021-05-26 04:38:20,030] [INFO] [utils.py:588:see_memory_usage] After initializing ZeRO optimizer
[2021-05-26 04:38:20,031] [INFO] [utils.py:589:see_memory_usage] MA 3.09 GB         Max_MA 4.09 GB         CA 6.92 GB         Max_CA 7 GB 
[2021-05-26 04:38:20,031] [INFO] [utils.py:597:see_memory_usage] CPU Virtual Memory:  used = 94.98 GB, percent = 50.7%
[2021-05-26 04:38:20,031] [INFO] [logging.py:60:log_dist] [Rank 0] DeepSpeed Final Optimizer = adamw
[2021-05-26 04:38:20,031] [INFO] [engine.py:449:_configure_lr_scheduler] DeepSpeed using configured LR scheduler = WarmupLR
[2021-05-26 04:38:20,031] [INFO] [logging.py:60:log_dist] [Rank 0] DeepSpeed LR Scheduler = <deepspeed.runtime.lr_schedules.WarmupLR object at 0x14b579df9a00>
[2021-05-26 04:38:20,031] [INFO] [logging.py:60:log_dist] [Rank 0] step=0, skipped=0, lr=[5e-05], mom=[[0.9, 0.999]]
[2021-05-26 04:38:20,031] [INFO] [config.py:748:print] DeepSpeedEngine configuration:
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   activation_checkpointing_config  {
    "partition_activations": false, 
    "contiguous_memory_optimization": false, 
    "cpu_checkpointing": false, 
    "number_checkpoints": null, 
    "synchronize_checkpoint_boundary": false, 
    "profile": false
}
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   aio_config ................... {'block_size': 1048576, 'queue_depth': 8, 'thread_count': 1, 'single_submit': False, 'overlap_events': True}
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   allreduce_always_fp32 ........ False
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   amp_enabled .................. False
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   amp_params ................... False
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   checkpoint_tag_validation_enabled  True
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   checkpoint_tag_validation_fail  False
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   disable_allgather ............ False
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   dump_state ................... False
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   dynamic_loss_scale_args ...... {'init_scale': 256, 'scale_window': 1000, 'delayed_shift': 2, 'min_scale': 1}
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   elasticity_enabled ........... False
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   flops_profiler_config ........ {
    "enabled": false, 
    "profile_step": 1, 
    "module_depth": -1, 
    "top_modules": 1, 
    "detailed": true, 
    "output_file": null
}
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   fp16_enabled ................. True
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   global_rank .................. 0
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   gradient_accumulation_steps .. 1
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   gradient_clipping ............ 1.0
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   gradient_predivide_factor .... 1.0
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   initial_dynamic_scale ........ 256
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   loss_scale ................... 0
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   memory_breakdown ............. False
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   optimizer_legacy_fusion ...... False
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   optimizer_name ............... adamw
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   optimizer_params ............. {'lr': 5e-05, 'betas': [0.9, 0.999], 'eps': 1e-08, 'weight_decay': 0.0}
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   pipeline ..................... {'stages': 'auto', 'partition': 'best', 'seed_layers': False, 'activation_checkpoint_interval': 0}
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   pld_enabled .................. False
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   pld_params ................... False
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   prescale_gradients ........... False
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   scheduler_name ............... WarmupLR
[2021-05-26 04:38:20,032] [INFO] [config.py:752:print]   scheduler_params ............. {'warmup_min_lr': 0, 'warmup_max_lr': 5e-05, 'warmup_num_steps': 8}
[2021-05-26 04:38:20,033] [INFO] [config.py:752:print]   sparse_attention ............. None
[2021-05-26 04:38:20,033] [INFO] [config.py:752:print]   sparse_gradients_enabled ..... False
[2021-05-26 04:38:20,033] [INFO] [config.py:752:print]   steps_per_print .............. 2000
[2021-05-26 04:38:20,033] [INFO] [config.py:752:print]   tensorboard_enabled .......... False
[2021-05-26 04:38:20,033] [INFO] [config.py:752:print]   tensorboard_job_name ......... DeepSpeedJobName
[2021-05-26 04:38:20,033] [INFO] [config.py:752:print]   tensorboard_output_path ...... 
[2021-05-26 04:38:20,033] [INFO] [config.py:752:print]   train_batch_size ............. 2048
[2021-05-26 04:38:20,033] [INFO] [config.py:752:print]   train_micro_batch_size_per_gpu  32
[2021-05-26 04:38:20,033] [INFO] [config.py:752:print]   wall_clock_breakdown ......... False
[2021-05-26 04:38:20,033] [INFO] [config.py:752:print]   world_size ................... 64
[2021-05-26 04:38:20,033] [INFO] [config.py:752:print]   zero_allow_untested_optimizer  False
[2021-05-26 04:38:20,033] [INFO] [config.py:752:print]   zero_config .................. {
    "stage": 3, 
    "contiguous_gradients": true, 
    "reduce_scatter": false, 
    "reduce_bucket_size": 6.710886e+07, 
    "allgather_partitions": true, 
    "allgather_bucket_size": 5.000000e+08, 
    "overlap_comm": true, 
    "load_from_fp32_weights": true, 
    "elastic_checkpoint": true, 
    "offload_param": {
        "device": "cpu", 
        "nvme_path": null, 
        "buffer_count": 5, 
        "buffer_size": 1.000000e+08, 
        "max_in_cpu": 1.000000e+09, 
        "pin_memory": true
    }, 
    "offload_optimizer": {
        "device": "cpu", 
        "nvme_path": null, 
        "buffer_count": 4, 
        "pin_memory": true, 
        "pipeline_read": false, 
        "pipeline_write": false, 
        "fast_init": false, 
        "pipeline": false
    }, 
    "sub_group_size": 1.000000e+14, 
    "prefetch_bucket_size": 6.039798e+07, 
    "param_persistence_threshold": 8.192000e+04, 
    "max_live_parameters": 1.000000e+09, 
    "max_reuse_distance": 1.000000e+09, 
    "gather_fp16_weights_on_model_save": false, 
    "ignore_unused_parameters": true
}
[2021-05-26 04:38:20,033] [INFO] [config.py:752:print]   zero_enabled ................. True
[2021-05-26 04:38:20,033] [INFO] [config.py:752:print]   zero_optimization_stage ...... 3
[2021-05-26 04:38:20,033] [INFO] [config.py:754:print]   json = {
    "fp16": {
        "enabled": true, 
        "loss_scale": 0, 
        "loss_scale_window": 1000, 
        "initial_scale_power": 8, 
        "hysteresis": 2, 
        "min_loss_scale": 1
    }, 
    "optimizer": {
        "type": "AdamW", 
        "params": {
            "lr": 5e-05, 
            "betas": [0.9, 0.999], 
            "eps": 1e-08, 
            "weight_decay": 0.0
        }
    }, 
    "scheduler": {
        "type": "WarmupLR", 
        "params": {
            "warmup_min_lr": 0, 
            "warmup_max_lr": 5e-05, 
            "warmup_num_steps": 8
        }
    }, 
    "zero_optimization": {
        "stage": 3, 
        "offload_optimizer": {
            "device": "cpu", 
            "pin_memory": true
        }, 
        "offload_param": {
            "device": "cpu", 
            "pin_memory": true
        }, 
        "overlap_comm": true, 
        "contiguous_gradients": true, 
        "sub_group_size": 1.000000e+14, 
        "reduce_bucket_size": 6.710886e+07, 
        "stage3_prefetch_bucket_size": 6.039798e+07, 
        "stage3_param_persistence_threshold": 8.192000e+04, 
        "stage3_max_live_parameters": 1.000000e+09, 
        "stage3_max_reuse_distance": 1.000000e+09, 
        "stage3_gather_fp16_weights_on_model_save": false
    }, 
    "gradient_accumulation_steps": 1, 
    "gradient_clipping": 1.0, 
    "steps_per_print": 2.000000e+03, 
    "train_batch_size": 2.048000e+03, 
    "train_micro_batch_size_per_gpu": 32, 
    "wall_clock_breakdown": false
}
Killing subprocess 22920
Killing subprocess 22921
Killing subprocess 22922
Killing subprocess 22923
Killing subprocess 63213
Killing subprocess 63214
Killing subprocess 63215
Killing subprocess 63216
Killing subprocess 46668
Killing subprocess 46669
Killing subprocess 46670
Killing subprocess 46671
Killing subprocess 67000
Killing subprocess 67001
Killing subprocess 67002
Killing subprocess 67003
Killing subprocess 13436
Killing subprocess 13437
Killing subprocess 13438
Killing subprocess 13439
Killing subprocess 12298
Killing subprocess 12299
Killing subprocess 12300
Killing subprocess 12301
Killing subprocess 9315
Killing subprocess 9316
Killing subprocess 9317
Killing subprocess 9318
Killing subprocess 28132
Killing subprocess 28133
Killing subprocess 28134
Killing subprocess 28135
Killing subprocess 50556
Killing subprocess 50557
Killing subprocess 50558
Killing subprocess 50559
Killing subprocess 13485
Killing subprocess 13486
Killing subprocess 13487
Killing subprocess 13488
Killing subprocess 33836
Killing subprocess 33837
Killing subprocess 33838
Killing subprocess 33839
Killing subprocess 59977
Killing subprocess 59978
Killing subprocess 59979
Killing subprocess 59980
Killing subprocess 21554
Killing subprocess 21555
Killing subprocess 21556
Killing subprocess 21557
Killing subprocess 75475
Killing subprocess 75476
Killing subprocess 75477
Killing subprocess 75478
Killing subprocess 49793
Killing subprocess 49794
Killing subprocess 49795
Killing subprocess 49796
Killing subprocess 62660
Killing subprocess 62661
Killing subprocess 62662
Killing subprocess 62663
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
[2021-05-26 04:40:19,839] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:19,839] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,034] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,036] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,052] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,053] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,057] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,059] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,063] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,072] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,072] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,072] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,073] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,074] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,074] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,074] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,080] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,081] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,082] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,083] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,083] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,086] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,089] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,094] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,097] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,097] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,099] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,101] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,101] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,103] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,115] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,116] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,122] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,123] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,124] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,124] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,125] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,125] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,127] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,128] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,129] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,140] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,141] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,143] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,143] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,144] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,153] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,153] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,202] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,205] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,211] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,215] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,216] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,229] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,239] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,239] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,415] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,418] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,427] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:20,433] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:21,852] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:21,856] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:22,741] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:40:22,761] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
nn.functional.linear has been overridden with a more memory efficient version. This will persist unless manually reset.
[2021-05-26 04:40:28,842] [INFO] [logging.py:60:log_dist] [Rank 0] DeepSpeed info: version=0.3.17+unknown, git-hash=unknown, git-branch=unknown
[2021-05-26 04:40:28,940] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,940] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,945] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,946] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,946] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,946] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,946] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,947] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,947] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,947] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,947] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,947] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,947] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,947] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,948] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,948] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,949] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,949] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,949] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,950] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,950] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,950] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,950] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,951] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,951] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,951] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,951] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,952] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,952] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,952] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,952] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,953] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,953] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,953] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,953] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,953] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,954] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,954] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,954] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,954] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,954] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,954] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,954] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,955] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,955] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,955] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,956] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,956] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,956] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,957] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,957] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,957] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,958] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,958] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,958] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,958] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,959] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,960] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,960] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,960] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,960] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,961] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,961] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:28,963] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:40:29,315] [INFO] [engine.py:164:__init__] DeepSpeed Flops Profiler Enabled: False
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
[2021-05-26 04:40:30,228] [INFO] [engine.py:636:_configure_optimizer] Using DeepSpeed Optimizer param name adamw as basic optimizer
[2021-05-26 04:40:30,228] [INFO] [engine.py:641:_configure_optimizer] DeepSpeed Basic Optimizer = DeepSpeedCPUAdam
Checking ZeRO support for optimizer=DeepSpeedCPUAdam type=<class 'deepspeed.ops.adam.cpu_adam.DeepSpeedCPUAdam'>
[2021-05-26 04:40:30,228] [INFO] [logging.py:60:log_dist] [Rank 0] Creating fp16 ZeRO stage 3 optimizer
Initializing ZeRO Stage 3
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
[2021-05-26 04:40:30,309] [INFO] [utils.py:588:see_memory_usage] Stage 3 initialize beginning
[2021-05-26 04:40:30,311] [INFO] [utils.py:589:see_memory_usage] MA 2.34 GB         Max_MA 3.81 GB         CA 5.4 GB         Max_CA 5 GB 
[2021-05-26 04:40:30,311] [INFO] [utils.py:597:see_memory_usage] CPU Virtual Memory:  used = 40.02 GB, percent = 21.4%
[2021-05-26 04:40:30,311] [INFO] [stage3.py:624:__init__] Reduce bucket size 67108864
[2021-05-26 04:40:30,311] [INFO] [stage3.py:625:__init__] Allgather bucket size 60397977.6
[2021-05-26 04:40:30,325] [INFO] [stage3.py:39:print_rank_0] FP16 params swapping is False, Max params in CPU is 1000000000.0
[2021-05-26 04:40:30,392] [INFO] [utils.py:588:see_memory_usage] Before creating fp16 partitions
[2021-05-26 04:40:30,392] [INFO] [utils.py:589:see_memory_usage] MA 2.34 GB         Max_MA 2.34 GB         CA 5.4 GB         Max_CA 5 GB 
[2021-05-26 04:40:30,393] [INFO] [utils.py:597:see_memory_usage] CPU Virtual Memory:  used = 40.02 GB, percent = 21.4%
[2021-05-26 04:40:31,251] [INFO] [stage3.py:39:print_rank_0] fp16 group 0 has 1 subgroups
[2021-05-26 04:40:32,871] [INFO] [stage3.py:39:print_rank_0] Swappable FP32 Partitions: count=0 size= 0.00 GB
[2021-05-26 04:40:32,871] [INFO] [stage3.py:39:print_rank_0] In-Memory FP32 Partitions: count=1 size= 3.02 GB
[2021-05-26 04:40:39,984] [INFO] [stage3.py:819:__init__] optimizer state initialized
[2021-05-26 04:40:39,984] [INFO] [stage3.py:39:print_rank_0] Largest partitioned param numel = 811977088
[2021-05-26 04:40:56,470] [INFO] [utils.py:588:see_memory_usage] After initializing ZeRO optimizer
[2021-05-26 04:40:56,471] [INFO] [utils.py:589:see_memory_usage] MA 3.09 GB         Max_MA 4.09 GB         CA 6.92 GB         Max_CA 7 GB 
[2021-05-26 04:40:56,471] [INFO] [utils.py:597:see_memory_usage] CPU Virtual Memory:  used = 94.99 GB, percent = 50.7%
[2021-05-26 04:40:56,472] [INFO] [logging.py:60:log_dist] [Rank 0] DeepSpeed Final Optimizer = adamw
[2021-05-26 04:40:56,472] [INFO] [engine.py:449:_configure_lr_scheduler] DeepSpeed using configured LR scheduler = WarmupLR
[2021-05-26 04:40:56,472] [INFO] [logging.py:60:log_dist] [Rank 0] DeepSpeed LR Scheduler = <deepspeed.runtime.lr_schedules.WarmupLR object at 0x15077cec1a00>
[2021-05-26 04:40:56,472] [INFO] [logging.py:60:log_dist] [Rank 0] step=0, skipped=0, lr=[5e-05], mom=[[0.9, 0.999]]
[2021-05-26 04:40:56,472] [INFO] [config.py:748:print] DeepSpeedEngine configuration:
[2021-05-26 04:40:56,472] [INFO] [config.py:752:print]   activation_checkpointing_config  {
    "partition_activations": false, 
    "contiguous_memory_optimization": false, 
    "cpu_checkpointing": false, 
    "number_checkpoints": null, 
    "synchronize_checkpoint_boundary": false, 
    "profile": false
}
[2021-05-26 04:40:56,472] [INFO] [config.py:752:print]   aio_config ................... {'block_size': 1048576, 'queue_depth': 8, 'thread_count': 1, 'single_submit': False, 'overlap_events': True}
[2021-05-26 04:40:56,472] [INFO] [config.py:752:print]   allreduce_always_fp32 ........ False
[2021-05-26 04:40:56,472] [INFO] [config.py:752:print]   amp_enabled .................. False
[2021-05-26 04:40:56,472] [INFO] [config.py:752:print]   amp_params ................... False
[2021-05-26 04:40:56,472] [INFO] [config.py:752:print]   checkpoint_tag_validation_enabled  True
[2021-05-26 04:40:56,472] [INFO] [config.py:752:print]   checkpoint_tag_validation_fail  False
[2021-05-26 04:40:56,472] [INFO] [config.py:752:print]   disable_allgather ............ False
[2021-05-26 04:40:56,472] [INFO] [config.py:752:print]   dump_state ................... False
[2021-05-26 04:40:56,472] [INFO] [config.py:752:print]   dynamic_loss_scale_args ...... {'init_scale': 256, 'scale_window': 1000, 'delayed_shift': 2, 'min_scale': 1}
[2021-05-26 04:40:56,472] [INFO] [config.py:752:print]   elasticity_enabled ........... False
[2021-05-26 04:40:56,472] [INFO] [config.py:752:print]   flops_profiler_config ........ {
    "enabled": false, 
    "profile_step": 1, 
    "module_depth": -1, 
    "top_modules": 1, 
    "detailed": true, 
    "output_file": null
}
[2021-05-26 04:40:56,472] [INFO] [config.py:752:print]   fp16_enabled ................. True
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   global_rank .................. 0
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   gradient_accumulation_steps .. 1
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   gradient_clipping ............ 1.0
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   gradient_predivide_factor .... 1.0
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   initial_dynamic_scale ........ 256
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   loss_scale ................... 0
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   memory_breakdown ............. False
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   optimizer_legacy_fusion ...... False
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   optimizer_name ............... adamw
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   optimizer_params ............. {'lr': 5e-05, 'betas': [0.9, 0.999], 'eps': 1e-08, 'weight_decay': 0.0}
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   pipeline ..................... {'stages': 'auto', 'partition': 'best', 'seed_layers': False, 'activation_checkpoint_interval': 0}
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   pld_enabled .................. False
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   pld_params ................... False
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   prescale_gradients ........... False
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   scheduler_name ............... WarmupLR
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   scheduler_params ............. {'warmup_min_lr': 0, 'warmup_max_lr': 5e-05, 'warmup_num_steps': 8}
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   sparse_attention ............. None
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   sparse_gradients_enabled ..... False
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   steps_per_print .............. 2000
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   tensorboard_enabled .......... False
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   tensorboard_job_name ......... DeepSpeedJobName
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   tensorboard_output_path ...... 
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   train_batch_size ............. 1024
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   train_micro_batch_size_per_gpu  16
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   wall_clock_breakdown ......... False
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   world_size ................... 64
[2021-05-26 04:40:56,473] [INFO] [config.py:752:print]   zero_allow_untested_optimizer  False
[2021-05-26 04:40:56,474] [INFO] [config.py:752:print]   zero_config .................. {
    "stage": 3, 
    "contiguous_gradients": true, 
    "reduce_scatter": false, 
    "reduce_bucket_size": 6.710886e+07, 
    "allgather_partitions": true, 
    "allgather_bucket_size": 5.000000e+08, 
    "overlap_comm": true, 
    "load_from_fp32_weights": true, 
    "elastic_checkpoint": true, 
    "offload_param": {
        "device": "cpu", 
        "nvme_path": null, 
        "buffer_count": 5, 
        "buffer_size": 1.000000e+08, 
        "max_in_cpu": 1.000000e+09, 
        "pin_memory": true
    }, 
    "offload_optimizer": {
        "device": "cpu", 
        "nvme_path": null, 
        "buffer_count": 4, 
        "pin_memory": true, 
        "pipeline_read": false, 
        "pipeline_write": false, 
        "fast_init": false, 
        "pipeline": false
    }, 
    "sub_group_size": 1.000000e+14, 
    "prefetch_bucket_size": 6.039798e+07, 
    "param_persistence_threshold": 8.192000e+04, 
    "max_live_parameters": 1.000000e+09, 
    "max_reuse_distance": 1.000000e+09, 
    "gather_fp16_weights_on_model_save": false, 
    "ignore_unused_parameters": true
}
[2021-05-26 04:40:56,474] [INFO] [config.py:752:print]   zero_enabled ................. True
[2021-05-26 04:40:56,474] [INFO] [config.py:752:print]   zero_optimization_stage ...... 3
[2021-05-26 04:40:56,474] [INFO] [config.py:754:print]   json = {
    "fp16": {
        "enabled": true, 
        "loss_scale": 0, 
        "loss_scale_window": 1000, 
        "initial_scale_power": 8, 
        "hysteresis": 2, 
        "min_loss_scale": 1
    }, 
    "optimizer": {
        "type": "AdamW", 
        "params": {
            "lr": 5e-05, 
            "betas": [0.9, 0.999], 
            "eps": 1e-08, 
            "weight_decay": 0.0
        }
    }, 
    "scheduler": {
        "type": "WarmupLR", 
        "params": {
            "warmup_min_lr": 0, 
            "warmup_max_lr": 5e-05, 
            "warmup_num_steps": 8
        }
    }, 
    "zero_optimization": {
        "stage": 3, 
        "offload_optimizer": {
            "device": "cpu", 
            "pin_memory": true
        }, 
        "offload_param": {
            "device": "cpu", 
            "pin_memory": true
        }, 
        "overlap_comm": true, 
        "contiguous_gradients": true, 
        "sub_group_size": 1.000000e+14, 
        "reduce_bucket_size": 6.710886e+07, 
        "stage3_prefetch_bucket_size": 6.039798e+07, 
        "stage3_param_persistence_threshold": 8.192000e+04, 
        "stage3_max_live_parameters": 1.000000e+09, 
        "stage3_max_reuse_distance": 1.000000e+09, 
        "stage3_gather_fp16_weights_on_model_save": false
    }, 
    "gradient_accumulation_steps": 1, 
    "gradient_clipping": 1.0, 
    "steps_per_print": 2.000000e+03, 
    "train_batch_size": 1.024000e+03, 
    "train_micro_batch_size_per_gpu": 16, 
    "wall_clock_breakdown": false
}
Killing subprocess 67163
Killing subprocess 67164
Killing subprocess 67166
Killing subprocess 67167
Killing subprocess 23075
Killing subprocess 23076
Killing subprocess 23077
Killing subprocess 23078
Killing subprocess 63366
Killing subprocess 63367
Killing subprocess 63368
Killing subprocess 63369
Killing subprocess 75633
Killing subprocess 75634
Killing subprocess 75635
Killing subprocess 75636
Killing subprocess 62906
Killing subprocess 62907
Killing subprocess 62908
Killing subprocess 62909
Killing subprocess 50730
Killing subprocess 50731
Killing subprocess 50732
Killing subprocess 50733
Killing subprocess 46815
Killing subprocess 46816
Killing subprocess 46817
Killing subprocess 46818
Killing subprocess 28296
Killing subprocess 28297
Killing subprocess 28298
Killing subprocess 28299
Killing subprocess 34107
Killing subprocess 34108
Killing subprocess 34109
Killing subprocess 34110
Killing subprocess 49946
Killing subprocess 49947
Killing subprocess 49948
Killing subprocess 49949
Killing subprocess 21767
Killing subprocess 21768
Killing subprocess 21769
Killing subprocess 21770
Killing subprocess 13683
Killing subprocess 13684
Killing subprocess 13685
Killing subprocess 13686
Killing subprocess 9471
Killing subprocess 9472
Killing subprocess 9473
Killing subprocess 9474
Killing subprocess 12445
Killing subprocess 12446
Killing subprocess 12447
Killing subprocess 12448
Killing subprocess 60136
Killing subprocess 60137
Killing subprocess 60138
Killing subprocess 60139
Killing subprocess 13587
Killing subprocess 13588
Killing subprocess 13589
Killing subprocess 13590
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
[2021-05-26 04:50:54,815] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,820] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,835] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,835] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,839] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,844] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,852] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,858] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,859] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,862] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,862] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,871] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,875] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,876] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,876] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,878] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,883] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,883] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,893] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,894] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,895] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,896] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,896] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,896] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,898] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,902] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,903] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,907] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,910] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,913] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,913] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,914] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,914] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,917] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,919] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,920] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,923] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,923] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,923] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,924] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,927] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,927] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,929] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,929] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,932] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,933] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,935] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,936] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,937] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,937] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,937] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,938] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,938] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,950] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,951] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,955] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,956] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,958] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:54,984] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:55,000] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:55,005] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:55,012] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:56,695] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
[2021-05-26 04:50:56,696] [INFO] [distributed.py:46:init_distributed] Initializing torch distributed with backend: nccl
nn.functional.linear has been overridden with a more memory efficient version. This will persist unless manually reset.
[2021-05-26 04:51:02,757] [INFO] [logging.py:60:log_dist] [Rank 0] DeepSpeed info: version=0.3.17+unknown, git-hash=unknown, git-branch=unknown
[2021-05-26 04:51:02,861] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,862] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,862] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,864] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,864] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,864] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,865] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,865] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,865] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,865] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,866] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,866] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,866] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,867] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,868] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,868] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,868] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,868] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,868] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,868] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,868] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,868] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,868] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,869] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,869] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,869] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,869] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,870] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,870] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,870] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,871] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,871] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,871] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,872] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,873] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,873] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,873] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,873] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,873] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,873] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,873] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,873] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,874] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,875] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,877] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,879] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:02,891] [INFO] [utils.py:11:_initialize_parameter_parallel_groups] data_parallel_size: 64, parameter_parallel_size: 64
[2021-05-26 04:51:03,229] [INFO] [engine.py:164:__init__] DeepSpeed Flops Profiler Enabled: False
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
[2021-05-26 04:51:04,136] [INFO] [engine.py:636:_configure_optimizer] Using DeepSpeed Optimizer param name adamw as basic optimizer
[2021-05-26 04:51:04,136] [INFO] [engine.py:641:_configure_optimizer] DeepSpeed Basic Optimizer = DeepSpeedCPUAdam
Checking ZeRO support for optimizer=DeepSpeedCPUAdam type=<class 'deepspeed.ops.adam.cpu_adam.DeepSpeedCPUAdam'>
[2021-05-26 04:51:04,136] [INFO] [logging.py:60:log_dist] [Rank 0] Creating fp16 ZeRO stage 3 optimizer
Initializing ZeRO Stage 3
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
[2021-05-26 04:51:04,212] [INFO] [utils.py:588:see_memory_usage] Stage 3 initialize beginning
[2021-05-26 04:51:04,213] [INFO] [utils.py:589:see_memory_usage] MA 2.34 GB         Max_MA 3.81 GB         CA 5.4 GB         Max_CA 5 GB 
[2021-05-26 04:51:04,213] [INFO] [utils.py:597:see_memory_usage] CPU Virtual Memory:  used = 40.07 GB, percent = 21.4%
[2021-05-26 04:51:04,213] [INFO] [stage3.py:624:__init__] Reduce bucket size 67108864
[2021-05-26 04:51:04,213] [INFO] [stage3.py:625:__init__] Allgather bucket size 60397977.6
Adam Optimizer #0 is created with AVX512 arithmetic capability.
Config: alpha=0.000050, betas=(0.900000, 0.999000), weight_decay=0.000000, adam_w=1
[2021-05-26 04:51:04,227] [INFO] [stage3.py:39:print_rank_0] FP16 params swapping is False, Max params in CPU is 1000000000.0
[2021-05-26 04:51:04,290] [INFO] [utils.py:588:see_memory_usage] Before creating fp16 partitions
[2021-05-26 04:51:04,291] [INFO] [utils.py:589:see_memory_usage] MA 2.34 GB         Max_MA 2.34 GB         CA 5.4 GB         Max_CA 5 GB 
[2021-05-26 04:51:04,291] [INFO] [utils.py:597:see_memory_usage] CPU Virtual Memory:  used = 40.07 GB, percent = 21.4%
[2021-05-26 04:51:05,151] [INFO] [stage3.py:39:print_rank_0] fp16 group 0 has 1 subgroups
[2021-05-26 04:51:06,824] [INFO] [stage3.py:39:print_rank_0] Swappable FP32 Partitions: count=0 size= 0.00 GB
[2021-05-26 04:51:06,824] [INFO] [stage3.py:39:print_rank_0] In-Memory FP32 Partitions: count=1 size= 3.02 GB
[2021-05-26 04:51:14,301] [INFO] [stage3.py:819:__init__] optimizer state initialized
[2021-05-26 04:51:14,302] [INFO] [stage3.py:39:print_rank_0] Largest partitioned param numel = 811977088
[2021-05-26 04:51:31,916] [INFO] [utils.py:588:see_memory_usage] After initializing ZeRO optimizer
[2021-05-26 04:51:31,917] [INFO] [utils.py:589:see_memory_usage] MA 3.09 GB         Max_MA 4.09 GB         CA 6.92 GB         Max_CA 7 GB 
[2021-05-26 04:51:31,917] [INFO] [utils.py:597:see_memory_usage] CPU Virtual Memory:  used = 95.03 GB, percent = 50.7%
[2021-05-26 04:51:31,918] [INFO] [logging.py:60:log_dist] [Rank 0] DeepSpeed Final Optimizer = adamw
[2021-05-26 04:51:31,918] [INFO] [engine.py:449:_configure_lr_scheduler] DeepSpeed using configured LR scheduler = WarmupLR
[2021-05-26 04:51:31,918] [INFO] [logging.py:60:log_dist] [Rank 0] DeepSpeed LR Scheduler = <deepspeed.runtime.lr_schedules.WarmupLR object at 0x149d553afa00>
[2021-05-26 04:51:31,918] [INFO] [logging.py:60:log_dist] [Rank 0] step=0, skipped=0, lr=[5e-05], mom=[[0.9, 0.999]]
[2021-05-26 04:51:31,918] [INFO] [config.py:748:print] DeepSpeedEngine configuration:
[2021-05-26 04:51:31,918] [INFO] [config.py:752:print]   activation_checkpointing_config  {
    "partition_activations": false, 
    "contiguous_memory_optimization": false, 
    "cpu_checkpointing": false, 
    "number_checkpoints": null, 
    "synchronize_checkpoint_boundary": false, 
    "profile": false
}
[2021-05-26 04:51:31,918] [INFO] [config.py:752:print]   aio_config ................... {'block_size': 1048576, 'queue_depth': 8, 'thread_count': 1, 'single_submit': False, 'overlap_events': True}
[2021-05-26 04:51:31,918] [INFO] [config.py:752:print]   allreduce_always_fp32 ........ False
[2021-05-26 04:51:31,918] [INFO] [config.py:752:print]   amp_enabled .................. False
[2021-05-26 04:51:31,918] [INFO] [config.py:752:print]   amp_params ................... False
[2021-05-26 04:51:31,918] [INFO] [config.py:752:print]   checkpoint_tag_validation_enabled  True
[2021-05-26 04:51:31,918] [INFO] [config.py:752:print]   checkpoint_tag_validation_fail  False
[2021-05-26 04:51:31,918] [INFO] [config.py:752:print]   disable_allgather ............ False
[2021-05-26 04:51:31,918] [INFO] [config.py:752:print]   dump_state ................... False
[2021-05-26 04:51:31,918] [INFO] [config.py:752:print]   dynamic_loss_scale_args ...... {'init_scale': 256, 'scale_window': 1000, 'delayed_shift': 2, 'min_scale': 1}
[2021-05-26 04:51:31,918] [INFO] [config.py:752:print]   elasticity_enabled ........... False
[2021-05-26 04:51:31,918] [INFO] [config.py:752:print]   flops_profiler_config ........ {
    "enabled": false, 
    "profile_step": 1, 
    "module_depth": -1, 
    "top_modules": 1, 
    "detailed": true, 
    "output_file": null
}
[2021-05-26 04:51:31,918] [INFO] [config.py:752:print]   fp16_enabled ................. True
[2021-05-26 04:51:31,918] [INFO] [config.py:752:print]   global_rank .................. 0
[2021-05-26 04:51:31,918] [INFO] [config.py:752:print]   gradient_accumulation_steps .. 1
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   gradient_clipping ............ 1.0
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   gradient_predivide_factor .... 1.0
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   initial_dynamic_scale ........ 256
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   loss_scale ................... 0
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   memory_breakdown ............. False
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   optimizer_legacy_fusion ...... False
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   optimizer_name ............... adamw
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   optimizer_params ............. {'lr': 5e-05, 'betas': [0.9, 0.999], 'eps': 1e-08, 'weight_decay': 0.0}
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   pipeline ..................... {'stages': 'auto', 'partition': 'best', 'seed_layers': False, 'activation_checkpoint_interval': 0}
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   pld_enabled .................. False
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   pld_params ................... False
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   prescale_gradients ........... False
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   scheduler_name ............... WarmupLR
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   scheduler_params ............. {'warmup_min_lr': 0, 'warmup_max_lr': 5e-05, 'warmup_num_steps': 8}
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   sparse_attention ............. None
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   sparse_gradients_enabled ..... False
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   steps_per_print .............. 2000
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   tensorboard_enabled .......... False
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   tensorboard_job_name ......... DeepSpeedJobName
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   tensorboard_output_path ...... 
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   train_batch_size ............. 512
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   train_micro_batch_size_per_gpu  8
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   wall_clock_breakdown ......... False
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   world_size ................... 64
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   zero_allow_untested_optimizer  False
[2021-05-26 04:51:31,919] [INFO] [config.py:752:print]   zero_config .................. {
    "stage": 3, 
    "contiguous_gradients": true, 
    "reduce_scatter": false, 
    "reduce_bucket_size": 6.710886e+07, 
    "allgather_partitions": true, 
    "allgather_bucket_size": 5.000000e+08, 
    "overlap_comm": true, 
    "load_from_fp32_weights": true, 
    "elastic_checkpoint": true, 
    "offload_param": {
        "device": "cpu", 
        "nvme_path": null, 
        "buffer_count": 5, 
        "buffer_size": 1.000000e+08, 
        "max_in_cpu": 1.000000e+09, 
        "pin_memory": true
    }, 
    "offload_optimizer": {
        "device": "cpu", 
        "nvme_path": null, 
        "buffer_count": 4, 
        "pin_memory": true, 
        "pipeline_read": false, 
        "pipeline_write": false, 
        "fast_init": false, 
        "pipeline": false
    }, 
    "sub_group_size": 1.000000e+14, 
    "prefetch_bucket_size": 6.039798e+07, 
    "param_persistence_threshold": 8.192000e+04, 
    "max_live_parameters": 1.000000e+09, 
    "max_reuse_distance": 1.000000e+09, 
    "gather_fp16_weights_on_model_save": false, 
    "ignore_unused_parameters": true
}
[2021-05-26 04:51:31,920] [INFO] [config.py:752:print]   zero_enabled ................. True
[2021-05-26 04:51:31,920] [INFO] [config.py:752:print]   zero_optimization_stage ...... 3
[2021-05-26 04:51:31,920] [INFO] [config.py:754:print]   json = {
    "fp16": {
        "enabled": true, 
        "loss_scale": 0, 
        "loss_scale_window": 1000, 
        "initial_scale_power": 8, 
        "hysteresis": 2, 
        "min_loss_scale": 1
    }, 
    "optimizer": {
        "type": "AdamW", 
        "params": {
            "lr": 5e-05, 
            "betas": [0.9, 0.999], 
            "eps": 1e-08, 
            "weight_decay": 0.0
        }
    }, 
    "scheduler": {
        "type": "WarmupLR", 
        "params": {
            "warmup_min_lr": 0, 
            "warmup_max_lr": 5e-05, 
            "warmup_num_steps": 8
        }
    }, 
    "zero_optimization": {
        "stage": 3, 
        "offload_optimizer": {
            "device": "cpu", 
            "pin_memory": true
        }, 
        "offload_param": {
            "device": "cpu", 
            "pin_memory": true
        }, 
        "overlap_comm": true, 
        "contiguous_gradients": true, 
        "sub_group_size": 1.000000e+14, 
        "reduce_bucket_size": 6.710886e+07, 
        "stage3_prefetch_bucket_size": 6.039798e+07, 
        "stage3_param_persistence_threshold": 8.192000e+04, 
        "stage3_max_live_parameters": 1.000000e+09, 
        "stage3_max_reuse_distance": 1.000000e+09, 
        "stage3_gather_fp16_weights_on_model_save": false
    }, 
    "gradient_accumulation_steps": 1, 
    "gradient_clipping": 1.0, 
    "steps_per_print": 2.000000e+03, 
    "train_batch_size": 512, 
    "train_micro_batch_size_per_gpu": 8, 
    "wall_clock_breakdown": false
}
[2021-05-26 04:56:04,854] [INFO] [stage3.py:2708:_overflow_clean_up] [deepscale] OVERFLOW! Rank 0 Skipping step. Attempted loss scale: 256, reducing to 256
{'train_runtime': 273.0148, 'train_samples_per_second': 3.663, 'train_steps_per_second': 0.007, 'epoch': 1.0}
{'train_runtime': 273.0144, 'train_samples_per_second': 3.663, 'train_steps_per_second': 0.007, 'epoch': 1.0}
{'train_runtime': 273.0144, 'train_samples_per_second': 3.663, 'train_steps_per_second': 0.007, 'epoch': 1.0}
{'train_runtime': 273.0145, 'train_samples_per_second': 3.663, 'train_steps_per_second': 0.007, 'epoch': 1.0}
{'train_runtime': 273.0148, 'train_samples_per_second': 3.663, 'train_steps_per_second': 0.007, 'epoch': 1.0}
{'train_runtime': 273.0146, 'train_samples_per_second': 3.663, 'train_steps_per_second': 0.007, 'epoch': 1.0}
{'train_runtime': 273.0144, 'train_samples_per_second': 3.663, 'train_steps_per_second': 0.007, 'epoch': 1.0}
{'train_runtime': 273.0145, 'train_samples_per_second': 3.663, 'train_steps_per_second': 0.007, 'epoch': 1.0}
{'train_runtime': 272.9369, 'train_samples_per_second': 3.664, 'train_steps_per_second': 0.007, 'epoch': 1.0}
{'train_runtime': 273.0143, 'train_samples_per_second': 3.663, 'train_steps_per_second': 0.007, 'epoch': 1.0}
{'train_runtime': 273.0147, 'train_samples_per_second': 3.663, 'train_steps_per_second': 0.007, 'epoch': 1.0}
{'train_runtime': 273.0148, 'train_samples_per_second': 3.663, 'train_steps_per_second': 0.007, 'epoch': 1.0}
{'train_runtime': 273.0125, 'train_samples_per_second': 3.663, 'train_steps_per_second': 0.007, 'epoch': 1.0}
{'train_runtime': 273.0143, 'train_samples_per_second': 3.663, 'train_steps_per_second': 0.007, 'epoch': 1.0}
{'train_runtime': 273.0146, 'train_samples_per_second': 3.663, 'train_steps_per_second': 0.007, 'epoch': 1.0}
{'train_runtime': 273.0149, 'train_samples_per_second': 3.663, 'train_steps_per_second': 0.007, 'epoch': 1.0}
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,861] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,862] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,863] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,864] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:04,866] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False
[2021-05-26 04:56:05,256] [INFO] [engine.py:1867:save_fp16_model] Did not save the model output_dir/pytorch_model.bin because `stage3_gather_fp16_weights_on_model_save` is False