loaded library: /usr/lib/x86_64-linux-gnu/libibverbs.so.1
loaded library: /usr/lib/x86_64-linux-gnu/libibverbs.so.1
------------------------ arguments ------------------------
  batches_per_epoch ............................... 5004
  channel_last .................................... False
  ddp ............................................. True
  exit_num ........................................ 300
  fuse_bn_add_relu ................................ False
  fuse_bn_relu .................................... False
  gpu_stat_file ................................... None
  grad_clipping ................................... 0.0
  graph ........................................... False
  label_smoothing ................................. 0.1
  learning_rate ................................... 0.256
  legacy_init ..................................... False
  load_path ....................................... None
  lr_decay_type ................................... cosine
  metric_local .................................... True
  metric_train_acc ................................ True
  momentum ........................................ 0.875
  nccl_fusion_max_ops ............................. 24
  nccl_fusion_threshold_mb ........................ 16
  num_classes ..................................... 1000
  num_devices_per_node ............................ 1
  num_epochs ...................................... 1
  num_nodes ....................................... 1
  ofrecord_part_num ............................... 256
  ofrecord_path ................................... /dataset/79846248
  print_interval .................................. 100
  print_timestamp ................................. False
  samples_per_epoch ............................... 1281167
  save_init ....................................... False
  save_path ....................................... None
  scale_grad ...................................... False
  skip_eval ....................................... True
  synthetic_data .................................. False
  total_batches ................................... -1
  train_batch_size ................................ 256
  train_global_batch_size ......................... 256
  use_fp16 ........................................ False
  use_gpu_decode .................................. False
  val_batch_size .................................. 50
  val_batches_per_epoch ........................... 1000
  val_global_batch_size ........................... 50
  val_samples_per_epoch ........................... 50000
  warmup_epochs ................................... 5
  weight_decay .................................... 3.0517578125e-05
  zero_init_residual .............................. True
-------------------- end of arguments ---------------------
***** Model Init *****
***** Model Init Finish, time escapled: 1.73611 s *****
W20220407 01:50:51.897284  2769 cudnn_conv_util.cpp:102] Currently available alogrithm (algo=0, require memory=0, idx=1) meeting requirments (max_workspace_size=1073741824, determinism=0) is not fastest. Fastest algorithm (3) requires memory 1520566288
[rank:0] [train], epoch: 0/1, iter: 100/5004, loss: 6.94325, lr: 0.000000, top1: 0.00051, throughput: 342.75 | 2022-04-07 01:52:04.166
timestamp, name, driver_version, utilization.gpu [%], utilization.memory [%], memory.total [MiB], memory.free [MiB], memory.used [MiB]
2022/04/07 01:52:04.247, Tesla V100-SXM2-32GB, 470.57.02, 38 %, 29 %, 32510 MiB, 5572 MiB, 26938 MiB
[rank:0] [train], epoch: 0/1, iter: 200/5004, loss: 6.94165, lr: 0.000000, top1: 0.00074, throughput: 351.68 | 2022-04-07 01:53:16.959
[rank:0] [train], epoch: 0/1, iter: 300/5004, loss: 6.94061, lr: 0.000000, top1: 0.00062, throughput: 356.57 | 2022-04-07 01:54:28.753