[QUESTION] RuntimeError: Timed out initializing process group in store based barrier on rank: 0, for key: store_based_barrier_key:1 (world_size=2, worker_count=1, timeout=0:10:00)

[JanryPei]

My question
I am trying to run Megatron multi-node on Docker.

My docker was established by the following command:

docker run -it -v ~/personal:/root/ --shm-size 128g --ipc=host --network=host --env NCCL_DEBUG=INFO --env NCCL_SOCKET_IFNAME=^docker*,lo --name container_name megatron_image_name

The pretrain.sh also had been setted like this:

GPUS_PER_NODE=1
MASTER_ADDR=10.x.x.x
MASTER_PORT=6000
NNODES=2
NODE_RANK=1
WORLD_SIZE=$(($GPUS_PER_NODE*$NNODES))
PP=1
TP=1
MICRO_BATCH_SIZE=1
GLOBAL_BATCH_SIZE=2

However, when I run the shell, the error occured:

using world size: 2, data-parallel-size: 2, tensor-model-parallel size: 1, pipeline-model-parallel size: 1 
using torch.float16 for parameters ...
------------------------ arguments ------------------------
  accumulate_allreduce_grads_in_fp32 .............. False
  adam_beta1 ...................................... 0.9
  adam_beta2 ...................................... 0.999
  adam_eps ........................................ 1e-08
  add_bias_linear ................................. True
  add_position_embedding .......................... True
  adlr_autoresume ................................. False
  adlr_autoresume_interval ........................ 1000
  apply_layernorm_1p .............................. False
  apply_query_key_layer_scaling ................... True
  apply_residual_connection_post_layernorm ........ False
  async_tensor_model_parallel_allreduce ........... True
  attention_dropout ............................... 0.1
  attention_softmax_in_fp32 ....................... False
  barrier_with_L1_time ............................ True
  bert_binary_head ................................ True
  bert_embedder_type .............................. megatron
  bert_load ....................................... None
  bf16 ............................................ False
  bias_dropout_fusion ............................. True
  bias_gelu_fusion ................................ True
  biencoder_projection_dim ........................ 0
  biencoder_shared_query_context_model ............ False
  block_data_path ................................. None
  check_for_nan_in_loss_and_grad .................. True
  classes_fraction ................................ 1.0
  clip_grad ....................................... 1.0
  consumed_train_samples .......................... 0
  consumed_valid_samples .......................... 0
  data_cache_path ................................. None
  data_parallel_random_init ....................... False
  data_parallel_size .............................. 2
  data_path ....................................... ['data/codeparrot_content_document']
  data_per_class_fraction ......................... 1.0
  data_sharding ................................... True
  dataloader_type ................................. single
  decoder_num_layers .............................. None
  decoder_seq_length .............................. None
  delay_grad_reduce ............................... True
  dino_bottleneck_size ............................ 256
  dino_freeze_last_layer .......................... 1
  dino_head_hidden_size ........................... 2048
  dino_local_crops_number ......................... 10
  dino_local_img_size ............................. 96
  dino_norm_last_layer ............................ False
  dino_teacher_temp ............................... 0.07
  dino_warmup_teacher_temp ........................ 0.04
  dino_warmup_teacher_temp_epochs ................. 30
  distribute_saved_activations .................... False
  distributed_backend ............................. nccl
  distributed_timeout_minutes ..................... 10
  embedding_path .................................. None
  empty_unused_memory_level ....................... 0
  encoder_num_layers .............................. 12
  encoder_seq_length .............................. 1024
  end_weight_decay ................................ 0.01
  eod_mask_loss ................................... False
  eval_interval ................................... 100
  eval_iters ...................................... 0
  evidence_data_path .............................. None
  exit_duration_in_mins ........................... None
  exit_interval ................................... None
  exit_on_missing_checkpoint ...................... False
  exit_signal_handler ............................. False
  expert_model_parallel_size ...................... 1
  expert_parallel ................................. False
  ffn_hidden_size ................................. 3072
  finetune ........................................ False
  fp16 ............................................ True
  fp16_lm_cross_entropy ........................... False
  fp32_residual_connection ........................ False
  fp8 ............................................. None
  fp8_amax_compute_algo ........................... most_recent
  fp8_amax_history_len ............................ 1
  fp8_interval .................................... 1
  fp8_margin ...................................... 0
  fp8_wgrad ....................................... True
  global_batch_size ............................... 2
  gradient_accumulation_fusion .................... True
  group_query_attention ........................... False
  head_lr_mult .................................... 1.0
  hidden_dropout .................................. 0.1
  hidden_size ..................................... 768
  hysteresis ...................................... 2
  ict_head_size ................................... None
  ict_load ........................................ None
  img_h ........................................... 224
  img_w ........................................... 224
  indexer_batch_size .............................. 128
  indexer_log_interval ............................ 1000
  inference_batch_times_seqlen_threshold .......... 512
  init_method_std ................................. 0.02
  init_method_xavier_uniform ...................... False
  initial_loss_scale .............................. 4294967296
  iter_per_epoch .................................. 1250
  kv_channels ..................................... 64
  lazy_mpu_init ................................... None
  load ............................................ None
  local_rank ...................................... None
  log_batch_size_to_tensorboard ................... False
  log_interval .................................... 10
  log_learning_rate_to_tensorboard ................ True
  log_loss_scale_to_tensorboard ................... True
  log_memory_to_tensorboard ....................... False
  log_num_zeros_in_grad ........................... False
  log_params_norm ................................. False
  log_timers_to_tensorboard ....................... False
  log_validation_ppl_to_tensorboard ............... False
  log_world_size_to_tensorboard ................... False
  loss_scale ...................................... None
  loss_scale_window ............................... 1000
  lr .............................................. 0.00015
  lr_decay_iters .................................. 1520000
  lr_decay_samples ................................ None
  lr_decay_style .................................. cosine
  lr_warmup_fraction .............................. 0.01
  lr_warmup_init .................................. 0.0
  lr_warmup_iters ................................. 0
  lr_warmup_samples ............................... 0
  make_vocab_size_divisible_by .................... 128
  mask_factor ..................................... 1.0
  mask_prob ....................................... 0.15
  mask_type ....................................... random
  masked_softmax_fusion ........................... True
  max_position_embeddings ......................... 1024
  max_tokens_to_oom ............................... 12000
  merge_file ...................................... data/merges.txt
  micro_batch_size ................................ 1
  min_loss_scale .................................. 1.0
  min_lr .......................................... 1e-05
  mmap_warmup ..................................... False
  model_spec ...................................... None
  no_load_optim ................................... None
  no_load_rng ..................................... None
  no_persist_layer_norm ........................... False
  no_save_optim ................................... None
  no_save_rng ..................................... None
  norm_epsilon .................................... 1e-05
  normalization ................................... LayerNorm
  num_attention_heads ............................. 12
  num_channels .................................... 3
  num_classes ..................................... 1000
  num_experts ..................................... None
  num_layers ...................................... 12
  num_layers_per_virtual_pipeline_stage ........... None
  num_query_groups ................................ 1
  num_workers ..................................... 2
  onnx_safe ....................................... None
  openai_gelu ..................................... False
  optimizer ....................................... adam
  output_bert_embeddings .......................... False
  overlap_grad_reduce ............................. False
  overlap_p2p_comm ................................ False
  override_opt_param_scheduler .................... False
  params_dtype .................................... torch.float16
  patch_dim ....................................... 16
  perform_initialization .......................... True
  pipeline_model_parallel_size .................... 1
  pipeline_model_parallel_split_rank .............. None
  position_embedding_type ......................... learned_absolute
  profile ......................................... False
  profile_ranks ................................... [0]
  profile_step_end ................................ 12
  profile_step_start .............................. 10
  query_in_block_prob ............................. 0.1
  rampup_batch_size ............................... None
  rank ............................................ 0
  recompute_granularity ........................... None
  recompute_method ................................ None
  recompute_num_layers ............................ None
  reset_attention_mask ............................ False
  reset_position_ids .............................. False
  retriever_report_topk_accuracies ................ []
  retriever_score_scaling ......................... False
  retriever_seq_length ............................ 256
  retro_add_retriever ............................. False
  retro_cyclic_train_iters ........................ None
  retro_encoder_attention_dropout ................. 0.1
  retro_encoder_hidden_dropout .................... 0.1
  retro_encoder_layers ............................ 2
  retro_num_neighbors ............................. 2
  retro_num_retrieved_chunks ...................... 2
  retro_return_doc_ids ............................ False
  retro_workdir ................................... None
  rotary_percent .................................. 1.0
  rotary_seq_len_interpolation_factor ............. None
  sample_rate ..................................... 1.0
  save ............................................ None
  save_interval ................................... None
  scatter_gather_tensors_in_pipeline .............. True
  seed ............................................ 1234
  seq_length ...................................... 1024
  sequence_parallel ............................... False
  sgd_momentum .................................... 0.9
  short_seq_prob .................................. 0.1
  skip_train ...................................... False
  split ........................................... 949,50,1
  squared_relu .................................... False
  standalone_embedding_stage ...................... False
  start_weight_decay .............................. 0.01
  swiglu .......................................... False
  swin_backbone_type .............................. tiny
  tensor_model_parallel_size ...................... 1
  tensorboard_dir ................................. None
  tensorboard_log_interval ........................ 1
  tensorboard_queue_size .......................... 1000
  test_data_path .................................. None
  timing_log_level ................................ 0
  timing_log_option ............................... minmax
  titles_data_path ................................ None
  tokenizer_model ................................. None
  tokenizer_type .................................. GPT2BPETokenizer
  train_data_path ................................. None
  train_iters ..................................... 10
  train_samples ................................... None
  transformer_impl ................................ local
  transformer_pipeline_model_parallel_size ........ 1
  untie_embeddings_and_output_weights ............. False
  use_checkpoint_args ............................. False
  use_checkpoint_opt_param_scheduler .............. False
  use_cpu_initialization .......................... None
  use_distributed_optimizer ....................... False
  use_flash_attn .................................. False
  use_mcore_models ................................ False
  use_one_sent_docs ............................... False
  use_ring_exchange_p2p ........................... False
  use_rotary_position_embeddings .................. False
  valid_data_path ................................. None
  variable_seq_lengths ............................ False
  virtual_pipeline_model_parallel_size ............ None
  vision_backbone_type ............................ vit
  vision_pretraining .............................. False
  vision_pretraining_type ......................... classify
  vocab_extra_ids ................................. 0
  vocab_file ...................................... data/vocab.json
  vocab_size ...................................... None
  wandb_exp_name .................................. 
  wandb_project ................................... 
  wandb_save_dir .................................. 
  weight_decay .................................... 0.01
  weight_decay_incr_style ......................... constant
  world_size ...................................... 2
-------------------- end of arguments ---------------------
setting number of micro-batches to constant 1
> building GPT2BPETokenizer tokenizer ...
 > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304)
> initializing torch distributed ...
Traceback (most recent call last):
  File "/root/Megatron-LM/pretrain_gpt.py", line 203, in <module>
    pretrain(train_valid_test_datasets_provider,
  File "/root/Megatron-LM/megatron/training.py", line 94, in pretrain
    initialize_megatron(extra_args_provider=extra_args_provider,
  File "/root/Megatron-LM/megatron/initialize.py", line 80, in initialize_megatron
    finish_mpu_init()
  File "/root/Megatron-LM/megatron/initialize.py", line 60, in finish_mpu_init
    _initialize_distributed()
  File "/root/Megatron-LM/megatron/initialize.py", line 196, in _initialize_distributed
    torch.distributed.init_process_group(
  File "/opt/conda/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py", line 932, in init_process_group
    _store_based_barrier(rank, store, timeout)
  File "/opt/conda/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py", line 469, in _store_based_barrier
    raise RuntimeError(
RuntimeError: Timed out initializing process group in store based barrier on rank: 0, for key: store_based_barrier_key:1 (world_size=2, worker_count=1, timeout=0:10:00)

[plegresl]

Could you try running a simple PyTorch Distributed test program to help determine if it's something wrong with your infrastructure and job startup, or something with Megatron itself.

`import os

import torch
import torch.distributed

print('MASTER_PORT = ', os.getenv('MASTER_PORT'))
print('MASTER_ADDR = ', os.getenv('MASTER_ADDR'))
print('WORLD_SIZE = ', os.getenv('WORLD_SIZE'))
print('RANK = ', os.getenv('RANK'))
print('LOCAL_RANK = ', os.getenv('LOCAL_RANK'))

print('About to initialize PyTorch Distributed...', flush=True)
torch.distributed.init_process_group(backend='nccl')
print('Completed initialization of PyTorch Distributed', flush=True)

print('Entering barrier...', flush=True)
torch.distributed.barrier()
print('Done with barrier', flush=True)
`

[felipeliliti]

Sei la