eleutherai

Preetham-gali's group workspace

Group: jMJvzfv6JDiRc9Yc8cj3Ec_3hhg1ht4

1-1

of 1

Tags

Notes

Author

stellaathena

State

Crashed

Start time

August 27th, 2021 3:31:52 PM

Runtime

11d 5h 49m 57s

Tracked hours

11d 5h 49m 53s

Run path

eleutherai/distilling/vdriqx5m

Linux-5.8.0-59-generic-x86_64-with-glibc2.29

Python version

3.8.5

Git repository

git clone https://github.com/EleutherAI/gpt-neox.git

Git state

git checkout -b "Large-to-Small" cbba9c69322ea18d1504dc1dba07b429f8b71c2d

Command

pretrain_gpt2.py --local_rank=0 --num_gpus 4 --deepspeed_config "{\"train_batch_size\": 160, \"train_micro_batch_size_per_gpu\": 2, \"gradient_accumulation_steps\": 20, \"optimizer\": {\"type\": \"Adam\", \"params\": {\"lr\": 0.0003, \"betas\": [0.9, 0.999], \"eps\": 1e-08}}, \"fp16\": {\"enabled\": true, \"loss_scale\": 0, \"loss_scale_window\": 1000, \"hysteresis\": 2, \"min_loss_scale\": 1}, \"gradient_clipping\": 1.0, \"zero_optimization\": {\"stage\": 1, \"allgather_partitions\": true, \"allgather_bucket_size\": 500000000, \"overlap_comm\": true, \"reduce_scatter\": true, \"reduce_bucket_size\": 500000000, \"contiguous_gradients\": true, \"cpu_offload\": false}, \"wall_clock_breakdown\": true}" --megatron_config "{\"num_gpus\": 4, \"train_batch_size\": 160, \"train_micro_batch_size_per_gpu\": 2, \"gradient_accumulation_steps\": 20, \"optimizer\": {\"type\": \"Adam\", \"params\": {\"lr\": 0.0003, \"betas\": [0.9, 0.999], \"eps\": 1e-08}}, \"fp16\": {\"enabled\": true, \"loss_scale\": 0, \"loss_scale_window\": 1000, \"hysteresis\": 2, \"min_loss_scale\": 1}, \"gradient_clipping\": 1.0, \"zero_optimization\": {\"stage\": 1, \"allgather_partitions\": true, \"allgather_bucket_size\": 500000000, \"overlap_comm\": true, \"reduce_scatter\": true, \"reduce_bucket_size\": 500000000, \"contiguous_gradients\": true, \"cpu_offload\": false}, \"wall_clock_breakdown\": true, \"precision\": \"fp16\", \"lr_decay_style\": \"cosine\", \"lr_decay_iters\": 250000, \"optimizer_type\": \"Adam\", \"zero_stage\": 1, \"zero_reduce_scatter\": true, \"zero_contiguous_gradients\": true, \"zero_reduce_bucket_size\": 500000000, \"zero_allgather_bucket_size\": 500000000, \"lr\": 0.0003, \"data_path\": \"/mnt/ssd-1/data/pile/pile_text_document\", \"data_impl\": \"mmap\", \"save\": \"checkpoints/distilling/large-to-small\", \"load\": \"checkpoints/distilling/large-to-small\", \"save_interval\": 10000, \"finetune\": true, \"batch_size\": 2, \"train_iters\": 250000, \"eval_iters\": 10, \"keep_last_n_checkpoints\": 4, \"split\": \"949,50,1\", \"vocab_file\": \"/mnt/ssd-1/data/gpt2-vocab.json\", \"merge_file\": \"/mnt/ssd-1/data/gpt2-merges.txt\", \"attention_dropout\": 0.0, \"hidden_dropout\": 0.0, \"weight_decay\": 0.1, \"checkpoint_activations\": true, \"synchronize_each_layer\": true, \"partition_activations\": true, \"gas\": 20, \"clip_grad\": 1.0, \"dynamic_loss_scale\": true, \"do_distillation\": true, \"teacher_model_args\": {\"precision\": null, \"num_layers\": 24, \"hidden_size\": 1536, \"num_attention_heads\": 16, \"seq_length\": 2048, \"max_position_embeddings\": 2048, \"norm\": \"layernorm\", \"layernorm_epsilon\": 1e-05, \"rms_norm_epsilon\": 1e-08, \"scalenorm_epsilon\": 1e-08, \"pos_emb\": \"rotary\", \"rpe_num_buckets\": 32, \"rpe_max_distance\": 128, \"no_weight_tying\": true, \"attention_config\": [\"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\"], \"sparsity_config\": {}, \"num_unique_layers\": null, \"param_sharing_style\": \"grouped\", \"make_vocab_size_divisible_by\": 128, \"apply_residual_connection_post_layernorm\": false, \"activation\": \"gelu\", \"scaled_upper_triang_masked_softmax_fusion\": false, \"scaled_masked_softmax_fusion\": false, \"bias_gelu_fusion\": false, \"bias_dropout_fusion\": false, \"fp16_lm_cross_entropy\": false, \"init_method_std\": 0.02, \"apply_query_key_layer_scaling\": false, \"use_cpu_initialization\": false, \"attention_softmax_in_fp32\": false, \"rotary_pct\": 1.0, \"rotary_emb_base\": 10000, \"init_method\": \"normal\", \"output_layer_init_method\": \"scaled_normal\", \"gmlp_attn_dim\": 64}, \"student_model_args\": {\"precision\": null, \"num_layers\": 12, \"hidden_size\": 1024, \"num_attention_heads\": 8, \"seq_length\": 2048, \"max_position_embeddings\": 2048, \"norm\": \"layernorm\", \"layernorm_epsilon\": 1e-05, \"rms_norm_epsilon\": 1e-08, \"scalenorm_epsilon\": 1e-08, \"pos_emb\": \"rotary\", \"rpe_num_buckets\": 32, \"rpe_max_distance\": 128, \"no_weight_tying\": true, \"attention_config\": [\"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\", \"global\"], \"sparsity_config\": {}, \"num_unique_layers\": null, \"param_sharing_style\": \"grouped\", \"make_vocab_size_divisible_by\": 128, \"apply_residual_connection_post_layernorm\": false, \"activation\": \"gelu\", \"scaled_upper_triang_masked_softmax_fusion\": false, \"scaled_masked_softmax_fusion\": false, \"bias_gelu_fusion\": false, \"bias_dropout_fusion\": false, \"fp16_lm_cross_entropy\": false, \"init_method_std\": 0.02, \"apply_query_key_layer_scaling\": false, \"use_cpu_initialization\": false, \"attention_softmax_in_fp32\": false, \"rotary_pct\": 1.0, \"rotary_emb_base\": 10000, \"init_method\": \"normal\", \"output_layer_init_method\": \"scaled_normal\", \"gmlp_attn_dim\": 64}, \"load_teacher\": \"/mnt/ssd-1/neox_checkpoints/dense_large_checkpoints/global_step250000\", \"alpha_lm\": 1.0, \"alpha_kld\": 1.0, \"pipe_parallel_size\": 1, \"is_pipe_parallel\": true, \"use_wandb\": true, \"wandb_group\": \"jMJvzfv6JDiRc9Yc8cj3Ec_3hhg1ht4\", \"wandb_team\": \"eleutherai\", \"wandb_project\": \"distilling\", \"log_dir\": \"logs\", \"log_interval\": 100, \"user_script\": \"pretrain_gpt2.py\"}"

System Hardware

CPU count	128
GPU count	4
GPU type	NVIDIA A100-SXM4-40GB

W&B CLI Version

0.10.28

Group

jMJvzfv6JDiRc9Yc8cj3Ec_3hhg1ht4

Config parameters are your model's inputs. Learn more

▶
Config parameters:{} 186 keys
- activation:
  "gelu"
- adlr_autoresume:
  false
- adlr_autoresume_interval:
  1,000
- alpha_kld:
  1
- alpha_lm:
  1
- alpha_mse:
  0
- amp:
  null
- apply_query_key_layer_scaling:
  false
- apply_residual_connection_post_layernorm:
  false
- attention_config:
  null
- attention_dropout:
  0
- attention_softmax_in_fp32:
  false
- batch_size:
  2
- bias_dropout_fusion:
  false
- bias_gelu_fusion:
  false
- checkpoint_activations:
  true
- checkpoint_in_cpu:
  false
- checkpoint_num_layers:
  1
- checkpoint_validation_with_forward_pass:
  false
- clip_grad:
  1
- contiguous_checkpointing:
  false
- data_impl:
  "mmap"
- data_path:
  "/mnt/ssd-1/data/pile/pile_text_document"
- deepscale:
  false
- deepscale_config:
  null
- deepspeed:
  true
- deepspeed_activation_checkpointing:
  true
- deepspeed_mpi:
  false
- detect_nvlink_pairs:
  false
- distributed_backend:
  "nccl"
- do_distillation:
  true
- do_test:
  null
- do_train:
  null
- do_valid:
  null
- dump_state:
  false
- dynamic_loss_scale:
  true
- eod_mask_loss:
  false
- eval_interval:
  1,000
- eval_iters:
  10
- eval_results_prefix:
  ""
- eval_tasks:
  null
- exclude:
  null
- exit_interval:
  null
- finetune:
  true
- flops_profiler:
  null
- ▶
  fp16:{} 5 keys
- ▶
  zero_optimization:{} 8 keys
- zero_reduce_bucket_size:
  500,000,000
- zero_reduce_scatter:
  true
- zero_stage:
  1

Summary metrics are your model's outputs. Learn more

▶
Summary metrics:{} 38 keys
- runtime/flops_per_sec_per_gpu:
  159,030,246,998,360.47
- runtime/iteration_time:
  4.1426050043106075
- runtime/samples_per_sec:
  38.623040293127445
- timers/_step_check_overflow:
  0.6463527679443359
- timers/_step_clipping:
  0.1201629638671875
- timers/_step_step:
  238.24477195739743
- timers/_step_zero_grad:
  0.4758834838867187
- timers/backward:
  13,090.79885482788
- timers/backward_allreduce:
  2.0503997802734375
- timers/backward_allreduce_microstep:
  5.942344665527344
- timers/backward_inner:
  13,074.701070785522
- timers/backward_inner_microstep:
  13,078.87053489685
- timers/backward_microstep:
  13,094.89607810974
- timers/batch_input:
  217.02313423156735
- timers/comms:
  360.1267337799072
- timers/forward:
  21,121.46496772766
- timers/pct_backward:
  31.39650713422838
- timers/pct_comms:
  0.8637151706121668
- timers/pct_fwd:
  50.65697157970237
- timers/pct_optimizer_step:
  0.57595745131733
- timers/reduce_grads:
  325.32191276550293
- timers/reduce_tied_grads:
  0.29754638671875
- timers/step:
  240.14592170715332
- timers/train_batch:
  0
- train/kld_loss:
  0.9171763062477112
- train/learning_rate:
  0.00000462944840916533
- train/lm_loss:
  3.231654167175293
- train/loss:
  4.148830890655518
- train/loss_scale:
  131,072
- train/mse_loss:
  0
- validation/kld_loss:
  0.9308838248252868
- validation/kld_loss_ppl:
  2.536750229951912
- validation/lm_loss:
  3.231205701828003
- validation/lm_loss_ppl:
  25.31015508170337
- validation/loss:
  4.162089824676514
- validation/loss_ppl:
  64.20556085837497
- validation/mse_loss:
  0
- validation/mse_loss_ppl:
  1

This run produced these artifacts as outputs. Total: 1. Learn more

wandb-history

run-vdriqx5m-history:v0