Hugging Face's logo

nanotron
/
old_bench

Model card Files Community
3outeille HF staff commited on
Commit
7293135
·
verified ·
1 Parent(s): a6e5fbf

Upload llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8

Browse files
Files changed (4) hide show
  1. llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/bench.slurm +111 -0
  2. llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/config.yaml +90 -0
  3. llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/log.out +370 -0
  4. llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/status.txt +1 -0
llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/bench.slurm ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+
3
+ #SBATCH --job-name=bench_cluster
4
+ #SBATCH --time=00:59:00
5
+ #SBATCH --partition=hopper-prod
6
+ #SBATCH --nodes=2
7
+ #SBATCH --gres=gpu:8
8
+ #SBATCH --qos=high
9
+ #SBATCH --ntasks-per-node=1
10
+ #SBATCH --cpus-per-task=96
11
+ #SBATCH --exclusive
12
+ #SBATCH --output=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/log.out
13
+ #SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/log.out
14
+
15
+ # Function to update status based on squeue output
16
+ update_status() {
17
+ job_id=$1
18
+ status_file=$2
19
+ # For unknown reasons, it doenst update status for pending. It only works for running
20
+ while true; do
21
+ job_status=$(squeue --job $job_id --noheader --format=%T)
22
+ echo "Job status: $job_status"
23
+ if [ -z "$job_status" ]; then
24
+ # Job has finished or is not found
25
+ break
26
+ elif [ "$job_status" = "RUNNING" ]; then
27
+ printf "running" > $status_file
28
+ break
29
+ fi
30
+ sleep 10
31
+ done
32
+ }
33
+
34
+ # Misc initializations.
35
+ echo "========================"
36
+ echo "START TIME: $(date)"
37
+ source /fsx/ferdinandmom/miniforge3/etc/profile.d/conda.sh
38
+ conda activate /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster
39
+ echo python3 version = $(python3 --version)
40
+ echo "========================"
41
+
42
+ # Slurm stuff
43
+ export HOSTNAMES=$(scontrol show hostnames "$SLURM_JOB_NODELIST")
44
+ export MASTER_ADDR=$(scontrol show hostnames "$SLURM_JOB_NODELIST" | head -n 1)
45
+ export MASTER_PORT=$((1024 + RANDOM % 64511))
46
+
47
+ export TMPDIR=/scratch
48
+ export HF_DATASETS_CACHE="/admin/home/ferdinand_mom/.cache"
49
+ export CUBLAS_WORKSPACE_CONFIG=":4096:8"
50
+ export CUDA_DEVICE_MAX_CONNECTIONS="1"
51
+
52
+ huggingface-cli login --token $HUGGINGFACE_TOKEN
53
+
54
+
55
+ NANOTRON_REPO="/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron"
56
+ CMD="$NANOTRON_REPO/run_train.py --config-file /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/config.yaml"
57
+
58
+ LAUNCHER="torchrun \
59
+ --nproc_per_node 8 \
60
+ --nnodes 2 \
61
+ --rdzv_endpoint ${MASTER_ADDR}:${MASTER_PORT} \
62
+ --rdzv_backend c10d \
63
+ --max_restarts 0 \
64
+ --tee 3 \
65
+ --node_rank ${SLURM_PROCID}"
66
+
67
+ # Checkout the bench_cluster branch
68
+ cd $NANOTRON_REPO
69
+ git checkout bench_cluster
70
+ cd ..
71
+ # Get the current job ID
72
+ job_id=${SLURM_JOB_ID}
73
+
74
+ # Update status to "pending" or "running" in the background
75
+ update_status $job_id /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/status.txt &
76
+
77
+ # Run the main command
78
+ srun -u $LAUNCHER $CMD
79
+ exit_status=$?
80
+
81
+ # Update status based on the exit status of `srun`
82
+ if [ $exit_status -eq 0 ]; then
83
+ printf "completed" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/status.txt
84
+ else
85
+ if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/log.out; then
86
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/status.txt
87
+ elif grep -q " CUDA error: an illegal memory access" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/log.out; then
88
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/status.txt
89
+ elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/log.out; then
90
+ printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/status.txt
91
+ else
92
+ printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/status.txt
93
+ fi
94
+ fi
95
+
96
+ # Run the report script if the job completed successfully
97
+ if [ $exit_status -eq 0 ]; then
98
+ python /fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py report --inp_dir /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8 --is_logs
99
+ python /fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py report --inp_dir /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8 --is_profiler
100
+ fi
101
+
102
+
103
+ # Push to hub the folder using huggingface_cli
104
+ huggingface-cli upload nanotron/bench_cluster /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8 llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8 --commit-message "Upload llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8"
105
+
106
+ # Verify the upload
107
+ if [ $? -eq 0 ]; then
108
+ echo "Uploading to Huggingface Hub successful"
109
+ else
110
+ echo "Failed to upload to Huggingface Hub"
111
+ fi
llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/config.yaml ADDED
@@ -0,0 +1,90 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ general:
2
+ project: bench_cluster
3
+ seed: 42
4
+ model:
5
+ ddp_bucket_cap_mb: 25
6
+ dtype: bfloat16
7
+ init_method:
8
+ std: 0.025
9
+ make_vocab_size_divisible_by: 1
10
+ model_config:
11
+ bos_token_id: 1
12
+ eos_token_id: 2
13
+ hidden_act: silu
14
+ hidden_size: 2048
15
+ initializer_range: 0.02
16
+ intermediate_size: 4096
17
+ is_llama_config: true
18
+ max_position_embeddings: 4096
19
+ num_attention_heads: 32
20
+ num_hidden_layers: 24
21
+ num_key_value_heads: 32
22
+ pad_token_id: null
23
+ pretraining_tp: 1
24
+ rms_norm_eps: 1.0e-05
25
+ rope_scaling: null
26
+ rope_theta: 10000.0
27
+ tie_word_embeddings: true
28
+ use_cache: true
29
+ vocab_size: 50257
30
+ optimizer:
31
+ accumulate_grad_in_fp32: true
32
+ clip_grad: 1.0
33
+ learning_rate_scheduler:
34
+ learning_rate: 0.0001
35
+ lr_decay_style: linear
36
+ lr_warmup_style: linear
37
+ lr_warmup_steps: 1
38
+ min_decay_lr: 1.0e-05
39
+ optimizer_factory:
40
+ adam_beta1: 0.9
41
+ adam_beta2: 0.95
42
+ adam_eps: 1.0e-08
43
+ name: adamW
44
+ torch_adam_is_fused: true
45
+ weight_decay: 0.01
46
+ zero_stage: 1
47
+ parallelism:
48
+ dp: 1
49
+ expert_parallel_size: 1
50
+ pp: 16
51
+ pp_engine: 1f1b
52
+ tp: 1
53
+ tp_linear_async_communication: false
54
+ tp_mode: REDUCE_SCATTER
55
+ profiler:
56
+ profiler_export_path: /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8
57
+ tokenizer:
58
+ tokenizer_max_length: null
59
+ tokenizer_name_or_path: openai-community/gpt2
60
+ tokenizer_revision: null
61
+ data_stages:
62
+ - name: Training Stage
63
+ start_training_step: 1
64
+ data:
65
+ dataset:
66
+ dataset_overwrite_cache: false
67
+ dataset_processing_num_proc_per_process: 64
68
+ hf_dataset_config_name: null
69
+ hf_dataset_or_datasets: roneneldan/TinyStories
70
+ hf_dataset_splits: train
71
+ text_column_name: text
72
+ num_loading_workers: 32
73
+ seed: 42
74
+ lighteval: null
75
+ tokens:
76
+ train_steps: 20
77
+ val_check_interval: -1
78
+ batch_accumulation_per_replica: 128
79
+ limit_test_batches: 0
80
+ limit_val_batches: 0
81
+ micro_batch_size: 8
82
+ sequence_length: 4096
83
+ logging:
84
+ iteration_step_info_interval: 1
85
+ log_level: info
86
+ log_level_replica: info
87
+ checkpoints:
88
+ checkpoint_interval: 100000
89
+ checkpoints_path: /dev/null
90
+ resume_checkpoint_path: null
llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/log.out ADDED
@@ -0,0 +1,370 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ========================
2
+ START TIME: Tue Jul 2 18:40:11 UTC 2024
3
+ python3 version = Python 3.10.14
4
+ ========================
5
+ The token has not been saved to the git credentials helper. Pass `add_to_git_credential=True` in this function directly or `--add-to-git-credential` if using via `huggingface-cli` if you want to set the git credential as well.
6
+ Token is valid (permission: write).
7
+ Your token has been saved to /admin/home/ferdinand_mom/.cache/huggingface/token
8
+ Login successful
9
+ Already on 'bench_cluster'
10
+ M examples/config_tiny_llama.py
11
+ M examples/config_tiny_llama.yaml
12
+ M examples/train_tiny_llama.sh
13
+ M src/nanotron/models/llama.py
14
+ M src/nanotron/trainer.py
15
+ Your branch is up to date with 'origin/bench_cluster'.
16
+ Job status: RUNNING
17
+ W0702 18:40:13.689000 140539943585600 torch/distributed/run.py:757]
18
+ W0702 18:40:13.689000 140539943585600 torch/distributed/run.py:757] *****************************************
19
+ W0702 18:40:13.689000 140539943585600 torch/distributed/run.py:757] 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.
20
+ W0702 18:40:13.689000 140539943585600 torch/distributed/run.py:757] *****************************************
21
+ W0702 18:40:13.690000 140293142153024 torch/distributed/run.py:757]
22
+ W0702 18:40:13.690000 140293142153024 torch/distributed/run.py:757] *****************************************
23
+ W0702 18:40:13.690000 140293142153024 torch/distributed/run.py:757] 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.
24
+ W0702 18:40:13.690000 140293142153024 torch/distributed/run.py:757] *****************************************
25
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Config:
26
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Config(general=GeneralArgs(project='bench_cluster',
27
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: run='%date_%jobid',
28
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: seed=42,
29
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: step=None,
30
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: consumed_train_samples=None,
31
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: benchmark_csv_path=None,
32
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: ignore_sanity_checks=True),
33
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: parallelism=ParallelismArgs(dp=1,
34
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pp=16,
35
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tp=1,
36
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f2067bb0910>,
37
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
38
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tp_linear_async_communication=False,
39
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: expert_parallel_size=1),
40
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
41
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: eos_token_id=2,
42
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hidden_act='silu',
43
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hidden_size=2048,
44
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: initializer_range=0.02,
45
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: intermediate_size=4096,
46
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: is_llama_config=True,
47
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: max_position_embeddings=4096,
48
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_attention_heads=32,
49
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_hidden_layers=24,
50
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_key_value_heads=32,
51
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pad_token_id=None,
52
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pretraining_tp=1,
53
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rms_norm_eps=1e-05,
54
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rope_scaling=None,
55
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rope_theta=10000.0,
56
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tie_word_embeddings=True,
57
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: use_cache=True,
58
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: vocab_size=50257),
59
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: init_method=RandomInit(std=0.025),
60
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: dtype=torch.bfloat16,
61
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: make_vocab_size_divisible_by=1,
62
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: ddp_bucket_cap_mb=25),
63
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
64
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tokenizer_revision=None,
65
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tokenizer_max_length=None),
66
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
67
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: checkpoint_interval=100000,
68
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: save_initial_state=False,
69
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: resume_checkpoint_path=None,
70
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: checkpoints_path_is_shared_file_system=False),
71
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: logging=LoggingArgs(log_level='info',
72
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: log_level_replica='info',
73
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration_step_info_interval=1),
74
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tokens=TokensArgs(sequence_length=4096,
75
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: train_steps=20,
76
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: micro_batch_size=8,
77
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: batch_accumulation_per_replica=128,
78
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: val_check_interval=-1,
79
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: limit_val_batches=0,
80
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: limit_test_batches=0),
81
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
82
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: adam_beta1=0.9,
83
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: adam_beta2=0.95,
84
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: torch_adam_is_fused=True,
85
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: name='adamW'),
86
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: zero_stage=1,
87
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: weight_decay=0.01,
88
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: clip_grad=1.0,
89
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: accumulate_grad_in_fp32=True,
90
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
91
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lr_warmup_steps=1,
92
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lr_warmup_style='linear',
93
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lr_decay_style='linear',
94
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lr_decay_steps=19,
95
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lr_decay_starting_step=None,
96
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: min_decay_lr=1e-05)),
97
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: data_stages=[DatasetStageArgs(name='Training Stage',
98
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: start_training_step=1,
99
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
100
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hf_dataset_splits='train',
101
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hf_dataset_config_name=None,
102
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: dataset_processing_num_proc_per_process=64,
103
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: dataset_overwrite_cache=False,
104
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: text_column_name='text'),
105
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: seed=42,
106
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_loading_workers=32))],
107
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8')),
108
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lighteval=None)
109
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Model Config:
110
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: LlamaConfig(bos_token_id=1,
111
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: eos_token_id=2,
112
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hidden_act='silu',
113
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hidden_size=2048,
114
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: initializer_range=0.02,
115
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: intermediate_size=4096,
116
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: is_llama_config=True,
117
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: max_position_embeddings=4096,
118
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_attention_heads=32,
119
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_hidden_layers=24,
120
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_key_value_heads=32,
121
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pad_token_id=None,
122
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pretraining_tp=1,
123
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rms_norm_eps=1e-05,
124
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rope_scaling=None,
125
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rope_theta=10000.0,
126
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tie_word_embeddings=True,
127
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: use_cache=True,
128
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: vocab_size=50257)
129
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Building model..
130
+ [default0]:07/02/2024 18:40:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Setting PP block ranks...
131
+ [default0]:07/02/2024 18:40:48 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Total number of parameters: 1.21G (2312.82MiB)
132
+ [default0]:07/02/2024 18:40:48 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Local number of parameters: 187M (356.33MiB)
133
+ [default0]:07/02/2024 18:40:48 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [After model building] Memory usage: 358.34MiB. Peak allocated: 360.37MiB Peak reserved: 368.00MiB
134
+ [default0]:07/02/2024 18:40:48 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: No checkpoint path provided.
135
+ [default0]:07/02/2024 18:40:48 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Parametrizing model parameters using StandardParametrizator
136
+ [default0]:07/02/2024 18:40:48 [INFO|DP=0|PP=8|TP=0|ip-26-0-171-88]: Local number of parameters: 41.9M (80.01MiB)
137
+ [default0]:07/02/2024 18:40:48 [INFO|DP=0|PP=8|TP=0|ip-26-0-171-88]: [After model building] Memory usage: 81.02MiB. Peak allocated: 83.05MiB Peak reserved: 96.00MiB
138
+ [default0]:07/02/2024 18:40:48 [INFO|DP=0|PP=8|TP=0|ip-26-0-171-88]: No checkpoint path provided.
139
+ [default2]:07/02/2024 18:40:48 [INFO|DP=0|PP=2|TP=0|ip-26-0-171-62]: Local number of parameters: 41.9M (80.01MiB)
140
+ [default2]:07/02/2024 18:40:48 [INFO|DP=0|PP=2|TP=0|ip-26-0-171-62]: [After model building] Memory usage: 81.02MiB. Peak allocated: 83.05MiB Peak reserved: 96.00MiB
141
+ [default2]:07/02/2024 18:40:48 [INFO|DP=0|PP=2|TP=0|ip-26-0-171-62]: No checkpoint path provided.
142
+ [default7]:07/02/2024 18:40:48 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-62]: Local number of parameters: 83.9M (160.02MiB)
143
+ [default7]:07/02/2024 18:40:48 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-62]: [After model building] Memory usage: 162.03MiB. Peak allocated: 164.06MiB Peak reserved: 170.00MiB
144
+ [default5]:07/02/2024 18:40:48 [INFO|DP=0|PP=5|TP=0|ip-26-0-171-62]: Local number of parameters: 41.9M (80.01MiB)
145
+ [default5]:07/02/2024 18:40:48 [INFO|DP=0|PP=5|TP=0|ip-26-0-171-62]: [After model building] Memory usage: 81.02MiB. Peak allocated: 83.05MiB Peak reserved: 96.00MiB
146
+ [default4]:07/02/2024 18:40:48 [INFO|DP=0|PP=4|TP=0|ip-26-0-171-62]: Local number of parameters: 83.9M (160.02MiB)
147
+ [default1]:07/02/2024 18:40:48 [INFO|DP=0|PP=1|TP=0|ip-26-0-171-62]: Local number of parameters: 83.9M (160.02MiB)
148
+ [default1]:07/02/2024 18:40:48 [INFO|DP=0|PP=1|TP=0|ip-26-0-171-62]: [After model building] Memory usage: 162.03MiB. Peak allocated: 164.06MiB Peak reserved: 170.00MiB
149
+ [default4]:07/02/2024 18:40:48 [INFO|DP=0|PP=4|TP=0|ip-26-0-171-62]: [After model building] Memory usage: 162.03MiB. Peak allocated: 164.06MiB Peak reserved: 170.00MiB
150
+ [default5]:07/02/2024 18:40:48 [INFO|DP=0|PP=5|TP=0|ip-26-0-171-62]: No checkpoint path provided.
151
+ [default1]:07/02/2024 18:40:48 [INFO|DP=0|PP=1|TP=0|ip-26-0-171-62]: No checkpoint path provided.
152
+ [default7]:07/02/2024 18:40:48 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-62]: No checkpoint path provided.
153
+ [default4]:07/02/2024 18:40:48 [INFO|DP=0|PP=4|TP=0|ip-26-0-171-62]: No checkpoint path provided.
154
+ [default3]:07/02/2024 18:40:48 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-62]: Local number of parameters: 83.9M (160.02MiB)
155
+ [default3]:07/02/2024 18:40:48 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-62]: [After model building] Memory usage: 162.03MiB. Peak allocated: 164.06MiB Peak reserved: 170.00MiB
156
+ [default3]:07/02/2024 18:40:48 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-62]: No checkpoint path provided.
157
+ [default6]:07/02/2024 18:40:48 [INFO|DP=0|PP=6|TP=0|ip-26-0-171-62]: Local number of parameters: 83.9M (160.02MiB)
158
+ [default6]:07/02/2024 18:40:48 [INFO|DP=0|PP=6|TP=0|ip-26-0-171-62]: [After model building] Memory usage: 162.03MiB. Peak allocated: 164.06MiB Peak reserved: 170.00MiB
159
+ [default6]:07/02/2024 18:40:48 [INFO|DP=0|PP=6|TP=0|ip-26-0-171-62]: No checkpoint path provided.
160
+ [default1]:07/02/2024 18:40:48 [INFO|DP=0|PP=9|TP=0|ip-26-0-171-88]: Local number of parameters: 83.9M (160.02MiB)
161
+ [default2]:07/02/2024 18:40:48 [INFO|DP=0|PP=10|TP=0|ip-26-0-171-88]: Local number of parameters: 83.9M (160.02MiB)
162
+ [default2]:07/02/2024 18:40:48 [INFO|DP=0|PP=10|TP=0|ip-26-0-171-88]: [After model building] Memory usage: 162.03MiB. Peak allocated: 164.06MiB Peak reserved: 170.00MiB
163
+ [default2]:07/02/2024 18:40:48 [INFO|DP=0|PP=10|TP=0|ip-26-0-171-88]: No checkpoint path provided.
164
+ [default1]:07/02/2024 18:40:48 [INFO|DP=0|PP=9|TP=0|ip-26-0-171-88]: [After model building] Memory usage: 162.03MiB. Peak allocated: 164.06MiB Peak reserved: 170.00MiB
165
+ [default1]:07/02/2024 18:40:48 [INFO|DP=0|PP=9|TP=0|ip-26-0-171-88]: No checkpoint path provided.
166
+ [default6]:07/02/2024 18:40:48 [INFO|DP=0|PP=14|TP=0|ip-26-0-171-88]: Local number of parameters: 103M (196.32MiB)
167
+ [default6]:07/02/2024 18:40:48 [INFO|DP=0|PP=14|TP=0|ip-26-0-171-88]: [After model building] Memory usage: 196.33MiB. Peak allocated: 196.34MiB Peak reserved: 200.00MiB
168
+ [default6]:07/02/2024 18:40:48 [INFO|DP=0|PP=14|TP=0|ip-26-0-171-88]: No checkpoint path provided.
169
+ [default4]:07/02/2024 18:40:48 [INFO|DP=0|PP=12|TP=0|ip-26-0-171-88]: Local number of parameters: 83.9M (160.02MiB)
170
+ [default4]:07/02/2024 18:40:48 [INFO|DP=0|PP=12|TP=0|ip-26-0-171-88]: [After model building] Memory usage: 162.03MiB. Peak allocated: 164.06MiB Peak reserved: 170.00MiB
171
+ [default4]:07/02/2024 18:40:48 [INFO|DP=0|PP=12|TP=0|ip-26-0-171-88]: No checkpoint path provided.
172
+ [default5]:07/02/2024 18:40:48 [INFO|DP=0|PP=13|TP=0|ip-26-0-171-88]: Local number of parameters: 83.9M (160.02MiB)
173
+ [default5]:07/02/2024 18:40:48 [INFO|DP=0|PP=13|TP=0|ip-26-0-171-88]: [After model building] Memory usage: 162.03MiB. Peak allocated: 164.06MiB Peak reserved: 170.00MiB
174
+ [default5]:07/02/2024 18:40:48 [INFO|DP=0|PP=13|TP=0|ip-26-0-171-88]: No checkpoint path provided.
175
+ [default7]:07/02/2024 18:40:48 [INFO|DP=0|PP=15|TP=0|ip-26-0-171-88]: Local number of parameters: 0 (0.00MiB)
176
+ [default7]:07/02/2024 18:40:48 [INFO|DP=0|PP=15|TP=0|ip-26-0-171-88]: [After model building] Memory usage: 0.01MiB. Peak allocated: 0.02MiB Peak reserved: 2.00MiB
177
+ [default7]:07/02/2024 18:40:48 [INFO|DP=0|PP=15|TP=0|ip-26-0-171-88]: No checkpoint path provided.
178
+ [default3]:07/02/2024 18:40:48 [INFO|DP=0|PP=11|TP=0|ip-26-0-171-88]: Local number of parameters: 41.9M (80.01MiB)
179
+ [default3]:07/02/2024 18:40:48 [INFO|DP=0|PP=11|TP=0|ip-26-0-171-88]: [After model building] Memory usage: 81.02MiB. Peak allocated: 83.05MiB Peak reserved: 96.00MiB
180
+ [default3]:07/02/2024 18:40:48 [INFO|DP=0|PP=11|TP=0|ip-26-0-171-88]: No checkpoint path provided.
181
+ [default0]:07/02/2024 18:40:49 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [Optimizer Building] Using LearningRateForSP as learning rate
182
+ [default0]:07/02/2024 18:40:49 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] Size of optimizer params per rank:
183
+ [default0]:07/02/2024 18:40:49 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 0 has 187M out of 187M (100.00%) params' optimizer states
184
+ [default0]:07/02/2024 18:40:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
185
+ [default0]:07/02/2024 18:40:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Using `datasets` library
186
+ [default0]:07/02/2024 18:40:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
187
+ [default0]:07/02/2024 18:40:50 [WARNING|DP=0|PP=0|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
188
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
189
+ [default0]:07/02/2024 18:40:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [Training Plan] There are 1 training stages
190
+ [default0]:07/02/2024 18:40:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [Stage Training Stage] start from step 1
191
+ [default0]:07/02/2024 18:40:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]:
192
+ [default0]:07/02/2024 18:40:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [Start training] datetime: 2024-07-02 18:40:50.903488 | mbs: 8 | grad_accum: 128 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
193
+ [default0]:07/02/2024 18:40:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
194
+ [default0]:07/02/2024 18:40:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 1783.67MiB. Peak allocated 1783.67MiB. Peak reserved: 1796.00MiB
195
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
196
+ [default5]:07/02/2024 18:40:51 [WARNING|DP=0|PP=5|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
197
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
198
+ [default5]:07/02/2024 18:40:51 [WARNING|DP=0|PP=13|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
199
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
200
+ [default1]:07/02/2024 18:40:51 [WARNING|DP=0|PP=9|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
201
+ [default0]:07/02/2024 18:40:51 [WARNING|DP=0|PP=8|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
202
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
203
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
204
+ [default4]:07/02/2024 18:40:51 [WARNING|DP=0|PP=4|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
205
+ [default2]:07/02/2024 18:40:51 [WARNING|DP=0|PP=2|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
206
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
207
+ [default7]:07/02/2024 18:40:51 [WARNING|DP=0|PP=7|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
208
+ [default1]:07/02/2024 18:40:51 [WARNING|DP=0|PP=1|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
209
+ [default3]:07/02/2024 18:40:51 [WARNING|DP=0|PP=3|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
210
+ [default6]:07/02/2024 18:40:51 [WARNING|DP=0|PP=6|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
211
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
212
+ [default2]:07/02/2024 18:40:51 [WARNING|DP=0|PP=10|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
213
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
214
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
215
+ [default4]:07/02/2024 18:40:51 [WARNING|DP=0|PP=12|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
216
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
217
+ [default6]:07/02/2024 18:40:51 [WARNING|DP=0|PP=14|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
218
+ [default7]:07/02/2024 18:40:51 [WARNING|DP=0|PP=15|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
219
+ [default3]:07/02/2024 18:40:51 [WARNING|DP=0|PP=11|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
220
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
221
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
222
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
223
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
224
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
225
+ [default0]:[rank0]: Traceback (most recent call last):
226
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
227
+ [default0]:[rank0]: trainer.train(dataloader)
228
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
229
+ [default0]:[rank0]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
230
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
231
+ [default0]:[rank0]: outputs = self.pipeline_engine.train_batch_iter(
232
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 252, in train_batch_iter
233
+ [default0]:[rank0]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
234
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
235
+ [default0]:[rank0]: output = model(**micro_batch)
236
+ [default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
237
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
238
+ [default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
239
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
240
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
241
+ [default0]:[rank0]: sharded_logits = self.model(
242
+ [default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
243
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
244
+ [default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
245
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
246
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
247
+ [default0]:[rank0]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
248
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
249
+ [default0]:[rank0]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
250
+ [default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
251
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
252
+ [default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
253
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
254
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
255
+ [default0]:[rank0]: output = self.pp_block(**new_kwargs)
256
+ [default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
257
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
258
+ [default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
259
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
260
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 631, in forward
261
+ [default0]:[rank0]: output = self.attn(hidden_states=hidden_states, sequence_mask=sequence_mask)
262
+ [default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
263
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
264
+ [default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
265
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
266
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 360, in forward
267
+ [default0]:[rank0]: qkv_states = self.qkv_proj(
268
+ [default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
269
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
270
+ [default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
271
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
272
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
273
+ [default0]:[rank0]: return column_linear(
274
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
275
+ [default0]:[rank0]: return F.linear(input, weight, bias)
276
+ [default0]:[rank0]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 384.00 MiB. GPU
277
+ [default6]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: c10d::allreduce_: an autograd kernel was not registered to the Autograd key(s) but we are trying to backprop through it. This may lead to silently incorrect behavior. This behavior is deprecated and will be removed in a future version of PyTorch. If your operator is differentiable, please ensure you have registered an autograd kernel to the correct Autograd key (e.g. DispatchKey::Autograd, DispatchKey::CompositeImplicitAutograd). If your operator is not differentiable, or to squash this warning and use the previous behavior, please register torch::CppFunction::makeFallthrough() to DispatchKey::Autograd. (Triggered internally at ../torch/csrc/autograd/autograd_not_implemented_fallback.cpp:63.)
278
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
279
+ W0702 18:41:24.865000 140539943585600 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 3707498 closing signal SIGTERM
280
+ W0702 18:41:24.865000 140539943585600 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 3707499 closing signal SIGTERM
281
+ W0702 18:41:24.866000 140539943585600 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 3707500 closing signal SIGTERM
282
+ W0702 18:41:24.867000 140539943585600 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 3707501 closing signal SIGTERM
283
+ W0702 18:41:24.867000 140539943585600 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 3707502 closing signal SIGTERM
284
+ W0702 18:41:24.868000 140539943585600 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 3707503 closing signal SIGTERM
285
+ W0702 18:41:24.870000 140539943585600 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 3707504 closing signal SIGTERM
286
+ [default5]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: c10d::allreduce_: an autograd kernel was not registered to the Autograd key(s) but we are trying to backprop through it. This may lead to silently incorrect behavior. This behavior is deprecated and will be removed in a future version of PyTorch. If your operator is differentiable, please ensure you have registered an autograd kernel to the correct Autograd key (e.g. DispatchKey::Autograd, DispatchKey::CompositeImplicitAutograd). If your operator is not differentiable, or to squash this warning and use the previous behavior, please register torch::CppFunction::makeFallthrough() to DispatchKey::Autograd. (Triggered internally at ../torch/csrc/autograd/autograd_not_implemented_fallback.cpp:63.)
287
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
288
+ [default4]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: c10d::allreduce_: an autograd kernel was not registered to the Autograd key(s) but we are trying to backprop through it. This may lead to silently incorrect behavior. This behavior is deprecated and will be removed in a future version of PyTorch. If your operator is differentiable, please ensure you have registered an autograd kernel to the correct Autograd key (e.g. DispatchKey::Autograd, DispatchKey::CompositeImplicitAutograd). If your operator is not differentiable, or to squash this warning and use the previous behavior, please register torch::CppFunction::makeFallthrough() to DispatchKey::Autograd. (Triggered internally at ../torch/csrc/autograd/autograd_not_implemented_fallback.cpp:63.)
289
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
290
+ E0702 18:41:27.800000 140539943585600 torch/distributed/elastic/multiprocessing/api.py:826] failed (exitcode: 1) local_rank: 0 (pid: 3707497) of binary: /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/python3.10
291
+ Traceback (most recent call last):
292
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/torchrun", line 8, in <module>
293
+ sys.exit(main())
294
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 347, in wrapper
295
+ return f(*args, **kwargs)
296
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 879, in main
297
+ run(args)
298
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 870, in run
299
+ elastic_launch(
300
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 132, in __call__
301
+ return launch_agent(self._config, self._entrypoint, list(args))
302
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 263, in launch_agent
303
+ raise ChildFailedError(
304
+ torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
305
+ ============================================================
306
+ /fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py FAILED
307
+ ------------------------------------------------------------
308
+ Failures:
309
+ <NO_OTHER_FAILURES>
310
+ ------------------------------------------------------------
311
+ Root Cause (first observed failure):
312
+ [0]:
313
+ time : 2024-07-02_18:41:24
314
+ host : ip-26-0-171-62.ec2.internal
315
+ rank : 0 (local_rank: 0)
316
+ exitcode : 1 (pid: 3707497)
317
+ error_file: <N/A>
318
+ traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
319
+ ============================================================
320
+ srun: error: ip-26-0-171-62: task 0: Exited with exit code 1
321
+ [default3]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: c10d::allreduce_: an autograd kernel was not registered to the Autograd key(s) but we are trying to backprop through it. This may lead to silently incorrect behavior. This behavior is deprecated and will be removed in a future version of PyTorch. If your operator is differentiable, please ensure you have registered an autograd kernel to the correct Autograd key (e.g. DispatchKey::Autograd, DispatchKey::CompositeImplicitAutograd). If your operator is not differentiable, or to squash this warning and use the previous behavior, please register torch::CppFunction::makeFallthrough() to DispatchKey::Autograd. (Triggered internally at ../torch/csrc/autograd/autograd_not_implemented_fallback.cpp:63.)
322
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
323
+ W0702 18:41:29.830000 140287475332864 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1252] The node 'ip-26-0-171-88.ec2.internal_696127_0' has failed to send a keep-alive heartbeat to the rendezvous 'none' due to an error of type RendezvousConnectionError.
324
+ W0702 18:41:29.866000 140293142153024 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 696196 closing signal SIGTERM
325
+ W0702 18:41:29.866000 140293142153024 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 696197 closing signal SIGTERM
326
+ W0702 18:41:29.867000 140293142153024 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 696198 closing signal SIGTERM
327
+ W0702 18:41:29.869000 140293142153024 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 696199 closing signal SIGTERM
328
+ W0702 18:41:29.870000 140293142153024 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 696200 closing signal SIGTERM
329
+ W0702 18:41:29.871000 140293142153024 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 696201 closing signal SIGTERM
330
+ W0702 18:41:29.872000 140293142153024 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 696202 closing signal SIGTERM
331
+ W0702 18:41:29.873000 140293142153024 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 696203 closing signal SIGTERM
332
+ W0702 18:41:32.911000 140293142153024 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1203] The node 'ip-26-0-171-88.ec2.internal_696127_0' has failed to shutdown the rendezvous 'none' due to an error of type RendezvousConnectionError.
333
+ W0702 18:41:32.923000 140293142153024 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1203] The node 'ip-26-0-171-88.ec2.internal_696127_0' has failed to shutdown the rendezvous 'none' due to an error of type RendezvousConnectionError.
334
+ Traceback (most recent call last):
335
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/rendezvous/c10d_rendezvous_backend.py", line 113, in _call_store
336
+ return getattr(self._store, store_op)(*args, **kwargs)
337
+ torch.distributed.DistNetworkError: Broken pipe
338
+
339
+ The above exception was the direct cause of the following exception:
340
+
341
+ Traceback (most recent call last):
342
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/torchrun", line 8, in <module>
343
+ sys.exit(main())
344
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 347, in wrapper
345
+ return f(*args, **kwargs)
346
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 879, in main
347
+ run(args)
348
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 870, in run
349
+ elastic_launch(
350
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 132, in __call__
351
+ return launch_agent(self._config, self._entrypoint, list(args))
352
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 254, in launch_agent
353
+ result = agent.run()
354
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 123, in wrapper
355
+ result = f(*args, **kwargs)
356
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 733, in run
357
+ result = self._invoke_run(role)
358
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 908, in _invoke_run
359
+ num_nodes_waiting = rdzv_handler.num_nodes_waiting()
360
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/rendezvous/dynamic_rendezvous.py", line 1174, in num_nodes_waiting
361
+ self._state_holder.sync()
362
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/rendezvous/dynamic_rendezvous.py", line 419, in sync
363
+ get_response = self._backend.get_state()
364
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/rendezvous/c10d_rendezvous_backend.py", line 73, in get_state
365
+ base64_state: bytes = self._call_store("get", self._key)
366
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/rendezvous/c10d_rendezvous_backend.py", line 115, in _call_store
367
+ raise RendezvousConnectionError(
368
+ torch.distributed.elastic.rendezvous.api.RendezvousConnectionError: The connection to the C10d store has failed. See inner exception for details.
369
+ srun: error: ip-26-0-171-88: task 1: Exited with exit code 1
370
+ Consider using `hf_transfer` for faster uploads. This solution comes with some limitations. See https://huggingface.co/docs/huggingface_hub/hf_transfer for more details.
llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-8/status.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ oom