Hugging Face's logo

nanotron
/
old_bench

Model card Files Community
3outeille HF staff commited on
Commit
1fd7d1a
•
1 Parent(s): 6a35671

Upload llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16

Browse files
Files changed (4) hide show
  1. llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16/bench.slurm +111 -0
  2. llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16/config.yaml +90 -0
  3. llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16/log.out +416 -0
  4. llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16/status.txt +1 -0
llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16/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-2_pp-8_mbz-16/log.out
13
+ #SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16/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-2_pp-8_mbz-16/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-2_pp-8_mbz-16/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-2_pp-8_mbz-16/status.txt
84
+ else
85
+ if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16/log.out; then
86
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16/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-2_pp-8_mbz-16/log.out; then
88
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16/status.txt
89
+ elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16/log.out; then
90
+ printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16/status.txt
91
+ else
92
+ printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16/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-2_pp-8_mbz-16 --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-2_pp-8_mbz-16 --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-2_pp-8_mbz-16 llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16 --commit-message "Upload llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16"
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-2_pp-8_mbz-16/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: 8
51
+ pp_engine: 1f1b
52
+ tp: 2
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-2_pp-8_mbz-16
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: 64
79
+ limit_test_batches: 0
80
+ limit_val_batches: 0
81
+ micro_batch_size: 16
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-2_pp-8_mbz-16/log.out ADDED
@@ -0,0 +1,416 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ========================
2
+ START TIME: Tue Jul 2 18:57:49 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 18:57:52.390000 140570191664960 torch/distributed/run.py:757]
18
+ W0702 18:57:52.390000 140570191664960 torch/distributed/run.py:757] *****************************************
19
+ W0702 18:57:52.390000 140570191664960 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 18:57:52.390000 140570191664960 torch/distributed/run.py:757] *****************************************
21
+ W0702 18:57:55.127000 139912312178496 torch/distributed/run.py:757]
22
+ W0702 18:57:55.127000 139912312178496 torch/distributed/run.py:757] *****************************************
23
+ W0702 18:57:55.127000 139912312178496 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 18:57:55.127000 139912312178496 torch/distributed/run.py:757] *****************************************
25
+ [default0]:07/02/2024 18:58:16 [WARNING|DP=0|PP=0|TP=0|ip-26-0-165-24]: [Vocab Size Padding] Padded vocab (size: 50257) with 1 dummy tokens (new size: 50258)
26
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Config:
27
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Config(general=GeneralArgs(project='bench_cluster',
28
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: run='%date_%jobid',
29
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: seed=42,
30
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: step=None,
31
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: consumed_train_samples=None,
32
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: benchmark_csv_path=None,
33
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: ignore_sanity_checks=True),
34
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: parallelism=ParallelismArgs(dp=1,
35
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: pp=8,
36
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tp=2,
37
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f1c504d4730>,
38
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
39
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tp_linear_async_communication=False,
40
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: expert_parallel_size=1),
41
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
42
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: eos_token_id=2,
43
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: hidden_act='silu',
44
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: hidden_size=2048,
45
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: initializer_range=0.02,
46
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: intermediate_size=4096,
47
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: is_llama_config=True,
48
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: max_position_embeddings=4096,
49
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: num_attention_heads=32,
50
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: num_hidden_layers=24,
51
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: num_key_value_heads=32,
52
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: pad_token_id=None,
53
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: pretraining_tp=1,
54
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: rms_norm_eps=1e-05,
55
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: rope_scaling=None,
56
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: rope_theta=10000.0,
57
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tie_word_embeddings=True,
58
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: use_cache=True,
59
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: vocab_size=50258),
60
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: init_method=RandomInit(std=0.025),
61
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: dtype=torch.bfloat16,
62
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: make_vocab_size_divisible_by=1,
63
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: ddp_bucket_cap_mb=25),
64
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
65
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tokenizer_revision=None,
66
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tokenizer_max_length=None),
67
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
68
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: checkpoint_interval=100000,
69
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: save_initial_state=False,
70
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: resume_checkpoint_path=None,
71
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: checkpoints_path_is_shared_file_system=False),
72
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: logging=LoggingArgs(log_level='info',
73
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: log_level_replica='info',
74
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration_step_info_interval=1),
75
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tokens=TokensArgs(sequence_length=4096,
76
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: train_steps=20,
77
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: micro_batch_size=16,
78
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: batch_accumulation_per_replica=64,
79
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: val_check_interval=-1,
80
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: limit_val_batches=0,
81
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: limit_test_batches=0),
82
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
83
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: adam_beta1=0.9,
84
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: adam_beta2=0.95,
85
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: torch_adam_is_fused=True,
86
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: name='adamW'),
87
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: zero_stage=1,
88
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: weight_decay=0.01,
89
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: clip_grad=1.0,
90
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: accumulate_grad_in_fp32=True,
91
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
92
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: lr_warmup_steps=1,
93
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: lr_warmup_style='linear',
94
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: lr_decay_style='linear',
95
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: lr_decay_steps=19,
96
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: lr_decay_starting_step=None,
97
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: min_decay_lr=1e-05)),
98
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: data_stages=[DatasetStageArgs(name='Training Stage',
99
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: start_training_step=1,
100
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
101
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: hf_dataset_splits='train',
102
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: hf_dataset_config_name=None,
103
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: dataset_processing_num_proc_per_process=64,
104
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: dataset_overwrite_cache=False,
105
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: text_column_name='text'),
106
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: seed=42,
107
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: num_loading_workers=32))],
108
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-16')),
109
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: lighteval=None)
110
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Model Config:
111
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: LlamaConfig(bos_token_id=1,
112
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: eos_token_id=2,
113
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: hidden_act='silu',
114
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: hidden_size=2048,
115
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: initializer_range=0.02,
116
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: intermediate_size=4096,
117
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: is_llama_config=True,
118
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: max_position_embeddings=4096,
119
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: num_attention_heads=32,
120
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: num_hidden_layers=24,
121
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: num_key_value_heads=32,
122
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: pad_token_id=None,
123
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: pretraining_tp=1,
124
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: rms_norm_eps=1e-05,
125
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: rope_scaling=None,
126
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: rope_theta=10000.0,
127
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tie_word_embeddings=True,
128
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: use_cache=True,
129
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: vocab_size=50258)
130
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Building model..
131
+ [default0]:07/02/2024 18:58:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Setting PP block ranks...
132
+ [default7]:07/02/2024 18:58:32 [INFO|DP=0|PP=7|TP=1|ip-26-0-170-160]: Local number of parameters: 51.5M (98.16MiB)
133
+ [default4]:07/02/2024 18:58:32 [INFO|DP=0|PP=6|TP=0|ip-26-0-170-160]: Local number of parameters: 83.9M (160.03MiB)
134
+ [default4]:07/02/2024 18:58:32 [INFO|DP=0|PP=6|TP=0|ip-26-0-170-160]: [After model building] Memory usage: 164.04MiB. Peak allocated: 166.07MiB Peak reserved: 180.00MiB
135
+ [default0]:07/02/2024 18:58:32 [INFO|DP=0|PP=4|TP=0|ip-26-0-170-160]: Local number of parameters: 62.9M (120.02MiB)
136
+ [default0]:07/02/2024 18:58:32 [INFO|DP=0|PP=4|TP=0|ip-26-0-170-160]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
137
+ [default0]:07/02/2024 18:58:32 [INFO|DP=0|PP=4|TP=0|ip-26-0-170-160]: No checkpoint path provided.
138
+ [default4]:07/02/2024 18:58:32 [INFO|DP=0|PP=6|TP=0|ip-26-0-170-160]: No checkpoint path provided.
139
+ [default7]:07/02/2024 18:58:32 [INFO|DP=0|PP=7|TP=1|ip-26-0-170-160]: [After model building] Memory usage: 98.17MiB. Peak allocated: 98.18MiB Peak reserved: 102.00MiB
140
+ [default7]:07/02/2024 18:58:32 [INFO|DP=0|PP=7|TP=1|ip-26-0-170-160]: No checkpoint path provided.
141
+ [default6]:07/02/2024 18:58:32 [INFO|DP=0|PP=7|TP=0|ip-26-0-170-160]: Local number of parameters: 51.5M (98.16MiB)
142
+ [default6]:07/02/2024 18:58:32 [INFO|DP=0|PP=7|TP=0|ip-26-0-170-160]: [After model building] Memory usage: 98.17MiB. Peak allocated: 98.18MiB Peak reserved: 102.00MiB
143
+ [default6]:07/02/2024 18:58:32 [INFO|DP=0|PP=7|TP=0|ip-26-0-170-160]: No checkpoint path provided.
144
+ [default1]:07/02/2024 18:58:32 [INFO|DP=0|PP=4|TP=1|ip-26-0-170-160]: Local number of parameters: 62.9M (120.02MiB)
145
+ [default1]:07/02/2024 18:58:32 [INFO|DP=0|PP=4|TP=1|ip-26-0-170-160]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
146
+ [default2]:07/02/2024 18:58:32 [INFO|DP=0|PP=5|TP=0|ip-26-0-170-160]: Local number of parameters: 62.9M (120.02MiB)
147
+ [default2]:07/02/2024 18:58:32 [INFO|DP=0|PP=5|TP=0|ip-26-0-170-160]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
148
+ [default2]:07/02/2024 18:58:32 [INFO|DP=0|PP=5|TP=0|ip-26-0-170-160]: No checkpoint path provided.
149
+ [default1]:07/02/2024 18:58:32 [INFO|DP=0|PP=4|TP=1|ip-26-0-170-160]: No checkpoint path provided.
150
+ [default5]:07/02/2024 18:58:32 [INFO|DP=0|PP=6|TP=1|ip-26-0-170-160]: Local number of parameters: 83.9M (160.03MiB)
151
+ [default5]:07/02/2024 18:58:32 [INFO|DP=0|PP=6|TP=1|ip-26-0-170-160]: [After model building] Memory usage: 164.04MiB. Peak allocated: 166.07MiB Peak reserved: 180.00MiB
152
+ [default5]:07/02/2024 18:58:32 [INFO|DP=0|PP=6|TP=1|ip-26-0-170-160]: No checkpoint path provided.
153
+ [default3]:07/02/2024 18:58:32 [INFO|DP=0|PP=5|TP=1|ip-26-0-170-160]: Local number of parameters: 62.9M (120.02MiB)
154
+ [default3]:07/02/2024 18:58:32 [INFO|DP=0|PP=5|TP=1|ip-26-0-170-160]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
155
+ [default3]:07/02/2024 18:58:32 [INFO|DP=0|PP=5|TP=1|ip-26-0-170-160]: No checkpoint path provided.
156
+ [default3]:07/02/2024 18:58:32 [INFO|DP=0|PP=1|TP=1|ip-26-0-165-24]: Local number of parameters: 62.9M (120.02MiB)
157
+ [default3]:07/02/2024 18:58:32 [INFO|DP=0|PP=1|TP=1|ip-26-0-165-24]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
158
+ [default0]:07/02/2024 18:58:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Total number of parameters: 1.21G (2313.02MiB)
159
+ [default0]:07/02/2024 18:58:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Local number of parameters: 135M (258.19MiB)
160
+ [default0]:07/02/2024 18:58:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [After model building] Memory usage: 262.20MiB. Peak allocated: 264.23MiB Peak reserved: 280.00MiB
161
+ [default3]:07/02/2024 18:58:32 [INFO|DP=0|PP=1|TP=1|ip-26-0-165-24]: No checkpoint path provided.
162
+ [default0]:07/02/2024 18:58:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: No checkpoint path provided.
163
+ [default0]:07/02/2024 18:58:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Parametrizing model parameters using StandardParametrizator
164
+ [default5]:07/02/2024 18:58:32 [INFO|DP=0|PP=2|TP=1|ip-26-0-165-24]: Local number of parameters: 62.9M (120.02MiB)
165
+ [default5]:07/02/2024 18:58:32 [INFO|DP=0|PP=2|TP=1|ip-26-0-165-24]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
166
+ [default4]:07/02/2024 18:58:32 [INFO|DP=0|PP=2|TP=0|ip-26-0-165-24]: Local number of parameters: 62.9M (120.02MiB)
167
+ [default4]:07/02/2024 18:58:32 [INFO|DP=0|PP=2|TP=0|ip-26-0-165-24]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
168
+ [default7]:07/02/2024 18:58:32 [INFO|DP=0|PP=3|TP=1|ip-26-0-165-24]: Local number of parameters: 83.9M (160.03MiB)
169
+ [default7]:07/02/2024 18:58:32 [INFO|DP=0|PP=3|TP=1|ip-26-0-165-24]: [After model building] Memory usage: 164.04MiB. Peak allocated: 166.07MiB Peak reserved: 180.00MiB
170
+ [default7]:07/02/2024 18:58:32 [INFO|DP=0|PP=3|TP=1|ip-26-0-165-24]: No checkpoint path provided.
171
+ [default6]:07/02/2024 18:58:32 [INFO|DP=0|PP=3|TP=0|ip-26-0-165-24]: Local number of parameters: 83.9M (160.03MiB)
172
+ [default6]:07/02/2024 18:58:32 [INFO|DP=0|PP=3|TP=0|ip-26-0-165-24]: [After model building] Memory usage: 164.04MiB. Peak allocated: 166.07MiB Peak reserved: 180.00MiB
173
+ [default6]:07/02/2024 18:58:32 [INFO|DP=0|PP=3|TP=0|ip-26-0-165-24]: No checkpoint path provided.
174
+ [default5]:07/02/2024 18:58:32 [INFO|DP=0|PP=2|TP=1|ip-26-0-165-24]: No checkpoint path provided.
175
+ [default1]:07/02/2024 18:58:32 [INFO|DP=0|PP=0|TP=1|ip-26-0-165-24]: Local number of parameters: 135M (258.19MiB)
176
+ [default1]:07/02/2024 18:58:32 [INFO|DP=0|PP=0|TP=1|ip-26-0-165-24]: [After model building] Memory usage: 262.20MiB. Peak allocated: 264.23MiB Peak reserved: 280.00MiB
177
+ [default4]:07/02/2024 18:58:32 [INFO|DP=0|PP=2|TP=0|ip-26-0-165-24]: No checkpoint path provided.
178
+ [default2]:07/02/2024 18:58:32 [INFO|DP=0|PP=1|TP=0|ip-26-0-165-24]: Local number of parameters: 62.9M (120.02MiB)
179
+ [default2]:07/02/2024 18:58:32 [INFO|DP=0|PP=1|TP=0|ip-26-0-165-24]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
180
+ [default1]:07/02/2024 18:58:32 [INFO|DP=0|PP=0|TP=1|ip-26-0-165-24]: No checkpoint path provided.
181
+ [default2]:07/02/2024 18:58:32 [INFO|DP=0|PP=1|TP=0|ip-26-0-165-24]: No checkpoint path provided.
182
+ [default0]:07/02/2024 18:58:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [Optimizer Building] Using LearningRateForSP as learning rate
183
+ [default0]:07/02/2024 18:58:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [ZeRO sharding] Size of optimizer params per rank:
184
+ [default0]:07/02/2024 18:58:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [ZeRO sharding] DP Rank 0 has 135M out of 135M (100.00%) params' optimizer states
185
+ [default0]:07/02/2024 18:58:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
186
+ [default0]:07/02/2024 18:58:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Using `datasets` library
187
+ [default0]:07/02/2024 18:58:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
188
+ [default0]:07/02/2024 18:58:35 [WARNING|DP=0|PP=0|TP=0|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
189
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
190
+ [default0]:07/02/2024 18:58:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [Training Plan] There are 1 training stages
191
+ [default0]:07/02/2024 18:58:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [Stage Training Stage] start from step 1
192
+ [default0]:07/02/2024 18:58:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]:
193
+ [default0]:07/02/2024 18:58:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [Start training] datetime: 2024-07-02 18:58:36.988045 | mbs: 16 | grad_accum: 64 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
194
+ [default0]:07/02/2024 18:58:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
195
+ [default0]:07/02/2024 18:58:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 1294.97MiB. Peak allocated 1294.97MiB. Peak reserved: 1316.00MiB
196
+ [default6]:07/02/2024 18:58:37 [WARNING|DP=0|PP=7|TP=0|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
197
+ [default0]:07/02/2024 18:58:37 [WARNING|DP=0|PP=4|TP=0|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
198
+ [default7]:07/02/2024 18:58:37 [WARNING|DP=0|PP=7|TP=1|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
199
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
200
+ [default1]:07/02/2024 18:58:37 [WARNING|DP=0|PP=4|TP=1|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
201
+ [default2]:07/02/2024 18:58:37 [WARNING|DP=0|PP=5|TP=0|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
202
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
203
+ [default5]:07/02/2024 18:58:37 [WARNING|DP=0|PP=6|TP=1|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
204
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
205
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
206
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
207
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
208
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
209
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
210
+ [default6]:07/02/2024 18:58:37 [WARNING|DP=0|PP=3|TP=0|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
211
+ [default4]:07/02/2024 18:58:37 [WARNING|DP=0|PP=2|TP=0|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
212
+ [default5]:07/02/2024 18:58:37 [WARNING|DP=0|PP=2|TP=1|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
213
+ [default7]:07/02/2024 18:58:37 [WARNING|DP=0|PP=3|TP=1|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
214
+ [default2]:07/02/2024 18:58:37 [WARNING|DP=0|PP=1|TP=0|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
215
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
216
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
217
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
218
+ [default3]:07/02/2024 18:58:37 [WARNING|DP=0|PP=1|TP=1|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
219
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
220
+ [default4]:07/02/2024 18:58:37 [WARNING|DP=0|PP=6|TP=0|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
221
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
222
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
223
+ [default3]:07/02/2024 18:58:37 [WARNING|DP=0|PP=5|TP=1|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
224
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
225
+ [default1]:07/02/2024 18:58:37 [WARNING|DP=0|PP=0|TP=1|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
226
+ [default1]:[rank1]: Traceback (most recent call last):
227
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
228
+ [default1]:[rank1]: trainer.train(dataloader)
229
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
230
+ [default1]:[rank1]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
231
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
232
+ [default1]:[rank1]: outputs = self.pipeline_engine.train_batch_iter(
233
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 252, in train_batch_iter
234
+ [default1]:[rank1]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
235
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
236
+ [default1]:[rank1]: output = model(**micro_batch)
237
+ [default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
238
+ [default1]:[rank1]: return self._call_impl(*args, **kwargs)
239
+ [default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
240
+ [default1]:[rank1]: return forward_call(*args, **kwargs)
241
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
242
+ [default1]:[rank1]: sharded_logits = self.model(
243
+ [default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
244
+ [default1]:[rank1]: return self._call_impl(*args, **kwargs)
245
+ [default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
246
+ [default1]:[rank1]: return forward_call(*args, **kwargs)
247
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
248
+ [default1]:[rank1]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
249
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
250
+ [default1]:[rank1]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
251
+ [default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
252
+ [default1]:[rank1]: return self._call_impl(*args, **kwargs)
253
+ [default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
254
+ [default1]:[rank1]: return forward_call(*args, **kwargs)
255
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
256
+ [default1]:[rank1]: output = self.pp_block(**new_kwargs)
257
+ [default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
258
+ [default1]:[rank1]: return self._call_impl(*args, **kwargs)
259
+ [default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
260
+ [default1]:[rank1]: return forward_call(*args, **kwargs)
261
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 631, in forward
262
+ [default1]:[rank1]: output = self.attn(hidden_states=hidden_states, sequence_mask=sequence_mask)
263
+ [default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
264
+ [default1]:[rank1]: return self._call_impl(*args, **kwargs)
265
+ [default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
266
+ [default1]:[rank1]: return forward_call(*args, **kwargs)
267
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 563, in forward
268
+ [default1]:[rank1]: key_value_states = torch.cat([key_states.unsqueeze(0), value_states.unsqueeze(0)], dim=0)
269
+ [default1]:[rank1]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 256.00 MiB. GPU  has a total capacity of 79.33 GiB of which 51.94 MiB is free. Including non-PyTorch memory, this process has 79.26 GiB memory in use. Of the allocated memory 70.91 GiB is allocated by PyTorch, and 297.81 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
270
+ [default0]:[rank0]: Traceback (most recent call last):
271
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
272
+ [default0]:[rank0]: trainer.train(dataloader)
273
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
274
+ [default0]:[rank0]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
275
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
276
+ [default0]:[rank0]: outputs = self.pipeline_engine.train_batch_iter(
277
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 252, in train_batch_iter
278
+ [default0]:[rank0]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
279
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
280
+ [default0]:[rank0]: output = model(**micro_batch)
281
+ [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
282
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
283
+ [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
284
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
285
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
286
+ [default0]:[rank0]: sharded_logits = self.model(
287
+ [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
288
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
289
+ [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
290
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
291
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
292
+ [default0]:[rank0]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
293
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
294
+ [default0]:[rank0]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
295
+ [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
296
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
297
+ [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
298
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
299
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
300
+ [default0]:[rank0]: output = self.pp_block(**new_kwargs)
301
+ [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
302
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
303
+ [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
304
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
305
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 631, in forward
306
+ [default0]:[rank0]: output = self.attn(hidden_states=hidden_states, sequence_mask=sequence_mask)
307
+ [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
308
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
309
+ [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
310
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
311
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 565, in forward
312
+ [default0]:[rank0]: key_value_states = key_value_states.permute(1, 2, 0, 3, 4).contiguous()
313
+ [default0]:[rank0]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 256.00 MiB. GPU
314
+ W0702 18:59:00.741000 140570191664960 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 676602 closing signal SIGTERM
315
+ W0702 18:59:00.741000 140570191664960 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 676603 closing signal SIGTERM
316
+ W0702 18:59:00.742000 140570191664960 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 676604 closing signal SIGTERM
317
+ W0702 18:59:00.742000 140570191664960 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 676605 closing signal SIGTERM
318
+ W0702 18:59:00.743000 140570191664960 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 676606 closing signal SIGTERM
319
+ W0702 18:59:00.743000 140570191664960 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 676607 closing signal SIGTERM
320
+ [default7]:/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.)
321
+ [default7]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
322
+ [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.)
323
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
324
+ [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.)
325
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
326
+ [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.)
327
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
328
+ E0702 18:59:03.171000 140570191664960 torch/distributed/elastic/multiprocessing/api.py:826] failed (exitcode: 1) local_rank: 0 (pid: 676600) of binary: /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/python3.10
329
+ Traceback (most recent call last):
330
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/torchrun", line 8, in <module>
331
+ sys.exit(main())
332
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 347, in wrapper
333
+ return f(*args, **kwargs)
334
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 879, in main
335
+ run(args)
336
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 870, in run
337
+ elastic_launch(
338
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 132, in __call__
339
+ return launch_agent(self._config, self._entrypoint, list(args))
340
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 263, in launch_agent
341
+ raise ChildFailedError(
342
+ torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
343
+ ============================================================
344
+ /fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py FAILED
345
+ ------------------------------------------------------------
346
+ Failures:
347
+ [1]:
348
+ time : 2024-07-02_18:59:00
349
+ host : ip-26-0-165-24.ec2.internal
350
+ rank : 1 (local_rank: 1)
351
+ exitcode : 1 (pid: 676601)
352
+ error_file: <N/A>
353
+ traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
354
+ ------------------------------------------------------------
355
+ Root Cause (first observed failure):
356
+ [0]:
357
+ time : 2024-07-02_18:59:00
358
+ host : ip-26-0-165-24.ec2.internal
359
+ rank : 0 (local_rank: 0)
360
+ exitcode : 1 (pid: 676600)
361
+ error_file: <N/A>
362
+ traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
363
+ ============================================================
364
+ srun: error: ip-26-0-165-24: task 0: Exited with exit code 1
365
+ [default2]:/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.)
366
+ [default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
367
+ [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.)
368
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
369
+ W0702 18:59:05.401000 139906645358336 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1252] The node 'ip-26-0-170-160.ec2.internal_700100_0' has failed to send a keep-alive heartbeat to the rendezvous 'none' due to an error of type RendezvousConnectionError.
370
+ W0702 18:59:05.756000 139912312178496 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 700172 closing signal SIGTERM
371
+ W0702 18:59:05.756000 139912312178496 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 700173 closing signal SIGTERM
372
+ W0702 18:59:05.757000 139912312178496 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 700174 closing signal SIGTERM
373
+ W0702 18:59:05.758000 139912312178496 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 700175 closing signal SIGTERM
374
+ W0702 18:59:05.759000 139912312178496 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 700176 closing signal SIGTERM
375
+ W0702 18:59:05.760000 139912312178496 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 700177 closing signal SIGTERM
376
+ W0702 18:59:05.761000 139912312178496 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 700178 closing signal SIGTERM
377
+ W0702 18:59:05.766000 139912312178496 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 700179 closing signal SIGTERM
378
+ W0702 18:59:09.385000 139912312178496 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1203] The node 'ip-26-0-170-160.ec2.internal_700100_0' has failed to shutdown the rendezvous 'none' due to an error of type RendezvousConnectionError.
379
+ W0702 18:59:09.395000 139912312178496 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1203] The node 'ip-26-0-170-160.ec2.internal_700100_0' has failed to shutdown the rendezvous 'none' due to an error of type RendezvousConnectionError.
380
+ Traceback (most recent call last):
381
+ 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
382
+ return getattr(self._store, store_op)(*args, **kwargs)
383
+ torch.distributed.DistNetworkError: Broken pipe
384
+
385
+ The above exception was the direct cause of the following exception:
386
+
387
+ Traceback (most recent call last):
388
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/torchrun", line 8, in <module>
389
+ sys.exit(main())
390
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 347, in wrapper
391
+ return f(*args, **kwargs)
392
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 879, in main
393
+ run(args)
394
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 870, in run
395
+ elastic_launch(
396
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 132, in __call__
397
+ return launch_agent(self._config, self._entrypoint, list(args))
398
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 254, in launch_agent
399
+ result = agent.run()
400
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 123, in wrapper
401
+ result = f(*args, **kwargs)
402
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 733, in run
403
+ result = self._invoke_run(role)
404
+ 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
405
+ num_nodes_waiting = rdzv_handler.num_nodes_waiting()
406
+ 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
407
+ self._state_holder.sync()
408
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/rendezvous/dynamic_rendezvous.py", line 419, in sync
409
+ get_response = self._backend.get_state()
410
+ 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
411
+ base64_state: bytes = self._call_store("get", self._key)
412
+ 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
413
+ raise RendezvousConnectionError(
414
+ torch.distributed.elastic.rendezvous.api.RendezvousConnectionError: The connection to the C10d store has failed. See inner exception for details.
415
+ srun: error: ip-26-0-170-160: task 1: Exited with exit code 1
416
+ 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-2_pp-8_mbz-16/status.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ oom