Hugging Face's logo

nanotron
/
old_bench

Model card Files Community
3outeille HF staff commited on
Commit
fa1c045
·
verified ·
1 Parent(s): 95c71a9

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

Browse files
Files changed (4) hide show
  1. llama-1B/8_GPUS/dp-2_tp-2_pp-2_mbz-16/bench.slurm +111 -0
  2. llama-1B/8_GPUS/dp-2_tp-2_pp-2_mbz-16/config.yaml +90 -0
  3. llama-1B/8_GPUS/dp-2_tp-2_pp-2_mbz-16/log.out +427 -0
  4. llama-1B/8_GPUS/dp-2_tp-2_pp-2_mbz-16/status.txt +1 -0
llama-1B/8_GPUS/dp-2_tp-2_pp-2_mbz-16/bench.slurm ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+
3
+ #SBATCH --job-name=bench_cluster
4
+ #SBATCH --time=02:00:00
5
+ #SBATCH --partition=hopper-prod
6
+ #SBATCH --nodes=1
7
+ #SBATCH --gres=gpu:8
8
+ #SBATCH --qos=normal
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/8_GPUS/dp-2_tp-2_pp-2_mbz-16/log.out
13
+ #SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-2_pp-2_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/8_GPUS/dp-2_tp-2_pp-2_mbz-16/config.yaml"
57
+
58
+ LAUNCHER="torchrun \
59
+ --nproc_per_node 8 \
60
+ --nnodes 1 \
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/8_GPUS/dp-2_tp-2_pp-2_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/8_GPUS/dp-2_tp-2_pp-2_mbz-16/status.txt
84
+ else
85
+ if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-2_pp-2_mbz-16/log.out; then
86
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-2_pp-2_mbz-16/status.txt
87
+ elif grep -q " CUDA error: an illegal memory access" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-2_pp-2_mbz-16/log.out; then
88
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-2_pp-2_mbz-16/status.txt
89
+ elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-2_pp-2_mbz-16/log.out; then
90
+ printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-2_pp-2_mbz-16/status.txt
91
+ else
92
+ printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-2_pp-2_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/8_GPUS/dp-2_tp-2_pp-2_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/8_GPUS/dp-2_tp-2_pp-2_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/8_GPUS/dp-2_tp-2_pp-2_mbz-16 llama-1B/8_GPUS/dp-2_tp-2_pp-2_mbz-16 --commit-message "Upload llama-1B/8_GPUS/dp-2_tp-2_pp-2_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/8_GPUS/dp-2_tp-2_pp-2_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: 2
49
+ expert_parallel_size: 1
50
+ pp: 2
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/8_GPUS/dp-2_tp-2_pp-2_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: 0
73
+ seed: 42
74
+ lighteval: null
75
+ tokens:
76
+ train_steps: 20
77
+ val_check_interval: -1
78
+ batch_accumulation_per_replica: 32
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/8_GPUS/dp-2_tp-2_pp-2_mbz-16/log.out ADDED
@@ -0,0 +1,427 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ========================
2
+ START TIME: Thu Jul 4 00:01:38 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
+ W0704 00:01:41.514000 139693704775488 torch/distributed/run.py:757]
18
+ W0704 00:01:41.514000 139693704775488 torch/distributed/run.py:757] *****************************************
19
+ W0704 00:01:41.514000 139693704775488 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
+ W0704 00:01:41.514000 139693704775488 torch/distributed/run.py:757] *****************************************
21
+ [default0]:07/04/2024 00:01:57 [WARNING|DP=0|PP=0|TP=0|ip-26-0-169-139]: [Vocab Size Padding] Padded vocab (size: 50257) with 1 dummy tokens (new size: 50258)
22
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: Config:
23
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: Config(general=GeneralArgs(project='bench_cluster',
24
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: run='%date_%jobid',
25
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: seed=42,
26
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: step=None,
27
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: consumed_train_samples=None,
28
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: benchmark_csv_path=None,
29
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: ignore_sanity_checks=True),
30
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: parallelism=ParallelismArgs(dp=2,
31
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: pp=2,
32
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: tp=2,
33
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f2bf27a5090>,
34
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
35
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: tp_linear_async_communication=False,
36
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: expert_parallel_size=1),
37
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
38
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: eos_token_id=2,
39
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: hidden_act='silu',
40
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: hidden_size=2048,
41
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: initializer_range=0.02,
42
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: intermediate_size=4096,
43
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: is_llama_config=True,
44
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: max_position_embeddings=4096,
45
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: num_attention_heads=32,
46
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: num_hidden_layers=24,
47
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: num_key_value_heads=32,
48
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: pad_token_id=None,
49
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: pretraining_tp=1,
50
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: rms_norm_eps=1e-05,
51
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: rope_scaling=None,
52
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: rope_theta=10000.0,
53
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: tie_word_embeddings=True,
54
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: use_cache=True,
55
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: vocab_size=50258),
56
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: init_method=RandomInit(std=0.025),
57
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: dtype=torch.bfloat16,
58
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: make_vocab_size_divisible_by=1,
59
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: ddp_bucket_cap_mb=25),
60
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
61
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: tokenizer_revision=None,
62
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: tokenizer_max_length=None),
63
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
64
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: checkpoint_interval=100000,
65
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: save_initial_state=False,
66
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: resume_checkpoint_path=None,
67
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: checkpoints_path_is_shared_file_system=False),
68
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: logging=LoggingArgs(log_level='info',
69
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: log_level_replica='info',
70
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: iteration_step_info_interval=1),
71
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: tokens=TokensArgs(sequence_length=4096,
72
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: train_steps=20,
73
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: micro_batch_size=16,
74
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: batch_accumulation_per_replica=32,
75
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: val_check_interval=-1,
76
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: limit_val_batches=0,
77
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: limit_test_batches=0),
78
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
79
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: adam_beta1=0.9,
80
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: adam_beta2=0.95,
81
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: torch_adam_is_fused=True,
82
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: name='adamW'),
83
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: zero_stage=1,
84
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: weight_decay=0.01,
85
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: clip_grad=1.0,
86
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: accumulate_grad_in_fp32=True,
87
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
88
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: lr_warmup_steps=1,
89
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: lr_warmup_style='linear',
90
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: lr_decay_style='linear',
91
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: lr_decay_steps=19,
92
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: lr_decay_starting_step=None,
93
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: min_decay_lr=1e-05)),
94
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: data_stages=[DatasetStageArgs(name='Training Stage',
95
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: start_training_step=1,
96
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
97
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: hf_dataset_splits='train',
98
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: hf_dataset_config_name=None,
99
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: dataset_processing_num_proc_per_process=64,
100
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: dataset_overwrite_cache=False,
101
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: text_column_name='text'),
102
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: seed=42,
103
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: num_loading_workers=0))],
104
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-2_pp-2_mbz-16')),
105
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: lighteval=None)
106
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: Model Config:
107
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: LlamaConfig(bos_token_id=1,
108
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: eos_token_id=2,
109
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: hidden_act='silu',
110
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: hidden_size=2048,
111
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: initializer_range=0.02,
112
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: intermediate_size=4096,
113
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: is_llama_config=True,
114
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: max_position_embeddings=4096,
115
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: num_attention_heads=32,
116
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: num_hidden_layers=24,
117
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: num_key_value_heads=32,
118
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: pad_token_id=None,
119
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: pretraining_tp=1,
120
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: rms_norm_eps=1e-05,
121
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: rope_scaling=None,
122
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: rope_theta=10000.0,
123
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: tie_word_embeddings=True,
124
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: use_cache=True,
125
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: vocab_size=50258)
126
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: Building model..
127
+ [default0]:07/04/2024 00:01:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: Setting PP block ranks...
128
+ [default0]:07/04/2024 00:02:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: Total number of parameters: 1.21G (2313.02MiB)
129
+ [default0]:07/04/2024 00:02:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: Local number of parameters: 345M (658.27MiB)
130
+ [default0]:07/04/2024 00:02:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: [After model building] Memory usage: 672.29MiB. Peak allocated: 674.32MiB Peak reserved: 690.00MiB
131
+ [default0]:07/04/2024 00:02:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: No checkpoint path provided.
132
+ [default0]:07/04/2024 00:02:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: Parametrizing model parameters using StandardParametrizator
133
+ [default3]:07/04/2024 00:02:09 [INFO|DP=1|PP=0|TP=1|ip-26-0-169-139]: No checkpoint path provided.
134
+ [default2]:07/04/2024 00:02:09 [INFO|DP=1|PP=0|TP=0|ip-26-0-169-139]: No checkpoint path provided.
135
+ [default4]:07/04/2024 00:02:09 [INFO|DP=0|PP=1|TP=0|ip-26-0-169-139]: Local number of parameters: 261M (498.24MiB)
136
+ [default4]:07/04/2024 00:02:09 [INFO|DP=0|PP=1|TP=0|ip-26-0-169-139]: [After model building] Memory usage: 508.26MiB. Peak allocated: 510.29MiB Peak reserved: 526.00MiB
137
+ [default4]:07/04/2024 00:02:09 [INFO|DP=0|PP=1|TP=0|ip-26-0-169-139]: No checkpoint path provided.
138
+ [default1]:07/04/2024 00:02:09 [INFO|DP=0|PP=0|TP=1|ip-26-0-169-139]: Local number of parameters: 345M (658.27MiB)
139
+ [default1]:07/04/2024 00:02:09 [INFO|DP=0|PP=0|TP=1|ip-26-0-169-139]: [After model building] Memory usage: 672.29MiB. Peak allocated: 674.32MiB Peak reserved: 690.00MiB
140
+ [default1]:07/04/2024 00:02:09 [INFO|DP=0|PP=0|TP=1|ip-26-0-169-139]: No checkpoint path provided.
141
+ [default5]:07/04/2024 00:02:09 [INFO|DP=0|PP=1|TP=1|ip-26-0-169-139]: Local number of parameters: 261M (498.24MiB)
142
+ [default5]:07/04/2024 00:02:09 [INFO|DP=0|PP=1|TP=1|ip-26-0-169-139]: [After model building] Memory usage: 508.26MiB. Peak allocated: 510.29MiB Peak reserved: 526.00MiB
143
+ [default5]:07/04/2024 00:02:09 [INFO|DP=0|PP=1|TP=1|ip-26-0-169-139]: No checkpoint path provided.
144
+ [default7]:07/04/2024 00:02:09 [INFO|DP=1|PP=1|TP=1|ip-26-0-169-139]: No checkpoint path provided.
145
+ [default6]:07/04/2024 00:02:09 [INFO|DP=1|PP=1|TP=0|ip-26-0-169-139]: No checkpoint path provided.
146
+ [default0]:07/04/2024 00:02:11 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: [Optimizer Building] Using LearningRateForSP as learning rate
147
+ [default0]:07/04/2024 00:02:11 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: [ZeRO sharding] Size of optimizer params per rank:
148
+ [default0]:07/04/2024 00:02:11 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: [ZeRO sharding] DP Rank 0 has 173M out of 345M (50.00%) params' optimizer states
149
+ [default0]:07/04/2024 00:02:11 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: [ZeRO sharding] DP Rank 1 has 173M out of 345M (50.00%) params' optimizer states
150
+ [default0]:07/04/2024 00:02:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
151
+ [default0]:07/04/2024 00:02:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: Using `datasets` library
152
+ [default0]:07/04/2024 00:02:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
153
+ [default0]:07/04/2024 00:02:13 [WARNING|DP=0|PP=0|TP=0|ip-26-0-169-139]: Repo card metadata block was not found. Setting CardData to empty.
154
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
155
+ [default0]:07/04/2024 00:02:14 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: [Training Plan] There are 1 training stages
156
+ [default0]:07/04/2024 00:02:14 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: [Stage Training Stage] start from step 1
157
+ [default0]:07/04/2024 00:02:14 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]:
158
+ [default0]:07/04/2024 00:02:14 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: [Start training] datetime: 2024-07-04 00:02:14.232785 | mbs: 16 | grad_accum: 32 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
159
+ [default0]:07/04/2024 00:02:14 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
160
+ [default0]:07/04/2024 00:02:14 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-139]: Memory usage: 2647.09MiB. Peak allocated 2647.09MiB. Peak reserved: 2668.00MiB
161
+ [default5]:07/04/2024 00:02:14 [WARNING|DP=0|PP=1|TP=1|ip-26-0-169-139]: Repo card metadata block was not found. Setting CardData to empty.
162
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
163
+ [default3]:07/04/2024 00:02:14 [WARNING|DP=1|PP=0|TP=1|ip-26-0-169-139]: Repo card metadata block was not found. Setting CardData to empty.
164
+ [default2]:07/04/2024 00:02:14 [WARNING|DP=1|PP=0|TP=0|ip-26-0-169-139]: Repo card metadata block was not found. Setting CardData to empty.
165
+ [default4]:07/04/2024 00:02:14 [WARNING|DP=0|PP=1|TP=0|ip-26-0-169-139]: Repo card metadata block was not found. Setting CardData to empty.
166
+ [default1]:07/04/2024 00:02:14 [WARNING|DP=0|PP=0|TP=1|ip-26-0-169-139]: Repo card metadata block was not found. Setting CardData to empty.
167
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
168
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
169
+ [default6]:07/04/2024 00:02:14 [WARNING|DP=1|PP=1|TP=0|ip-26-0-169-139]: Repo card metadata block was not found. Setting CardData to empty.
170
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
171
+ [default7]:07/04/2024 00:02:14 [WARNING|DP=1|PP=1|TP=1|ip-26-0-169-139]: Repo card metadata block was not found. Setting CardData to empty.
172
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
173
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
174
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
175
+ [default0]:[rank0]: Traceback (most recent call last):
176
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
177
+ [default0]:[rank0]: trainer.train(dataloader)
178
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
179
+ [default0]:[rank0]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
180
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
181
+ [default0]:[rank0]: outputs = self.pipeline_engine.train_batch_iter(
182
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
183
+ [default0]:[rank0]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
184
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
185
+ [default0]:[rank0]: output = model(**micro_batch)
186
+ [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
187
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
188
+ [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
189
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
190
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
191
+ [default0]:[rank0]: sharded_logits = self.model(
192
+ [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
193
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
194
+ [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
195
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
196
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
197
+ [default0]:[rank0]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
198
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
199
+ [default0]:[rank0]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
200
+ [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
201
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
202
+ [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
203
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
204
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
205
+ [default0]:[rank0]: output = self.pp_block(**new_kwargs)
206
+ [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
207
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
208
+ [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
209
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
210
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
211
+ [default0]:[rank0]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
212
+ [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
213
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
214
+ [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
215
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
216
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
217
+ [default0]:[rank0]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
218
+ [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
219
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
220
+ [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
221
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
222
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 128, in forward
223
+ [default0]:[rank0]: return self.act(gate_states) * up_states
224
+ [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
225
+ [default0]:[rank0]: return self._call_impl(*args, **kwargs)
226
+ [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
227
+ [default0]:[rank0]: return forward_call(*args, **kwargs)
228
+ [default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/nn/activations.py", line 149, in forward
229
+ [default0]:[rank0]: return nn.functional.silu(input)
230
+ [default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/functional.py", line 2102, in silu
231
+ [default0]:[rank0]: return torch._C._nn.silu(input)
232
+ [default0]:[rank0]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 256.00 MiB. GPU
233
+ [default1]:[rank1]: Traceback (most recent call last):
234
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
235
+ [default1]:[rank1]: trainer.train(dataloader)
236
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
237
+ [default1]:[rank1]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
238
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
239
+ [default1]:[rank1]: outputs = self.pipeline_engine.train_batch_iter(
240
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
241
+ [default1]:[rank1]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
242
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
243
+ [default1]:[rank1]: output = model(**micro_batch)
244
+ [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
245
+ [default1]:[rank1]: return self._call_impl(*args, **kwargs)
246
+ [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
247
+ [default1]:[rank1]: return forward_call(*args, **kwargs)
248
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
249
+ [default1]:[rank1]: sharded_logits = self.model(
250
+ [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
251
+ [default1]:[rank1]: return self._call_impl(*args, **kwargs)
252
+ [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
253
+ [default1]:[rank1]: return forward_call(*args, **kwargs)
254
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
255
+ [default1]:[rank1]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
256
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
257
+ [default1]:[rank1]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
258
+ [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
259
+ [default1]:[rank1]: return self._call_impl(*args, **kwargs)
260
+ [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
261
+ [default1]:[rank1]: return forward_call(*args, **kwargs)
262
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
263
+ [default1]:[rank1]: output = self.pp_block(**new_kwargs)
264
+ [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
265
+ [default1]:[rank1]: return self._call_impl(*args, **kwargs)
266
+ [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
267
+ [default1]:[rank1]: return forward_call(*args, **kwargs)
268
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
269
+ [default1]:[rank1]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
270
+ [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
271
+ [default1]:[rank1]: return self._call_impl(*args, **kwargs)
272
+ [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
273
+ [default1]:[rank1]: return forward_call(*args, **kwargs)
274
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
275
+ [default1]:[rank1]: merged_states = self.gate_up_proj(hidden_states)
276
+ [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
277
+ [default1]:[rank1]: return self._call_impl(*args, **kwargs)
278
+ [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
279
+ [default1]:[rank1]: return forward_call(*args, **kwargs)
280
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
281
+ [default1]:[rank1]: return column_linear(
282
+ [default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
283
+ [default1]:[rank1]: return F.linear(input, weight, bias)
284
+ [default1]:[rank1]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 512.00 MiB. GPU  has a total capacity of 79.33 GiB of which 497.94 MiB is free. Including non-PyTorch memory, this process has 78.83 GiB memory in use. Of the allocated memory 67.60 GiB is allocated by PyTorch, and 301.87 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)
285
+ [default3]:[rank3]: Traceback (most recent call last):
286
+ [default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
287
+ [default3]:[rank3]: trainer.train(dataloader)
288
+ [default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
289
+ [default3]:[rank3]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
290
+ [default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
291
+ [default3]:[rank3]: outputs = self.pipeline_engine.train_batch_iter(
292
+ [default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
293
+ [default3]:[rank3]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
294
+ [default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
295
+ [default3]:[rank3]: output = model(**micro_batch)
296
+ [default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
297
+ [default3]:[rank3]: return self._call_impl(*args, **kwargs)
298
+ [default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
299
+ [default3]:[rank3]: return forward_call(*args, **kwargs)
300
+ [default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
301
+ [default3]:[rank3]: sharded_logits = self.model(
302
+ [default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
303
+ [default3]:[rank3]: return self._call_impl(*args, **kwargs)
304
+ [default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
305
+ [default3]:[rank3]: return forward_call(*args, **kwargs)
306
+ [default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
307
+ [default3]:[rank3]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
308
+ [default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
309
+ [default3]:[rank3]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
310
+ [default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
311
+ [default3]:[rank3]: return self._call_impl(*args, **kwargs)
312
+ [default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
313
+ [default3]:[rank3]: return forward_call(*args, **kwargs)
314
+ [default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
315
+ [default3]:[rank3]: output = self.pp_block(**new_kwargs)
316
+ [default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
317
+ [default3]:[rank3]: return self._call_impl(*args, **kwargs)
318
+ [default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
319
+ [default3]:[rank3]: return forward_call(*args, **kwargs)
320
+ [default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
321
+ [default3]:[rank3]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
322
+ [default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
323
+ [default3]:[rank3]: return self._call_impl(*args, **kwargs)
324
+ [default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
325
+ [default3]:[rank3]: return forward_call(*args, **kwargs)
326
+ [default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
327
+ [default3]:[rank3]: merged_states = self.gate_up_proj(hidden_states)
328
+ [default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
329
+ [default3]:[rank3]: return self._call_impl(*args, **kwargs)
330
+ [default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
331
+ [default3]:[rank3]: return forward_call(*args, **kwargs)
332
+ [default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
333
+ [default3]:[rank3]: return column_linear(
334
+ [default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
335
+ [default3]:[rank3]: return F.linear(input, weight, bias)
336
+ [default3]:[rank3]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 512.00 MiB. GPU  has a total capacity of 79.33 GiB of which 497.94 MiB is free. Including non-PyTorch memory, this process has 78.83 GiB memory in use. Of the allocated memory 67.60 GiB is allocated by PyTorch, and 301.87 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)
337
+ [default2]:[rank2]: Traceback (most recent call last):
338
+ [default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
339
+ [default2]:[rank2]: trainer.train(dataloader)
340
+ [default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
341
+ [default2]:[rank2]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
342
+ [default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
343
+ [default2]:[rank2]: outputs = self.pipeline_engine.train_batch_iter(
344
+ [default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
345
+ [default2]:[rank2]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
346
+ [default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
347
+ [default2]:[rank2]: output = model(**micro_batch)
348
+ [default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
349
+ [default2]:[rank2]: return self._call_impl(*args, **kwargs)
350
+ [default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
351
+ [default2]:[rank2]: return forward_call(*args, **kwargs)
352
+ [default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
353
+ [default2]:[rank2]: sharded_logits = self.model(
354
+ [default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
355
+ [default2]:[rank2]: return self._call_impl(*args, **kwargs)
356
+ [default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
357
+ [default2]:[rank2]: return forward_call(*args, **kwargs)
358
+ [default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
359
+ [default2]:[rank2]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
360
+ [default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
361
+ [default2]:[rank2]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
362
+ [default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
363
+ [default2]:[rank2]: return self._call_impl(*args, **kwargs)
364
+ [default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
365
+ [default2]:[rank2]: return forward_call(*args, **kwargs)
366
+ [default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
367
+ [default2]:[rank2]: output = self.pp_block(**new_kwargs)
368
+ [default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
369
+ [default2]:[rank2]: return self._call_impl(*args, **kwargs)
370
+ [default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
371
+ [default2]:[rank2]: return forward_call(*args, **kwargs)
372
+ [default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
373
+ [default2]:[rank2]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
374
+ [default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
375
+ [default2]:[rank2]: return self._call_impl(*args, **kwargs)
376
+ [default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
377
+ [default2]:[rank2]: return forward_call(*args, **kwargs)
378
+ [default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
379
+ [default2]:[rank2]: merged_states = self.gate_up_proj(hidden_states)
380
+ [default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
381
+ [default2]:[rank2]: return self._call_impl(*args, **kwargs)
382
+ [default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
383
+ [default2]:[rank2]: return forward_call(*args, **kwargs)
384
+ [default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
385
+ [default2]:[rank2]: return column_linear(
386
+ [default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
387
+ [default2]:[rank2]: return F.linear(input, weight, bias)
388
+ [default2]:[rank2]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 512.00 MiB. GPU  has a total capacity of 79.33 GiB of which 497.94 MiB is free. Including non-PyTorch memory, this process has 78.83 GiB memory in use. Of the allocated memory 67.60 GiB is allocated by PyTorch, and 301.87 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)
389
+ W0704 00:02:26.667000 139693704775488 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 677683 closing signal SIGTERM
390
+ W0704 00:02:26.668000 139693704775488 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 677684 closing signal SIGTERM
391
+ W0704 00:02:26.668000 139693704775488 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 677685 closing signal SIGTERM
392
+ W0704 00:02:26.668000 139693704775488 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 677686 closing signal SIGTERM
393
+ W0704 00:02:26.668000 139693704775488 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 677687 closing signal SIGTERM
394
+ W0704 00:02:26.668000 139693704775488 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 677688 closing signal SIGTERM
395
+ W0704 00:02:26.670000 139693704775488 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 677689 closing signal SIGTERM
396
+ E0704 00:02:28.584000 139693704775488 torch/distributed/elastic/multiprocessing/api.py:826] failed (exitcode: 1) local_rank: 0 (pid: 677682) of binary: /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/python3.10
397
+ Traceback (most recent call last):
398
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/torchrun", line 8, in <module>
399
+ sys.exit(main())
400
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 347, in wrapper
401
+ return f(*args, **kwargs)
402
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 879, in main
403
+ run(args)
404
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 870, in run
405
+ elastic_launch(
406
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 132, in __call__
407
+ return launch_agent(self._config, self._entrypoint, list(args))
408
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 263, in launch_agent
409
+ raise ChildFailedError(
410
+ torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
411
+ ============================================================
412
+ /fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py FAILED
413
+ ------------------------------------------------------------
414
+ Failures:
415
+ <NO_OTHER_FAILURES>
416
+ ------------------------------------------------------------
417
+ Root Cause (first observed failure):
418
+ [0]:
419
+ time : 2024-07-04_00:02:26
420
+ host : ip-26-0-169-139.ec2.internal
421
+ rank : 0 (local_rank: 0)
422
+ exitcode : 1 (pid: 677682)
423
+ error_file: <N/A>
424
+ traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
425
+ ============================================================
426
+ srun: error: ip-26-0-169-139: task 0: Exited with exit code 1
427
+ 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/8_GPUS/dp-2_tp-2_pp-2_mbz-16/status.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ oom