3outeille HF staff commited on
Commit
45a5376
1 Parent(s): 6266f26

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

Browse files
llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-2/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-1_tp-8_pp-1_mbz-2/log.out
13
+ #SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-2/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-1_tp-8_pp-1_mbz-2/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-1_tp-8_pp-1_mbz-2/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-1_tp-8_pp-1_mbz-2/status.txt
84
+ else
85
+ if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-2/log.out; then
86
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-2/status.txt
87
+ elif grep -q " CUDA error: an illegal memory access" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-2/log.out; then
88
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-2/status.txt
89
+ elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-2/log.out; then
90
+ printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-2/status.txt
91
+ else
92
+ printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-2/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-1_tp-8_pp-1_mbz-2 --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-1_tp-8_pp-1_mbz-2 --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-1_tp-8_pp-1_mbz-2 llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-2 --commit-message "Upload llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-2"
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-1_tp-8_pp-1_mbz-2/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: 1
51
+ pp_engine: 1f1b
52
+ tp: 8
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-1_tp-8_pp-1_mbz-2
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: 512
79
+ limit_test_batches: 0
80
+ limit_val_batches: 0
81
+ micro_batch_size: 2
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-1_tp-8_pp-1_mbz-2/log.out ADDED
@@ -0,0 +1,499 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ========================
2
+ START TIME: Wed Jul 3 22:29:06 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
+ W0703 22:29:15.909000 140073828640576 torch/distributed/run.py:757]
18
+ W0703 22:29:15.909000 140073828640576 torch/distributed/run.py:757] *****************************************
19
+ W0703 22:29:15.909000 140073828640576 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
+ W0703 22:29:15.909000 140073828640576 torch/distributed/run.py:757] *****************************************
21
+ [default0]:07/03/2024 22:29:38 [WARNING|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Vocab Size Padding] Padded vocab (size: 50257) with 7 dummy tokens (new size: 50264)
22
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Config:
23
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Config(general=GeneralArgs(project='bench_cluster',
24
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: run='%date_%jobid',
25
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: seed=42,
26
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: step=None,
27
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: consumed_train_samples=None,
28
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: benchmark_csv_path=None,
29
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: ignore_sanity_checks=True),
30
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: parallelism=ParallelismArgs(dp=1,
31
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pp=1,
32
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tp=8,
33
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f61d8e64940>,
34
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
35
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tp_linear_async_communication=False,
36
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: expert_parallel_size=1),
37
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
38
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: eos_token_id=2,
39
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hidden_act='silu',
40
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hidden_size=2048,
41
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: initializer_range=0.02,
42
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: intermediate_size=4096,
43
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: is_llama_config=True,
44
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: max_position_embeddings=4096,
45
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_attention_heads=32,
46
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_hidden_layers=24,
47
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_key_value_heads=32,
48
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pad_token_id=None,
49
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pretraining_tp=1,
50
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rms_norm_eps=1e-05,
51
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rope_scaling=None,
52
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rope_theta=10000.0,
53
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tie_word_embeddings=True,
54
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: use_cache=True,
55
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: vocab_size=50264),
56
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: init_method=RandomInit(std=0.025),
57
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: dtype=torch.bfloat16,
58
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: make_vocab_size_divisible_by=1,
59
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: ddp_bucket_cap_mb=25),
60
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
61
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tokenizer_revision=None,
62
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tokenizer_max_length=None),
63
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
64
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: checkpoint_interval=100000,
65
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: save_initial_state=False,
66
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: resume_checkpoint_path=None,
67
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: checkpoints_path_is_shared_file_system=False),
68
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: logging=LoggingArgs(log_level='info',
69
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: log_level_replica='info',
70
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: iteration_step_info_interval=1),
71
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tokens=TokensArgs(sequence_length=4096,
72
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: train_steps=20,
73
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: micro_batch_size=2,
74
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: batch_accumulation_per_replica=512,
75
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: val_check_interval=-1,
76
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: limit_val_batches=0,
77
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: limit_test_batches=0),
78
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
79
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: adam_beta1=0.9,
80
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: adam_beta2=0.95,
81
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: torch_adam_is_fused=True,
82
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: name='adamW'),
83
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: zero_stage=1,
84
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: weight_decay=0.01,
85
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: clip_grad=1.0,
86
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: accumulate_grad_in_fp32=True,
87
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
88
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lr_warmup_steps=1,
89
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lr_warmup_style='linear',
90
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lr_decay_style='linear',
91
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lr_decay_steps=19,
92
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lr_decay_starting_step=None,
93
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: min_decay_lr=1e-05)),
94
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: data_stages=[DatasetStageArgs(name='Training Stage',
95
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: start_training_step=1,
96
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
97
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hf_dataset_splits='train',
98
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hf_dataset_config_name=None,
99
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: dataset_processing_num_proc_per_process=64,
100
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: dataset_overwrite_cache=False,
101
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: text_column_name='text'),
102
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: seed=42,
103
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_loading_workers=0))],
104
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-2')),
105
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lighteval=None)
106
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Model Config:
107
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: LlamaConfig(bos_token_id=1,
108
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: eos_token_id=2,
109
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hidden_act='silu',
110
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hidden_size=2048,
111
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: initializer_range=0.02,
112
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: intermediate_size=4096,
113
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: is_llama_config=True,
114
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: max_position_embeddings=4096,
115
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_attention_heads=32,
116
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_hidden_layers=24,
117
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_key_value_heads=32,
118
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pad_token_id=None,
119
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pretraining_tp=1,
120
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rms_norm_eps=1e-05,
121
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rope_scaling=None,
122
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rope_theta=10000.0,
123
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tie_word_embeddings=True,
124
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: use_cache=True,
125
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: vocab_size=50264)
126
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Building model..
127
+ [default0]:07/03/2024 22:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Setting PP block ranks...
128
+ [default5]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=5|ip-26-0-160-225]: Local number of parameters: 139M (264.73MiB)
129
+ [default5]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=5|ip-26-0-160-225]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
130
+ [default6]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=6|ip-26-0-160-225]: Local number of parameters: 139M (264.73MiB)
131
+ [default6]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=6|ip-26-0-160-225]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
132
+ [default5]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=5|ip-26-0-160-225]: No checkpoint path provided.
133
+ [default6]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=6|ip-26-0-160-225]: No checkpoint path provided.
134
+ [default3]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=3|ip-26-0-160-225]: Local number of parameters: 139M (264.73MiB)
135
+ [default3]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=3|ip-26-0-160-225]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
136
+ [default3]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=3|ip-26-0-160-225]: No checkpoint path provided.
137
+ [default7]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=7|ip-26-0-160-225]: Local number of parameters: 139M (264.73MiB)
138
+ [default7]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=7|ip-26-0-160-225]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
139
+ [default7]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=7|ip-26-0-160-225]: No checkpoint path provided.
140
+ [default1]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=1|ip-26-0-160-225]: Local number of parameters: 139M (264.73MiB)
141
+ [default1]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=1|ip-26-0-160-225]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
142
+ [default1]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=1|ip-26-0-160-225]: No checkpoint path provided.
143
+ [default0]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Total number of parameters: 1.11G (2117.88MiB)
144
+ [default0]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Local number of parameters: 139M (264.73MiB)
145
+ [default0]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
146
+ [default0]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: No checkpoint path provided.
147
+ [default0]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Parametrizing model parameters using StandardParametrizator
148
+ [default0]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Optimizer Building] Using LearningRateForSP as learning rate
149
+ [default0]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [ZeRO sharding] Size of optimizer params per rank:
150
+ [default0]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [ZeRO sharding] DP Rank 0 has 139M out of 139M (100.00%) params' optimizer states
151
+ [default4]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=4|ip-26-0-160-225]: Local number of parameters: 139M (264.73MiB)
152
+ [default4]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=4|ip-26-0-160-225]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
153
+ [default4]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=4|ip-26-0-160-225]: No checkpoint path provided.
154
+ [default2]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=2|ip-26-0-160-225]: Local number of parameters: 139M (264.73MiB)
155
+ [default2]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=2|ip-26-0-160-225]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
156
+ [default2]:07/03/2024 22:29:54 [INFO|DP=0|PP=0|TP=2|ip-26-0-160-225]: No checkpoint path provided.
157
+ [default0]:07/03/2024 22:29:55 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
158
+ [default0]:07/03/2024 22:29:55 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Using `datasets` library
159
+ [default0]:07/03/2024 22:29:55 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
160
+ [default0]:07/03/2024 22:29:56 [WARNING|DP=0|PP=0|TP=0|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
161
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
162
+ [default0]:07/03/2024 22:29:58 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Training Plan] There are 1 training stages
163
+ [default0]:07/03/2024 22:29:58 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Stage Training Stage] start from step 1
164
+ [default0]:07/03/2024 22:29:58 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]:
165
+ [default0]:07/03/2024 22:29:58 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Start training] datetime: 2024-07-03 22:29:58.364589 | mbs: 2 | grad_accum: 512 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
166
+ [default0]:07/03/2024 22:29:58 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
167
+ [default0]:07/03/2024 22:29:58 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 1350.75MiB. Peak allocated 1350.76MiB. Peak reserved: 1384.00MiB
168
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
169
+ [default5]:07/03/2024 22:29:58 [WARNING|DP=0|PP=0|TP=5|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
170
+ [default7]:07/03/2024 22:29:58 [WARNING|DP=0|PP=0|TP=7|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
171
+ [default3]:07/03/2024 22:29:58 [WARNING|DP=0|PP=0|TP=3|ip-26-0-160-225]: 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
+ [default1]:07/03/2024 22:29:58 [WARNING|DP=0|PP=0|TP=1|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
174
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
175
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
176
+ [default2]:07/03/2024 22:29:58 [WARNING|DP=0|PP=0|TP=2|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
177
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
178
+ [default6]:07/03/2024 22:29:58 [WARNING|DP=0|PP=0|TP=6|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
179
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
180
+ [default4]:07/03/2024 22:29:58 [WARNING|DP=0|PP=0|TP=4|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
181
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
182
+ [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.)
183
+ [default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
184
+ [default1]:/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.)
185
+ [default1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
186
+ [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.)
187
+ [default0]:/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.)
188
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
189
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
190
+ [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.)
191
+ [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.)
192
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
193
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
194
+ [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.)
195
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
196
+ [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.)
197
+ [default7]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
198
+ [default0]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py:2261: UserWarning: torch.distributed.all_reduce_coalesced will be deprecated. If you must use it, please revisit our documentation later at https://pytorch.org/docs/master/distributed.html#collective-functions
199
+ [default0]: warnings.warn(
200
+ [default1]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py:2261: UserWarning: torch.distributed.all_reduce_coalesced will be deprecated. If you must use it, please revisit our documentation later at https://pytorch.org/docs/master/distributed.html#collective-functions
201
+ [default1]: warnings.warn(
202
+ [default3]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py:2261: UserWarning: torch.distributed.all_reduce_coalesced will be deprecated. If you must use it, please revisit our documentation later at https://pytorch.org/docs/master/distributed.html#collective-functions
203
+ [default3]: warnings.warn(
204
+ [default5]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py:2261: UserWarning: torch.distributed.all_reduce_coalesced will be deprecated. If you must use it, please revisit our documentation later at https://pytorch.org/docs/master/distributed.html#collective-functions
205
+ [default5]: warnings.warn(
206
+ [default0]:07/03/2024 22:31:17 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 1427.15MiB. Peak allocated 5028.02MiB. Peak reserved: 5396.00MiB
207
+ [default2]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py:2261: UserWarning: torch.distributed.all_reduce_coalesced will be deprecated. If you must use it, please revisit our documentation later at https://pytorch.org/docs/master/distributed.html#collective-functions
208
+ [default2]: warnings.warn(
209
+ [default6]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py:2261: UserWarning: torch.distributed.all_reduce_coalesced will be deprecated. If you must use it, please revisit our documentation later at https://pytorch.org/docs/master/distributed.html#collective-functions
210
+ [default6]: warnings.warn(
211
+ [default7]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py:2261: UserWarning: torch.distributed.all_reduce_coalesced will be deprecated. If you must use it, please revisit our documentation later at https://pytorch.org/docs/master/distributed.html#collective-functions
212
+ [default7]: warnings.warn(
213
+ [default4]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py:2261: UserWarning: torch.distributed.all_reduce_coalesced will be deprecated. If you must use it, please revisit our documentation later at https://pytorch.org/docs/master/distributed.html#collective-functions
214
+ [default4]: warnings.warn(
215
+ [default0]:07/03/2024 22:31:17 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: iteration: 1 / 20 | consumed_tokens: 4.19M | elapsed_time_per_iteration_ms: 79.2K | tokens_per_sec: 53K | tokens_per_sec_per_gpu: 6.62K | global_batch_size: 1.02K | lm_loss: 11.5 | lr: 0.0001 | model_tflops_per_gpu: 60.1 | hardware_tflops_per_gpu: 60.1 | grad_norm: 15.7 | cuda_memory_allocated: 2.61G | cuda_max_memory_reserved: 5.66G | hd_total_memory_tb: 312G | hd_used_memory_tb: 66.5G | hd_free_memory_tb: 246G
216
+ [default0]:07/03/2024 22:31:17 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 2486.16MiB. Peak allocated 2486.16MiB. Peak reserved: 5396.00MiB
217
+ [default0]:07/03/2024 22:32:28 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 2486.41MiB. Peak allocated 6087.28MiB. Peak reserved: 6248.00MiB
218
+ [default0]:07/03/2024 22:32:28 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: iteration: 2 / 20 | consumed_tokens: 8.39M | elapsed_time_per_iteration_ms: 71K | tokens_per_sec: 59.1K | tokens_per_sec_per_gpu: 7.38K | global_batch_size: 1.02K | lm_loss: 11.5 | lr: 9.53e-05 | model_tflops_per_gpu: 67 | hardware_tflops_per_gpu: 67 | grad_norm: 16 | cuda_memory_allocated: 2.61G | cuda_max_memory_reserved: 6.55G | hd_total_memory_tb: 312G | hd_used_memory_tb: 66.5G | hd_free_memory_tb: 246G
219
+ [default0]:07/03/2024 22:32:28 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 2486.16MiB. Peak allocated 2486.48MiB. Peak reserved: 6248.00MiB
220
+ [default0]:07/03/2024 22:33:48 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 2486.41MiB. Peak allocated 6087.28MiB. Peak reserved: 6248.00MiB
221
+ [default0]:07/03/2024 22:33:48 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: iteration: 3 / 20 | consumed_tokens: 12.6M | elapsed_time_per_iteration_ms: 80.3K | tokens_per_sec: 52.2K | tokens_per_sec_per_gpu: 6.53K | global_batch_size: 1.02K | lm_loss: 12.8 | lr: 9.05e-05 | model_tflops_per_gpu: 59.2 | hardware_tflops_per_gpu: 59.2 | grad_norm: 137 | cuda_memory_allocated: 2.61G | cuda_max_memory_reserved: 6.55G | hd_total_memory_tb: 312G | hd_used_memory_tb: 66.5G | hd_free_memory_tb: 246G
222
+ [default0]:07/03/2024 22:33:48 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 2486.16MiB. Peak allocated 2486.48MiB. Peak reserved: 6248.00MiB
223
+ [default0]:STAGE:2024-07-03 22:33:48 343298:343298 ActivityProfilerController.cpp:314] Completed Stage: Warm Up
224
+ [default0]:07/03/2024 22:35:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 2486.41MiB. Peak allocated 6087.28MiB. Peak reserved: 6248.00MiB
225
+ [default0]:07/03/2024 22:35:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: iteration: 4 / 20 | consumed_tokens: 16.8M | elapsed_time_per_iteration_ms: 107K | tokens_per_sec: 39.2K | tokens_per_sec_per_gpu: 4.9K | global_batch_size: 1.02K | lm_loss: 12.2 | lr: 8.58e-05 | model_tflops_per_gpu: 44.5 | hardware_tflops_per_gpu: 44.5 | grad_norm: 22.4 | cuda_memory_allocated: 2.61G | cuda_max_memory_reserved: 6.55G | hd_total_memory_tb: 312G | hd_used_memory_tb: 66.5G | hd_free_memory_tb: 246G
226
+ [default0]:07/03/2024 22:35:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 2486.16MiB. Peak allocated 2486.48MiB. Peak reserved: 6248.00MiB
227
+ [default0]:07/03/2024 22:37:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: iteration: 5 / 20 | consumed_tokens: 21M | elapsed_time_per_iteration_ms: 107K | tokens_per_sec: 39.2K | tokens_per_sec_per_gpu: 4.9K | global_batch_size: 1.02K | lm_loss: 12.4 | lr: 8.11e-05 | model_tflops_per_gpu: 44.5 | hardware_tflops_per_gpu: 44.5 | grad_norm: 42.8
228
+ [default0]:07/03/2024 22:37:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 2486.16MiB. Peak allocated 6087.28MiB. Peak reserved: 6248.00MiB
229
+ [default0]:07/03/2024 22:39:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: iteration: 6 / 20 | consumed_tokens: 25.2M | elapsed_time_per_iteration_ms: 108K | tokens_per_sec: 39K | tokens_per_sec_per_gpu: 4.87K | global_batch_size: 1.02K | lm_loss: 11.1 | lr: 7.63e-05 | model_tflops_per_gpu: 44.2 | hardware_tflops_per_gpu: 44.2 | grad_norm: 24.8
230
+ [default0]:STAGE:2024-07-03 22:44:09 343298:343298 ActivityProfilerController.cpp:320] Completed Stage: Collection
231
+ [default0]:STAGE:2024-07-03 22:44:39 343298:343298 ActivityProfilerController.cpp:324] Completed Stage: Post Processing
232
+ [default1]:[rank1]:[E ProcessGroupNCCL.cpp:563] [Rank 1] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600054 milliseconds before timing out.
233
+ [default5]:[rank5]:[E ProcessGroupNCCL.cpp:563] [Rank 5] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600056 milliseconds before timing out.
234
+ [default4]:[rank4]:[E ProcessGroupNCCL.cpp:563] [Rank 4] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600041 milliseconds before timing out.
235
+ [default3]:[rank3]:[E ProcessGroupNCCL.cpp:563] [Rank 3] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600005 milliseconds before timing out.
236
+ [default7]:[rank7]:[E ProcessGroupNCCL.cpp:563] [Rank 7] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600024 milliseconds before timing out.
237
+ [default2]:[rank2]:[E ProcessGroupNCCL.cpp:563] [Rank 2] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600030 milliseconds before timing out.
238
+ [default6]:[rank6]:[E ProcessGroupNCCL.cpp:563] [Rank 6] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600046 milliseconds before timing out.
239
+ [default7]:[rank7]:[E ProcessGroupNCCL.cpp:1537] [PG 2 Rank 7] Timeout at NCCL work: 611392, last enqueued NCCL work: 611508, last completed NCCL work: 611391.
240
+ [default7]:[rank7]:[E ProcessGroupNCCL.cpp:577] [Rank 7] Some NCCL operations have failed or timed out. Due to the asynchronous nature of CUDA kernels, subsequent GPU operations might run on corrupted/incomplete data.
241
+ [default7]:[rank7]:[E ProcessGroupNCCL.cpp:583] [Rank 7] To avoid data inconsistency, we are taking the entire process down.
242
+ [default7]:[rank7]:[E ProcessGroupNCCL.cpp:1414] [PG 2 Rank 7] Process group watchdog thread terminated with exception: [Rank 7] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600024 milliseconds before timing out.
243
+ [default7]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
244
+ [default7]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7f902227c897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
245
+ [default7]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7f9023555c62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
246
+ [default7]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7f902355aa80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
247
+ [default7]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7f902355bdcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
248
+ [default7]:frame #4: <unknown function> + 0xd3e95 (0x7f906eff4e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
249
+ [default7]:frame #5: <unknown function> + 0x8609 (0x7f907403b609 in /lib/x86_64-linux-gnu/libpthread.so.0)
250
+ [default7]:frame #6: clone + 0x43 (0x7f9073e06353 in /lib/x86_64-linux-gnu/libc.so.6)
251
+ [default7]:
252
+ [default7]:terminate called after throwing an instance of 'c10::DistBackendError'
253
+ [default7]: what(): [PG 2 Rank 7] Process group watchdog thread terminated with exception: [Rank 7] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600024 milliseconds before timing out.
254
+ [default7]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
255
+ [default7]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7f902227c897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
256
+ [default7]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7f9023555c62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
257
+ [default7]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7f902355aa80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
258
+ [default7]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7f902355bdcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
259
+ [default7]:frame #4: <unknown function> + 0xd3e95 (0x7f906eff4e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
260
+ [default7]:frame #5: <unknown function> + 0x8609 (0x7f907403b609 in /lib/x86_64-linux-gnu/libpthread.so.0)
261
+ [default7]:frame #6: clone + 0x43 (0x7f9073e06353 in /lib/x86_64-linux-gnu/libc.so.6)
262
+ [default7]:
263
+ [default7]:Exception raised from ncclCommWatchdog at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:1418 (most recent call first):
264
+ [default7]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7f902227c897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
265
+ [default7]:frame #1: <unknown function> + 0xe32119 (0x7f90231df119 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
266
+ [default7]:frame #2: <unknown function> + 0xd3e95 (0x7f906eff4e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
267
+ [default7]:frame #3: <unknown function> + 0x8609 (0x7f907403b609 in /lib/x86_64-linux-gnu/libpthread.so.0)
268
+ [default7]:frame #4: clone + 0x43 (0x7f9073e06353 in /lib/x86_64-linux-gnu/libc.so.6)
269
+ [default7]:
270
+ [default2]:[rank2]:[E ProcessGroupNCCL.cpp:1537] [PG 2 Rank 2] Timeout at NCCL work: 611392, last enqueued NCCL work: 611508, last completed NCCL work: 611391.
271
+ [default2]:[rank2]:[E ProcessGroupNCCL.cpp:577] [Rank 2] Some NCCL operations have failed or timed out. Due to the asynchronous nature of CUDA kernels, subsequent GPU operations might run on corrupted/incomplete data.
272
+ [default2]:[rank2]:[E ProcessGroupNCCL.cpp:583] [Rank 2] To avoid data inconsistency, we are taking the entire process down.
273
+ [default2]:[rank2]:[E ProcessGroupNCCL.cpp:1414] [PG 2 Rank 2] Process group watchdog thread terminated with exception: [Rank 2] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600030 milliseconds before timing out.
274
+ [default2]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
275
+ [default2]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7fcf8bdea897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
276
+ [default2]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7fcf8d0c3c62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
277
+ [default2]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7fcf8d0c8a80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
278
+ [default2]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7fcf8d0c9dcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
279
+ [default2]:frame #4: <unknown function> + 0xd3e95 (0x7fcfd8b62e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
280
+ [default2]:frame #5: <unknown function> + 0x8609 (0x7fcfddba9609 in /lib/x86_64-linux-gnu/libpthread.so.0)
281
+ [default2]:frame #6: clone + 0x43 (0x7fcfdd974353 in /lib/x86_64-linux-gnu/libc.so.6)
282
+ [default2]:
283
+ [default2]:terminate called after throwing an instance of 'c10::DistBackendError'
284
+ [default2]: what(): [PG 2 Rank 2] Process group watchdog thread terminated with exception: [Rank 2] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600030 milliseconds before timing out.
285
+ [default2]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
286
+ [default2]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7fcf8bdea897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
287
+ [default2]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7fcf8d0c3c62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
288
+ [default2]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7fcf8d0c8a80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
289
+ [default2]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7fcf8d0c9dcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
290
+ [default2]:frame #4: <unknown function> + 0xd3e95 (0x7fcfd8b62e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
291
+ [default2]:frame #5: <unknown function> + 0x8609 (0x7fcfddba9609 in /lib/x86_64-linux-gnu/libpthread.so.0)
292
+ [default2]:frame #6: clone + 0x43 (0x7fcfdd974353 in /lib/x86_64-linux-gnu/libc.so.6)
293
+ [default2]:
294
+ [default2]:Exception raised from ncclCommWatchdog at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:1418 (most recent call first):
295
+ [default2]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7fcf8bdea897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
296
+ [default2]:frame #1: <unknown function> + 0xe32119 (0x7fcf8cd4d119 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
297
+ [default2]:frame #2: <unknown function> + 0xd3e95 (0x7fcfd8b62e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
298
+ [default2]:frame #3: <unknown function> + 0x8609 (0x7fcfddba9609 in /lib/x86_64-linux-gnu/libpthread.so.0)
299
+ [default2]:frame #4: clone + 0x43 (0x7fcfdd974353 in /lib/x86_64-linux-gnu/libc.so.6)
300
+ [default2]:
301
+ [default6]:[rank6]:[E ProcessGroupNCCL.cpp:1537] [PG 2 Rank 6] Timeout at NCCL work: 611392, last enqueued NCCL work: 611508, last completed NCCL work: 611391.
302
+ [default6]:[rank6]:[E ProcessGroupNCCL.cpp:577] [Rank 6] Some NCCL operations have failed or timed out. Due to the asynchronous nature of CUDA kernels, subsequent GPU operations might run on corrupted/incomplete data.
303
+ [default6]:[rank6]:[E ProcessGroupNCCL.cpp:583] [Rank 6] To avoid data inconsistency, we are taking the entire process down.
304
+ [default6]:[rank6]:[E ProcessGroupNCCL.cpp:1414] [PG 2 Rank 6] Process group watchdog thread terminated with exception: [Rank 6] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600046 milliseconds before timing out.
305
+ [default6]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
306
+ [default6]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7f1c1eae9897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
307
+ [default6]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7f1c1fdc2c62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
308
+ [default6]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7f1c1fdc7a80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
309
+ [default6]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7f1c1fdc8dcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
310
+ [default6]:frame #4: <unknown function> + 0xd3e95 (0x7f1c6b861e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
311
+ [default6]:frame #5: <unknown function> + 0x8609 (0x7f1c708a8609 in /lib/x86_64-linux-gnu/libpthread.so.0)
312
+ [default6]:frame #6: clone + 0x43 (0x7f1c70673353 in /lib/x86_64-linux-gnu/libc.so.6)
313
+ [default6]:
314
+ [default6]:terminate called after throwing an instance of 'c10::DistBackendError'
315
+ [default6]: what(): [PG 2 Rank 6] Process group watchdog thread terminated with exception: [Rank 6] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600046 milliseconds before timing out.
316
+ [default6]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
317
+ [default6]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7f1c1eae9897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
318
+ [default6]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7f1c1fdc2c62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
319
+ [default6]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7f1c1fdc7a80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
320
+ [default6]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7f1c1fdc8dcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
321
+ [default6]:frame #4: <unknown function> + 0xd3e95 (0x7f1c6b861e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
322
+ [default6]:frame #5: <unknown function> + 0x8609 (0x7f1c708a8609 in /lib/x86_64-linux-gnu/libpthread.so.0)
323
+ [default6]:frame #6: clone + 0x43 (0x7f1c70673353 in /lib/x86_64-linux-gnu/libc.so.6)
324
+ [default6]:
325
+ [default6]:Exception raised from ncclCommWatchdog at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:1418 (most recent call first):
326
+ [default6]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7f1c1eae9897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
327
+ [default6]:frame #1: <unknown function> + 0xe32119 (0x7f1c1fa4c119 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
328
+ [default6]:frame #2: <unknown function> + 0xd3e95 (0x7f1c6b861e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
329
+ [default6]:frame #3: <unknown function> + 0x8609 (0x7f1c708a8609 in /lib/x86_64-linux-gnu/libpthread.so.0)
330
+ [default6]:frame #4: clone + 0x43 (0x7f1c70673353 in /lib/x86_64-linux-gnu/libc.so.6)
331
+ [default6]:
332
+ [default5]:[rank5]:[E ProcessGroupNCCL.cpp:1537] [PG 2 Rank 5] Timeout at NCCL work: 611392, last enqueued NCCL work: 611508, last completed NCCL work: 611391.
333
+ [default5]:[rank5]:[E ProcessGroupNCCL.cpp:577] [Rank 5] Some NCCL operations have failed or timed out. Due to the asynchronous nature of CUDA kernels, subsequent GPU operations might run on corrupted/incomplete data.
334
+ [default1]:[rank1]:[E ProcessGroupNCCL.cpp:1537] [PG 2 Rank 1] Timeout at NCCL work: 611392, last enqueued NCCL work: 611508, last completed NCCL work: 611391.
335
+ [default1]:[rank1]:[E ProcessGroupNCCL.cpp:577] [Rank 1] Some NCCL operations have failed or timed out. Due to the asynchronous nature of CUDA kernels, subsequent GPU operations might run on corrupted/incomplete data.
336
+ [default1]:[rank1]:[E ProcessGroupNCCL.cpp:583] [Rank 1] To avoid data inconsistency, we are taking the entire process down.
337
+ [default5]:[rank5]:[E ProcessGroupNCCL.cpp:583] [Rank 5] To avoid data inconsistency, we are taking the entire process down.
338
+ [default5]:[rank5]:[E ProcessGroupNCCL.cpp:1414] [PG 2 Rank 5] Process group watchdog thread terminated with exception: [Rank 5] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600056 milliseconds before timing out.
339
+ [default5]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
340
+ [default5]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7fef091b4897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
341
+ [default1]:[rank1]:[E ProcessGroupNCCL.cpp:1414] [PG 2 Rank 1] Process group watchdog thread terminated with exception: [Rank 1] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600054 milliseconds before timing out.
342
+ [default1]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
343
+ [default1]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7fc0c8b0c897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
344
+ [default1]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7fc0c9de5c62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
345
+ [default1]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7fc0c9deaa80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
346
+ [default1]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7fc0c9debdcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
347
+ [default1]:frame #4: <unknown function> + 0xd3e95 (0x7fc115884e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
348
+ [default1]:frame #5: <unknown function> + 0x8609 (0x7fc11a8cb609 in /lib/x86_64-linux-gnu/libpthread.so.0)
349
+ [default5]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7fef0a48dc62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
350
+ [default1]:frame #6: clone + 0x43 (0x7fc11a696353 in /lib/x86_64-linux-gnu/libc.so.6)
351
+ [default5]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7fef0a492a80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
352
+ [default1]:
353
+ [default1]:terminate called after throwing an instance of 'c10::DistBackendError'
354
+ [default1]: what(): [PG 2 Rank 1] Process group watchdog thread terminated with exception: [Rank 1] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600054 milliseconds before timing out.
355
+ [default1]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
356
+ [default1]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7fc0c8b0c897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
357
+ [default1]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7fc0c9de5c62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
358
+ [default1]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7fc0c9deaa80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
359
+ [default1]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7fc0c9debdcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
360
+ [default1]:frame #4: <unknown function> + 0xd3e95 (0x7fc115884e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
361
+ [default1]:frame #5: <unknown function> + 0x8609 (0x7fc11a8cb609 in /lib/x86_64-linux-gnu/libpthread.so.0)
362
+ [default5]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7fef0a493dcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
363
+ [default1]:frame #6: clone + 0x43 (0x7fc11a696353 in /lib/x86_64-linux-gnu/libc.so.6)
364
+ [default1]:
365
+ [default5]:frame #4: <unknown function> + 0xd3e95 (0x7fef55f2ce95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
366
+ [default5]:frame #5: <unknown function> + 0x8609 (0x7fef5af73609 in /lib/x86_64-linux-gnu/libpthread.so.0)
367
+ [default5]:frame #6: clone + 0x43 (0x7fef5ad3e353 in /lib/x86_64-linux-gnu/libc.so.6)
368
+ [default1]:Exception raised from ncclCommWatchdog at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:1418 (most recent call first):
369
+ [default1]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7fc0c8b0c897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
370
+ [default1]:frame #1: <unknown function> + 0xe32119 (0x7fc0c9a6f119 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
371
+ [default1]:frame #2: <unknown function> + 0xd3e95 (0x7fc115884e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
372
+ [default1]:frame #3: <unknown function> + 0x8609 (0x7fc11a8cb609 in /lib/x86_64-linux-gnu/libpthread.so.0)
373
+ [default1]:frame #4: clone + 0x43 (0x7fc11a696353 in /lib/x86_64-linux-gnu/libc.so.6)
374
+ [default5]:
375
+ [default1]:
376
+ [default5]:terminate called after throwing an instance of 'c10::DistBackendError'
377
+ [default5]: what(): [PG 2 Rank 5] Process group watchdog thread terminated with exception: [Rank 5] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600056 milliseconds before timing out.
378
+ [default5]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
379
+ [default5]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7fef091b4897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
380
+ [default5]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7fef0a48dc62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
381
+ [default5]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7fef0a492a80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
382
+ [default5]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7fef0a493dcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
383
+ [default5]:frame #4: <unknown function> + 0xd3e95 (0x7fef55f2ce95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
384
+ [default5]:frame #5: <unknown function> + 0x8609 (0x7fef5af73609 in /lib/x86_64-linux-gnu/libpthread.so.0)
385
+ [default5]:frame #6: clone + 0x43 (0x7fef5ad3e353 in /lib/x86_64-linux-gnu/libc.so.6)
386
+ [default5]:
387
+ [default5]:Exception raised from ncclCommWatchdog at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:1418 (most recent call first):
388
+ [default5]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7fef091b4897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
389
+ [default5]:frame #1: <unknown function> + 0xe32119 (0x7fef0a117119 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
390
+ [default5]:frame #2: <unknown function> + 0xd3e95 (0x7fef55f2ce95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
391
+ [default5]:frame #3: <unknown function> + 0x8609 (0x7fef5af73609 in /lib/x86_64-linux-gnu/libpthread.so.0)
392
+ [default5]:frame #4: clone + 0x43 (0x7fef5ad3e353 in /lib/x86_64-linux-gnu/libc.so.6)
393
+ [default5]:
394
+ [default4]:[rank4]:[E ProcessGroupNCCL.cpp:1537] [PG 2 Rank 4] Timeout at NCCL work: 611392, last enqueued NCCL work: 611508, last completed NCCL work: 611391.
395
+ [default4]:[rank4]:[E ProcessGroupNCCL.cpp:577] [Rank 4] Some NCCL operations have failed or timed out. Due to the asynchronous nature of CUDA kernels, subsequent GPU operations might run on corrupted/incomplete data.
396
+ [default4]:[rank4]:[E ProcessGroupNCCL.cpp:583] [Rank 4] To avoid data inconsistency, we are taking the entire process down.
397
+ [default4]:[rank4]:[E ProcessGroupNCCL.cpp:1414] [PG 2 Rank 4] Process group watchdog thread terminated with exception: [Rank 4] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600041 milliseconds before timing out.
398
+ [default4]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
399
+ [default4]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7f632db7b897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
400
+ [default4]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7f632ee54c62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
401
+ [default4]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7f632ee59a80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
402
+ [default4]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7f632ee5adcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
403
+ [default4]:frame #4: <unknown function> + 0xd3e95 (0x7f637a8f3e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
404
+ [default4]:frame #5: <unknown function> + 0x8609 (0x7f637f93a609 in /lib/x86_64-linux-gnu/libpthread.so.0)
405
+ [default4]:frame #6: clone + 0x43 (0x7f637f705353 in /lib/x86_64-linux-gnu/libc.so.6)
406
+ [default4]:
407
+ [default4]:terminate called after throwing an instance of 'c10::DistBackendError'
408
+ [default4]: what(): [PG 2 Rank 4] Process group watchdog thread terminated with exception: [Rank 4] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600041 milliseconds before timing out.
409
+ [default4]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
410
+ [default4]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7f632db7b897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
411
+ [default4]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7f632ee54c62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
412
+ [default4]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7f632ee59a80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
413
+ [default4]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7f632ee5adcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
414
+ [default4]:frame #4: <unknown function> + 0xd3e95 (0x7f637a8f3e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
415
+ [default4]:frame #5: <unknown function> + 0x8609 (0x7f637f93a609 in /lib/x86_64-linux-gnu/libpthread.so.0)
416
+ [default4]:frame #6: clone + 0x43 (0x7f637f705353 in /lib/x86_64-linux-gnu/libc.so.6)
417
+ [default4]:
418
+ [default4]:Exception raised from ncclCommWatchdog at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:1418 (most recent call first):
419
+ [default4]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7f632db7b897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
420
+ [default4]:frame #1: <unknown function> + 0xe32119 (0x7f632eade119 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
421
+ [default4]:frame #2: <unknown function> + 0xd3e95 (0x7f637a8f3e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
422
+ [default4]:frame #3: <unknown function> + 0x8609 (0x7f637f93a609 in /lib/x86_64-linux-gnu/libpthread.so.0)
423
+ [default4]:frame #4: clone + 0x43 (0x7f637f705353 in /lib/x86_64-linux-gnu/libc.so.6)
424
+ [default4]:
425
+ [default3]:[rank3]:[E ProcessGroupNCCL.cpp:1537] [PG 2 Rank 3] Timeout at NCCL work: 611392, last enqueued NCCL work: 611508, last completed NCCL work: 611391.
426
+ [default3]:[rank3]:[E ProcessGroupNCCL.cpp:577] [Rank 3] Some NCCL operations have failed or timed out. Due to the asynchronous nature of CUDA kernels, subsequent GPU operations might run on corrupted/incomplete data.
427
+ [default3]:[rank3]:[E ProcessGroupNCCL.cpp:583] [Rank 3] To avoid data inconsistency, we are taking the entire process down.
428
+ [default3]:[rank3]:[E ProcessGroupNCCL.cpp:1414] [PG 2 Rank 3] Process group watchdog thread terminated with exception: [Rank 3] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600005 milliseconds before timing out.
429
+ [default3]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
430
+ [default3]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7fa191dfa897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
431
+ [default3]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7fa1930d3c62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
432
+ [default3]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7fa1930d8a80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
433
+ [default3]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7fa1930d9dcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
434
+ [default3]:frame #4: <unknown function> + 0xd3e95 (0x7fa1deb72e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
435
+ [default3]:frame #5: <unknown function> + 0x8609 (0x7fa1e3bb9609 in /lib/x86_64-linux-gnu/libpthread.so.0)
436
+ [default3]:frame #6: clone + 0x43 (0x7fa1e3984353 in /lib/x86_64-linux-gnu/libc.so.6)
437
+ [default3]:
438
+ [default3]:terminate called after throwing an instance of 'c10::DistBackendError'
439
+ [default3]: what(): [PG 2 Rank 3] Process group watchdog thread terminated with exception: [Rank 3] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=611392, OpType=_REDUCE_SCATTER_BASE, NumelIn=16777216, NumelOut=2097152, Timeout(ms)=600000) ran for 600005 milliseconds before timing out.
440
+ [default3]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
441
+ [default3]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7fa191dfa897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
442
+ [default3]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7fa1930d3c62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
443
+ [default3]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7fa1930d8a80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
444
+ [default3]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7fa1930d9dcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
445
+ [default3]:frame #4: <unknown function> + 0xd3e95 (0x7fa1deb72e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
446
+ [default3]:frame #5: <unknown function> + 0x8609 (0x7fa1e3bb9609 in /lib/x86_64-linux-gnu/libpthread.so.0)
447
+ [default3]:frame #6: clone + 0x43 (0x7fa1e3984353 in /lib/x86_64-linux-gnu/libc.so.6)
448
+ [default3]:
449
+ [default3]:Exception raised from ncclCommWatchdog at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:1418 (most recent call first):
450
+ [default3]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7fa191dfa897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
451
+ [default3]:frame #1: <unknown function> + 0xe32119 (0x7fa192d5d119 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
452
+ [default3]:frame #2: <unknown function> + 0xd3e95 (0x7fa1deb72e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
453
+ [default3]:frame #3: <unknown function> + 0x8609 (0x7fa1e3bb9609 in /lib/x86_64-linux-gnu/libpthread.so.0)
454
+ [default3]:frame #4: clone + 0x43 (0x7fa1e3984353 in /lib/x86_64-linux-gnu/libc.so.6)
455
+ [default3]:
456
+ W0703 22:49:42.263000 140073828640576 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 343298 closing signal SIGTERM
457
+ W0703 22:49:42.264000 140073828640576 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 343300 closing signal SIGTERM
458
+ W0703 22:49:42.264000 140073828640576 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 343301 closing signal SIGTERM
459
+ W0703 22:49:42.264000 140073828640576 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 343302 closing signal SIGTERM
460
+ W0703 22:49:42.264000 140073828640576 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 343303 closing signal SIGTERM
461
+ W0703 22:49:42.264000 140073828640576 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 343304 closing signal SIGTERM
462
+ E0703 22:49:49.660000 140073828640576 torch/distributed/elastic/multiprocessing/api.py:826] failed (exitcode: -6) local_rank: 1 (pid: 343299) of binary: /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/python3.10
463
+ Traceback (most recent call last):
464
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/torchrun", line 8, in <module>
465
+ sys.exit(main())
466
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 347, in wrapper
467
+ return f(*args, **kwargs)
468
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 879, in main
469
+ run(args)
470
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 870, in run
471
+ elastic_launch(
472
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 132, in __call__
473
+ return launch_agent(self._config, self._entrypoint, list(args))
474
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 263, in launch_agent
475
+ raise ChildFailedError(
476
+ torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
477
+ ============================================================
478
+ /fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py FAILED
479
+ ------------------------------------------------------------
480
+ Failures:
481
+ [1]:
482
+ time : 2024-07-03_22:49:42
483
+ host : ip-26-0-160-225.ec2.internal
484
+ rank : 7 (local_rank: 7)
485
+ exitcode : -6 (pid: 343305)
486
+ error_file: <N/A>
487
+ traceback : Signal 6 (SIGABRT) received by PID 343305
488
+ ------------------------------------------------------------
489
+ Root Cause (first observed failure):
490
+ [0]:
491
+ time : 2024-07-03_22:49:42
492
+ host : ip-26-0-160-225.ec2.internal
493
+ rank : 1 (local_rank: 1)
494
+ exitcode : -6 (pid: 343299)
495
+ error_file: <N/A>
496
+ traceback : Signal 6 (SIGABRT) received by PID 343299
497
+ ============================================================
498
+ srun: error: ip-26-0-160-225: task 0: Exited with exit code 1
499
+ 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-1_tp-8_pp-1_mbz-2/status.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ timeout