Hugging Face's logo

nanotron
/
old_bench

Model card Files Community
3outeille HF staff commited on
Commit
6f8ac19
•
1 Parent(s): 02510ec

Upload llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1

Browse files
Files changed (6) hide show
  1. .gitattributes +1 -0
  2. llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/bench.slurm +111 -0
  3. llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/config.yaml +90 -0
  4. llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/log.out +339 -0
  5. llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/profiler/ip-26-0-169-86_2278697.1720051013128234436.pt.trace.json.tmp +3 -0
  6. llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/status.txt +1 -0
.gitattributes CHANGED
@@ -119,3 +119,4 @@ llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-32/profiler/ip-26-0-171-88_1097996.1720048094
119
  llama-1B/8_GPUS/dp-1_tp-4_pp-2_mbz-4/profiler/ip-26-0-169-86_2274198.1720050009949646955.pt.trace.json.tmp filter=lfs diff=lfs merge=lfs -text
120
  llama-1B/8_GPUS/dp-1_tp-4_pp-2_mbz-16/profiler/ip-26-0-174-36_488999.1720049954801989090.pt.trace.json filter=lfs diff=lfs merge=lfs -text
121
  llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-1/profiler/ip-26-0-160-225_351538.1720048932925630211.pt.trace.json filter=lfs diff=lfs merge=lfs -text
 
 
119
  llama-1B/8_GPUS/dp-1_tp-4_pp-2_mbz-4/profiler/ip-26-0-169-86_2274198.1720050009949646955.pt.trace.json.tmp filter=lfs diff=lfs merge=lfs -text
120
  llama-1B/8_GPUS/dp-1_tp-4_pp-2_mbz-16/profiler/ip-26-0-174-36_488999.1720049954801989090.pt.trace.json filter=lfs diff=lfs merge=lfs -text
121
  llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-1/profiler/ip-26-0-160-225_351538.1720048932925630211.pt.trace.json filter=lfs diff=lfs merge=lfs -text
122
+ llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/profiler/ip-26-0-169-86_2278697.1720051013128234436.pt.trace.json.tmp filter=lfs diff=lfs merge=lfs -text
llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/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-4_tp-1_pp-2_mbz-1/log.out
13
+ #SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/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-4_tp-1_pp-2_mbz-1/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-4_tp-1_pp-2_mbz-1/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-4_tp-1_pp-2_mbz-1/status.txt
84
+ else
85
+ if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/log.out; then
86
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/status.txt
87
+ elif grep -q " CUDA error: an illegal memory access" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/log.out; then
88
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/status.txt
89
+ elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/log.out; then
90
+ printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/status.txt
91
+ else
92
+ printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/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-4_tp-1_pp-2_mbz-1 --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-4_tp-1_pp-2_mbz-1 --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-4_tp-1_pp-2_mbz-1 llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1 --commit-message "Upload llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1"
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-4_tp-1_pp-2_mbz-1/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: 4
49
+ expert_parallel_size: 1
50
+ pp: 2
51
+ pp_engine: 1f1b
52
+ tp: 1
53
+ tp_linear_async_communication: false
54
+ tp_mode: REDUCE_SCATTER
55
+ profiler:
56
+ profiler_export_path: /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1
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: 256
79
+ limit_test_batches: 0
80
+ limit_val_batches: 0
81
+ micro_batch_size: 1
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-4_tp-1_pp-2_mbz-1/log.out ADDED
@@ -0,0 +1,339 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ========================
2
+ START TIME: Wed Jul 3 23:43:53 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 23:43:55.830000 140140090980160 torch/distributed/run.py:757]
18
+ W0703 23:43:55.830000 140140090980160 torch/distributed/run.py:757] *****************************************
19
+ W0703 23:43:55.830000 140140090980160 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 23:43:55.830000 140140090980160 torch/distributed/run.py:757] *****************************************
21
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Config:
22
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Config(general=GeneralArgs(project='bench_cluster',
23
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: run='%date_%jobid',
24
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: seed=42,
25
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: step=None,
26
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: consumed_train_samples=None,
27
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: benchmark_csv_path=None,
28
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: ignore_sanity_checks=True),
29
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: parallelism=ParallelismArgs(dp=4,
30
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: pp=2,
31
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: tp=1,
32
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f9805e348b0>,
33
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
34
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: tp_linear_async_communication=False,
35
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: expert_parallel_size=1),
36
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
37
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: eos_token_id=2,
38
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: hidden_act='silu',
39
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: hidden_size=2048,
40
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: initializer_range=0.02,
41
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: intermediate_size=4096,
42
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: is_llama_config=True,
43
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: max_position_embeddings=4096,
44
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: num_attention_heads=32,
45
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: num_hidden_layers=24,
46
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: num_key_value_heads=32,
47
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: pad_token_id=None,
48
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: pretraining_tp=1,
49
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: rms_norm_eps=1e-05,
50
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: rope_scaling=None,
51
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: rope_theta=10000.0,
52
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: tie_word_embeddings=True,
53
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: use_cache=True,
54
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: vocab_size=50257),
55
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: init_method=RandomInit(std=0.025),
56
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: dtype=torch.bfloat16,
57
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: make_vocab_size_divisible_by=1,
58
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: ddp_bucket_cap_mb=25),
59
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
60
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: tokenizer_revision=None,
61
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: tokenizer_max_length=None),
62
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
63
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: checkpoint_interval=100000,
64
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: save_initial_state=False,
65
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: resume_checkpoint_path=None,
66
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: checkpoints_path_is_shared_file_system=False),
67
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: logging=LoggingArgs(log_level='info',
68
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: log_level_replica='info',
69
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: iteration_step_info_interval=1),
70
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: tokens=TokensArgs(sequence_length=4096,
71
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: train_steps=20,
72
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: micro_batch_size=1,
73
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: batch_accumulation_per_replica=256,
74
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: val_check_interval=-1,
75
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: limit_val_batches=0,
76
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: limit_test_batches=0),
77
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
78
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: adam_beta1=0.9,
79
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: adam_beta2=0.95,
80
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: torch_adam_is_fused=True,
81
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: name='adamW'),
82
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: zero_stage=1,
83
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: weight_decay=0.01,
84
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: clip_grad=1.0,
85
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: accumulate_grad_in_fp32=True,
86
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
87
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: lr_warmup_steps=1,
88
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: lr_warmup_style='linear',
89
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: lr_decay_style='linear',
90
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: lr_decay_steps=19,
91
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: lr_decay_starting_step=None,
92
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: min_decay_lr=1e-05)),
93
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: data_stages=[DatasetStageArgs(name='Training Stage',
94
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: start_training_step=1,
95
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
96
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: hf_dataset_splits='train',
97
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: hf_dataset_config_name=None,
98
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: dataset_processing_num_proc_per_process=64,
99
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: dataset_overwrite_cache=False,
100
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: text_column_name='text'),
101
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: seed=42,
102
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: num_loading_workers=0))],
103
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1')),
104
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: lighteval=None)
105
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Model Config:
106
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: LlamaConfig(bos_token_id=1,
107
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: eos_token_id=2,
108
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: hidden_act='silu',
109
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: hidden_size=2048,
110
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: initializer_range=0.02,
111
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: intermediate_size=4096,
112
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: is_llama_config=True,
113
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: max_position_embeddings=4096,
114
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: num_attention_heads=32,
115
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: num_hidden_layers=24,
116
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: num_key_value_heads=32,
117
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: pad_token_id=None,
118
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: pretraining_tp=1,
119
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: rms_norm_eps=1e-05,
120
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: rope_scaling=None,
121
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: rope_theta=10000.0,
122
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: tie_word_embeddings=True,
123
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: use_cache=True,
124
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: vocab_size=50257)
125
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Building model..
126
+ [default0]:07/03/2024 23:44:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Setting PP block ranks...
127
+ [default1]:07/03/2024 23:44:23 [INFO|DP=1|PP=0|TP=0|ip-26-0-169-86]: No checkpoint path provided.
128
+ [default5]:07/03/2024 23:44:23 [INFO|DP=1|PP=1|TP=0|ip-26-0-169-86]: No checkpoint path provided.
129
+ [default2]:07/03/2024 23:44:23 [INFO|DP=2|PP=0|TP=0|ip-26-0-169-86]: No checkpoint path provided.
130
+ [default0]:07/03/2024 23:44:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Total number of parameters: 1.21G (2312.82MiB)
131
+ [default0]:07/03/2024 23:44:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Local number of parameters: 690M (1316.43MiB)
132
+ [default4]:07/03/2024 23:44:23 [INFO|DP=0|PP=1|TP=0|ip-26-0-169-86]: Local number of parameters: 522M (996.40MiB)
133
+ [default4]:07/03/2024 23:44:23 [INFO|DP=0|PP=1|TP=0|ip-26-0-169-86]: [After model building] Memory usage: 1006.41MiB. Peak allocated: 1008.44MiB Peak reserved: 1032.00MiB
134
+ [default4]:07/03/2024 23:44:23 [INFO|DP=0|PP=1|TP=0|ip-26-0-169-86]: No checkpoint path provided.
135
+ [default0]:07/03/2024 23:44:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: [After model building] Memory usage: 1330.44MiB. Peak allocated: 1332.47MiB Peak reserved: 1364.00MiB
136
+ [default0]:07/03/2024 23:44:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: No checkpoint path provided.
137
+ [default0]:07/03/2024 23:44:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Parametrizing model parameters using StandardParametrizator
138
+ [default7]:07/03/2024 23:44:23 [INFO|DP=3|PP=1|TP=0|ip-26-0-169-86]: No checkpoint path provided.
139
+ [default6]:07/03/2024 23:44:23 [INFO|DP=2|PP=1|TP=0|ip-26-0-169-86]: No checkpoint path provided.
140
+ [default3]:07/03/2024 23:44:23 [INFO|DP=3|PP=0|TP=0|ip-26-0-169-86]: No checkpoint path provided.
141
+ [default0]:07/03/2024 23:44:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: [Optimizer Building] Using LearningRateForSP as learning rate
142
+ [default0]:07/03/2024 23:44:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: [ZeRO sharding] Size of optimizer params per rank:
143
+ [default0]:07/03/2024 23:44:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: [ZeRO sharding] DP Rank 0 has 173M out of 690M (25.00%) params' optimizer states
144
+ [default0]:07/03/2024 23:44:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: [ZeRO sharding] DP Rank 1 has 173M out of 690M (25.00%) params' optimizer states
145
+ [default0]:07/03/2024 23:44:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: [ZeRO sharding] DP Rank 2 has 173M out of 690M (25.00%) params' optimizer states
146
+ [default0]:07/03/2024 23:44:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: [ZeRO sharding] DP Rank 3 has 173M out of 690M (25.00%) params' optimizer states
147
+ [default0]:07/03/2024 23:44:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
148
+ [default0]:07/03/2024 23:44:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Using `datasets` library
149
+ [default0]:07/03/2024 23:44:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
150
+ [default0]:07/03/2024 23:44:29 [WARNING|DP=0|PP=0|TP=0|ip-26-0-169-86]: Repo card metadata block was not found. Setting CardData to empty.
151
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
152
+ [default0]:07/03/2024 23:44:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: [Training Plan] There are 1 training stages
153
+ [default0]:07/03/2024 23:44:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: [Stage Training Stage] start from step 1
154
+ [default0]:07/03/2024 23:44:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]:
155
+ [default0]:07/03/2024 23:44:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: [Start training] datetime: 2024-07-03 23:44:31.610525 | mbs: 1 | grad_accum: 256 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
156
+ [default0]:07/03/2024 23:44:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
157
+ [default0]:07/03/2024 23:44:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Memory usage: 4621.51MiB. Peak allocated 4621.51MiB. Peak reserved: 4658.00MiB
158
+ [default1]:07/03/2024 23:44:31 [WARNING|DP=1|PP=0|TP=0|ip-26-0-169-86]: Repo card metadata block was not found. Setting CardData to empty.
159
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
160
+ [default2]:07/03/2024 23:44:31 [WARNING|DP=2|PP=0|TP=0|ip-26-0-169-86]: Repo card metadata block was not found. Setting CardData to empty.
161
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
162
+ [default7]:07/03/2024 23:44:32 [WARNING|DP=3|PP=1|TP=0|ip-26-0-169-86]: Repo card metadata block was not found. Setting CardData to empty.
163
+ [default5]:07/03/2024 23:44:32 [WARNING|DP=1|PP=1|TP=0|ip-26-0-169-86]: Repo card metadata block was not found. Setting CardData to empty.
164
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
165
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
166
+ [default3]:07/03/2024 23:44:32 [WARNING|DP=3|PP=0|TP=0|ip-26-0-169-86]: Repo card metadata block was not found. Setting CardData to empty.
167
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
168
+ [default6]:07/03/2024 23:44:32 [WARNING|DP=2|PP=1|TP=0|ip-26-0-169-86]: Repo card metadata block was not found. Setting CardData to empty.
169
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
170
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
171
+ [default4]:07/03/2024 23:44:32 [WARNING|DP=0|PP=1|TP=0|ip-26-0-169-86]: Repo card metadata block was not found. Setting CardData to empty.
172
+ [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.)
173
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
174
+ [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.)
175
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
176
+ [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.)
177
+ [default7]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
178
+ [default0]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: Attempting to run cuBLAS, but there was no current CUDA context! Attempting to set the primary context... (Triggered internally at ../aten/src/ATen/cuda/CublasHandlePool.cpp:135.)
179
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
180
+ [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.)
181
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
182
+ [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.)
183
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
184
+ [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.)
185
+ [default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
186
+ [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.)
187
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
188
+ [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.)
189
+ [default1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
190
+ [default0]:07/03/2024 23:45:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Memory usage: 4692.54MiB. Peak allocated 12968.78MiB. Peak reserved: 13170.00MiB
191
+ [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
192
+ [default4]: warnings.warn(
193
+ [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
194
+ [default0]: warnings.warn(
195
+ [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
196
+ [default6]: warnings.warn(
197
+ [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
198
+ [default2]: warnings.warn(
199
+ [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
200
+ [default3]: warnings.warn(
201
+ [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
202
+ [default7]: warnings.warn(
203
+ [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
204
+ [default5]: warnings.warn(
205
+ [default4]:07/03/2024 23:45:03 [INFO|DP=0|PP=1|TP=0|ip-26-0-169-86]: iteration: 1 / 20 | consumed_tokens: 4.19M | elapsed_time_per_iteration_ms: 28.7K | tokens_per_sec: 146K | tokens_per_sec_per_gpu: 18.3K | global_batch_size: 1.02K | lm_loss: 11.1 | lr: 0.0001 | model_tflops_per_gpu: 166 | hardware_tflops_per_gpu: 166 | grad_norm: 24.6 | cuda_memory_allocated: 4.78G | cuda_max_memory_reserved: 10.7G | hd_total_memory_tb: 312G | hd_used_memory_tb: 68.8G | hd_free_memory_tb: 243G
206
+ [default0]:07/03/2024 23:45:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Memory usage: 6009.01MiB. Peak allocated 8975.43MiB. Peak reserved: 16138.00MiB
207
+ [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
208
+ [default1]: warnings.warn(
209
+ [default0]:07/03/2024 23:45:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Memory usage: 6009.01MiB. Peak allocated 14183.96MiB. Peak reserved: 16138.00MiB
210
+ [default0]:07/03/2024 23:45:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Memory usage: 6009.01MiB. Peak allocated 8975.43MiB. Peak reserved: 16138.00MiB
211
+ [default4]:07/03/2024 23:45:23 [INFO|DP=0|PP=1|TP=0|ip-26-0-169-86]: iteration: 2 / 20 | consumed_tokens: 8.39M | elapsed_time_per_iteration_ms: 19.9K | tokens_per_sec: 210K | tokens_per_sec_per_gpu: 26.3K | global_batch_size: 1.02K | lm_loss: 11.1 | lr: 9.53e-05 | model_tflops_per_gpu: 239 | hardware_tflops_per_gpu: 239 | grad_norm: 24.8 | cuda_memory_allocated: 4.78G | cuda_max_memory_reserved: 10.7G | hd_total_memory_tb: 312G | hd_used_memory_tb: 68.8G | hd_free_memory_tb: 243G
212
+ [default4]:07/03/2024 23:45:45 [INFO|DP=0|PP=1|TP=0|ip-26-0-169-86]: iteration: 3 / 20 | consumed_tokens: 12.6M | elapsed_time_per_iteration_ms: 21.9K | tokens_per_sec: 191K | tokens_per_sec_per_gpu: 23.9K | global_batch_size: 1.02K | lm_loss: 10.5 | lr: 9.05e-05 | model_tflops_per_gpu: 217 | hardware_tflops_per_gpu: 217 | grad_norm: 195 | cuda_memory_allocated: 4.78G | cuda_max_memory_reserved: 10.7G | hd_total_memory_tb: 312G | hd_used_memory_tb: 68.8G | hd_free_memory_tb: 243G
213
+ [default0]:07/03/2024 23:45:45 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Memory usage: 6009.01MiB. Peak allocated 14183.96MiB. Peak reserved: 16138.00MiB
214
+ [default0]:07/03/2024 23:45:45 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Memory usage: 6009.01MiB. Peak allocated 8975.43MiB. Peak reserved: 16138.00MiB
215
+ [default0]:STAGE:2024-07-03 23:45:45 2278697:2278697 ActivityProfilerController.cpp:314] Completed Stage: Warm Up
216
+ [default4]:07/03/2024 23:46:13 [INFO|DP=0|PP=1|TP=0|ip-26-0-169-86]: iteration: 4 / 20 | consumed_tokens: 16.8M | elapsed_time_per_iteration_ms: 28.3K | tokens_per_sec: 148K | tokens_per_sec_per_gpu: 18.5K | global_batch_size: 1.02K | lm_loss: 13.9 | lr: 8.58e-05 | model_tflops_per_gpu: 168 | hardware_tflops_per_gpu: 168 | grad_norm: 18 | cuda_memory_allocated: 4.78G | cuda_max_memory_reserved: 10.7G | hd_total_memory_tb: 312G | hd_used_memory_tb: 68.8G | hd_free_memory_tb: 243G
217
+ [default0]:07/03/2024 23:46:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Memory usage: 6009.01MiB. Peak allocated 14183.96MiB. Peak reserved: 16138.00MiB
218
+ [default0]:07/03/2024 23:46:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Memory usage: 6009.01MiB. Peak allocated 8975.43MiB. Peak reserved: 16138.00MiB
219
+ [default4]:07/03/2024 23:46:41 [INFO|DP=0|PP=1|TP=0|ip-26-0-169-86]: iteration: 5 / 20 | consumed_tokens: 21M | elapsed_time_per_iteration_ms: 28.2K | tokens_per_sec: 149K | tokens_per_sec_per_gpu: 18.6K | global_batch_size: 1.02K | lm_loss: 9.72 | lr: 8.11e-05 | model_tflops_per_gpu: 169 | hardware_tflops_per_gpu: 169 | grad_norm: 20.2
220
+ [default0]:07/03/2024 23:46:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-169-86]: Memory usage: 6009.01MiB. Peak allocated 14183.96MiB. Peak reserved: 16138.00MiB
221
+ [default4]:07/03/2024 23:47:09 [INFO|DP=0|PP=1|TP=0|ip-26-0-169-86]: iteration: 6 / 20 | consumed_tokens: 25.2M | elapsed_time_per_iteration_ms: 28.3K | tokens_per_sec: 148K | tokens_per_sec_per_gpu: 18.5K | global_batch_size: 1.02K | lm_loss: 13.6 | lr: 7.63e-05 | model_tflops_per_gpu: 168 | hardware_tflops_per_gpu: 168 | grad_norm: 98.4
222
+ [default0]:STAGE:2024-07-03 23:48:38 2278697:2278697 ActivityProfilerController.cpp:320] Completed Stage: Collection
223
+ [default0]:STAGE:2024-07-03 23:48:45 2278697:2278697 ActivityProfilerController.cpp:324] Completed Stage: Post Processing
224
+ [default4]:[rank4]:[E ProcessGroupNCCL.cpp:563] [Rank 1] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=13840, OpType=RECV, NumelIn=7, NumelOut=7, Timeout(ms)=600000) ran for 600011 milliseconds before timing out.
225
+ [default4]:[rank4]: Traceback (most recent call last):
226
+ [default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
227
+ [default4]:[rank4]: trainer.train(dataloader)
228
+ [default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
229
+ [default4]:[rank4]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
230
+ [default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
231
+ [default4]:[rank4]: outputs = self.pipeline_engine.train_batch_iter(
232
+ [default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
233
+ [default4]:[rank4]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
234
+ [default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
235
+ [default4]:[rank4]: output = model(**micro_batch)
236
+ [default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
237
+ [default4]:[rank4]: return self._call_impl(*args, **kwargs)
238
+ [default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
239
+ [default4]:[rank4]: return forward_call(*args, **kwargs)
240
+ [default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
241
+ [default4]:[rank4]: sharded_logits = self.model(
242
+ [default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
243
+ [default4]:[rank4]: return self._call_impl(*args, **kwargs)
244
+ [default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
245
+ [default4]:[rank4]: return forward_call(*args, **kwargs)
246
+ [default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
247
+ [default4]:[rank4]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
248
+ [default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
249
+ [default4]:[rank4]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
250
+ [default4]:[rank4]: 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
+ [default4]:[rank4]: return self._call_impl(*args, **kwargs)
252
+ [default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
253
+ [default4]:[rank4]: return forward_call(*args, **kwargs)
254
+ [default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 126, in forward
255
+ [default4]:[rank4]: new_kwargs[name] = recv_from_pipeline_state_buffer(
256
+ [default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/functional.py", line 117, in recv_from_pipeline_state_buffer
257
+ [default4]:[rank4]: pipeline_state.run_communication()
258
+ [default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/state.py", line 150, in run_communication
259
+ [default4]:[rank4]: recv_activation_tensor = recv_activation()
260
+ [default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/state.py", line 31, in __call__
261
+ [default4]:[rank4]: return self.p2p.recv_tensors(num_tensors=1, from_rank=self.from_rank)[0]
262
+ [default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/p2p.py", line 353, in recv_tensors
263
+ [default4]:[rank4]: buffers, futures = self.irecv_tensors(num_tensors=num_tensors, from_rank=from_rank, tag=tag)
264
+ [default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/p2p.py", line 326, in irecv_tensors
265
+ [default4]:[rank4]: meta = self._recv_meta(from_rank=from_rank, tag=tag)
266
+ [default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/p2p.py", line 267, in _recv_meta
267
+ [default4]:[rank4]: self.second_metadata = torch.empty(second_metadata_num_elements, dtype=torch.long, device=self.device)
268
+ [default4]:[rank4]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate more than 1EB memory.
269
+ [default4]:[rank4]:[E ProcessGroupNCCL.cpp:1537] [PG 4 Rank 1] Timeout at NCCL work: 13840, last enqueued NCCL work: 13840, last completed NCCL work: 13839.
270
+ [default4]:[rank4]:[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.
271
+ [default4]:[rank4]:[E ProcessGroupNCCL.cpp:583] [Rank 1] To avoid data inconsistency, we are taking the entire process down.
272
+ [default4]:[rank4]:[E ProcessGroupNCCL.cpp:1414] [PG 4 Rank 1] Process group watchdog thread terminated with exception: [Rank 1] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=13840, OpType=RECV, NumelIn=7, NumelOut=7, Timeout(ms)=600000) ran for 600011 milliseconds before timing out.
273
+ [default4]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
274
+ [default4]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7f2df746a897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
275
+ [default4]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7f2df8743c62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
276
+ [default4]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7f2df8748a80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
277
+ [default4]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7f2df8749dcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
278
+ [default4]:frame #4: <unknown function> + 0xd3e95 (0x7f2e441e2e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
279
+ [default4]:frame #5: <unknown function> + 0x8609 (0x7f2e49229609 in /lib/x86_64-linux-gnu/libpthread.so.0)
280
+ [default4]:frame #6: clone + 0x43 (0x7f2e48ff4353 in /lib/x86_64-linux-gnu/libc.so.6)
281
+ [default4]:
282
+ [default4]:terminate called after throwing an instance of 'c10::DistBackendError'
283
+ [default4]: what(): [PG 4 Rank 1] Process group watchdog thread terminated with exception: [Rank 1] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=13840, OpType=RECV, NumelIn=7, NumelOut=7, Timeout(ms)=600000) ran for 600011 milliseconds before timing out.
284
+ [default4]:Exception raised from checkTimeout at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:565 (most recent call first):
285
+ [default4]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7f2df746a897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
286
+ [default4]:frame #1: c10d::ProcessGroupNCCL::WorkNCCL::checkTimeout(std::optional<std::chrono::duration<long, std::ratio<1l, 1000l> > >) + 0x1d2 (0x7f2df8743c62 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
287
+ [default4]:frame #2: c10d::ProcessGroupNCCL::watchdogHandler() + 0x1a0 (0x7f2df8748a80 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
288
+ [default4]:frame #3: c10d::ProcessGroupNCCL::ncclCommWatchdog() + 0x10c (0x7f2df8749dcc in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
289
+ [default4]:frame #4: <unknown function> + 0xd3e95 (0x7f2e441e2e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
290
+ [default4]:frame #5: <unknown function> + 0x8609 (0x7f2e49229609 in /lib/x86_64-linux-gnu/libpthread.so.0)
291
+ [default4]:frame #6: clone + 0x43 (0x7f2e48ff4353 in /lib/x86_64-linux-gnu/libc.so.6)
292
+ [default4]:
293
+ [default4]:Exception raised from ncclCommWatchdog at ../torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:1418 (most recent call first):
294
+ [default4]:frame #0: c10::Error::Error(c10::SourceLocation, std::string) + 0x57 (0x7f2df746a897 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libc10.so)
295
+ [default4]:frame #1: <unknown function> + 0xe32119 (0x7f2df83cd119 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/lib/libtorch_cuda.so)
296
+ [default4]:frame #2: <unknown function> + 0xd3e95 (0x7f2e441e2e95 in /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/../lib/libstdc++.so.6)
297
+ [default4]:frame #3: <unknown function> + 0x8609 (0x7f2e49229609 in /lib/x86_64-linux-gnu/libpthread.so.0)
298
+ [default4]:frame #4: clone + 0x43 (0x7f2e48ff4353 in /lib/x86_64-linux-gnu/libc.so.6)
299
+ [default4]:
300
+ W0703 23:57:11.869000 140140090980160 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2278697 closing signal SIGTERM
301
+ W0703 23:57:11.869000 140140090980160 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2278698 closing signal SIGTERM
302
+ W0703 23:57:11.869000 140140090980160 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2278699 closing signal SIGTERM
303
+ W0703 23:57:11.869000 140140090980160 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2278700 closing signal SIGTERM
304
+ W0703 23:57:11.871000 140140090980160 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2278702 closing signal SIGTERM
305
+ W0703 23:57:11.873000 140140090980160 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2278703 closing signal SIGTERM
306
+ W0703 23:57:11.874000 140140090980160 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2278704 closing signal SIGTERM
307
+ E0703 23:57:16.467000 140140090980160 torch/distributed/elastic/multiprocessing/api.py:826] failed (exitcode: -6) local_rank: 4 (pid: 2278701) of binary: /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/python3.10
308
+ Traceback (most recent call last):
309
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/torchrun", line 8, in <module>
310
+ sys.exit(main())
311
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 347, in wrapper
312
+ return f(*args, **kwargs)
313
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 879, in main
314
+ run(args)
315
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 870, in run
316
+ elastic_launch(
317
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 132, in __call__
318
+ return launch_agent(self._config, self._entrypoint, list(args))
319
+ File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 263, in launch_agent
320
+ raise ChildFailedError(
321
+ torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
322
+ ============================================================
323
+ /fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py FAILED
324
+ ------------------------------------------------------------
325
+ Failures:
326
+ <NO_OTHER_FAILURES>
327
+ ------------------------------------------------------------
328
+ Root Cause (first observed failure):
329
+ [0]:
330
+ time : 2024-07-03_23:57:11
331
+ host : ip-26-0-169-86.ec2.internal
332
+ rank : 4 (local_rank: 4)
333
+ exitcode : -6 (pid: 2278701)
334
+ error_file: <N/A>
335
+ traceback : Signal 6 (SIGABRT) received by PID 2278701
336
+ ============================================================
337
+ srun: error: ip-26-0-169-86: task 0: Exited with exit code 1
338
+ 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.
339
+
llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/profiler/ip-26-0-169-86_2278697.1720051013128234436.pt.trace.json.tmp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:67ace9238a799b090ece45c559db0f4d545f4af885cc301f5d2296a345930607
3
+ size 2548537898
llama-1B/8_GPUS/dp-4_tp-1_pp-2_mbz-1/status.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ oom