Hugging Face's logo

nanotron
/
old_bench

Model card Files Community
3outeille HF staff commited on
Commit
8a76c11
•
1 Parent(s): 1fd7d1a

Upload llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8

Browse files
Files changed (6) hide show
  1. .gitattributes +1 -0
  2. llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/bench.slurm +111 -0
  3. llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/config.yaml +90 -0
  4. llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/log.out +333 -0
  5. llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/profiler/ip-26-0-171-62_3731127.1719946421954651053.pt.trace.json +3 -0
  6. llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/status.txt +1 -0
.gitattributes CHANGED
@@ -48,3 +48,4 @@ llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/profiler/ip-26-0-171-62_3711718.171994589
48
  llama-1B/16_GPUS/dp-4_tp-2_pp-2_mbz-2/profiler/ip-26-0-171-102_3600759.1719945485932222180.pt.trace.json filter=lfs diff=lfs merge=lfs -text
49
  llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-2/profiler/ip-26-0-160-192_925006.1719945618152196843.pt.trace.json filter=lfs diff=lfs merge=lfs -text
50
  llama-1B/16_GPUS/dp-1_tp-8_pp-2_mbz-16/profiler/ip-26-0-165-24_655682.1719946219098357028.pt.trace.json filter=lfs diff=lfs merge=lfs -text
 
 
48
  llama-1B/16_GPUS/dp-4_tp-2_pp-2_mbz-2/profiler/ip-26-0-171-102_3600759.1719945485932222180.pt.trace.json filter=lfs diff=lfs merge=lfs -text
49
  llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-2/profiler/ip-26-0-160-192_925006.1719945618152196843.pt.trace.json filter=lfs diff=lfs merge=lfs -text
50
  llama-1B/16_GPUS/dp-1_tp-8_pp-2_mbz-16/profiler/ip-26-0-165-24_655682.1719946219098357028.pt.trace.json filter=lfs diff=lfs merge=lfs -text
51
+ llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/profiler/ip-26-0-171-62_3731127.1719946421954651053.pt.trace.json filter=lfs diff=lfs merge=lfs -text
llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/bench.slurm ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+
3
+ #SBATCH --job-name=bench_cluster
4
+ #SBATCH --time=00:59:00
5
+ #SBATCH --partition=hopper-prod
6
+ #SBATCH --nodes=2
7
+ #SBATCH --gres=gpu:8
8
+ #SBATCH --qos=high
9
+ #SBATCH --ntasks-per-node=1
10
+ #SBATCH --cpus-per-task=96
11
+ #SBATCH --exclusive
12
+ #SBATCH --output=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/log.out
13
+ #SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/log.out
14
+
15
+ # Function to update status based on squeue output
16
+ update_status() {
17
+ job_id=$1
18
+ status_file=$2
19
+ # For unknown reasons, it doenst update status for pending. It only works for running
20
+ while true; do
21
+ job_status=$(squeue --job $job_id --noheader --format=%T)
22
+ echo "Job status: $job_status"
23
+ if [ -z "$job_status" ]; then
24
+ # Job has finished or is not found
25
+ break
26
+ elif [ "$job_status" = "RUNNING" ]; then
27
+ printf "running" > $status_file
28
+ break
29
+ fi
30
+ sleep 10
31
+ done
32
+ }
33
+
34
+ # Misc initializations.
35
+ echo "========================"
36
+ echo "START TIME: $(date)"
37
+ source /fsx/ferdinandmom/miniforge3/etc/profile.d/conda.sh
38
+ conda activate /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster
39
+ echo python3 version = $(python3 --version)
40
+ echo "========================"
41
+
42
+ # Slurm stuff
43
+ export HOSTNAMES=$(scontrol show hostnames "$SLURM_JOB_NODELIST")
44
+ export MASTER_ADDR=$(scontrol show hostnames "$SLURM_JOB_NODELIST" | head -n 1)
45
+ export MASTER_PORT=$((1024 + RANDOM % 64511))
46
+
47
+ export TMPDIR=/scratch
48
+ export HF_DATASETS_CACHE="/admin/home/ferdinand_mom/.cache"
49
+ export CUBLAS_WORKSPACE_CONFIG=":4096:8"
50
+ export CUDA_DEVICE_MAX_CONNECTIONS="1"
51
+
52
+ huggingface-cli login --token $HUGGINGFACE_TOKEN
53
+
54
+
55
+ NANOTRON_REPO="/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron"
56
+ CMD="$NANOTRON_REPO/run_train.py --config-file /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/config.yaml"
57
+
58
+ LAUNCHER="torchrun \
59
+ --nproc_per_node 8 \
60
+ --nnodes 2 \
61
+ --rdzv_endpoint ${MASTER_ADDR}:${MASTER_PORT} \
62
+ --rdzv_backend c10d \
63
+ --max_restarts 0 \
64
+ --tee 3 \
65
+ --node_rank ${SLURM_PROCID}"
66
+
67
+ # Checkout the bench_cluster branch
68
+ cd $NANOTRON_REPO
69
+ git checkout bench_cluster
70
+ cd ..
71
+ # Get the current job ID
72
+ job_id=${SLURM_JOB_ID}
73
+
74
+ # Update status to "pending" or "running" in the background
75
+ update_status $job_id /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/status.txt &
76
+
77
+ # Run the main command
78
+ srun -u $LAUNCHER $CMD
79
+ exit_status=$?
80
+
81
+ # Update status based on the exit status of `srun`
82
+ if [ $exit_status -eq 0 ]; then
83
+ printf "completed" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/status.txt
84
+ else
85
+ if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/log.out; then
86
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/status.txt
87
+ elif grep -q " CUDA error: an illegal memory access" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/log.out; then
88
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/status.txt
89
+ elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/log.out; then
90
+ printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/status.txt
91
+ else
92
+ printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/status.txt
93
+ fi
94
+ fi
95
+
96
+ # Run the report script if the job completed successfully
97
+ if [ $exit_status -eq 0 ]; then
98
+ python /fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py report --inp_dir /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8 --is_logs
99
+ python /fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py report --inp_dir /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8 --is_profiler
100
+ fi
101
+
102
+
103
+ # Push to hub the folder using huggingface_cli
104
+ huggingface-cli upload nanotron/bench_cluster /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8 llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8 --commit-message "Upload llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8"
105
+
106
+ # Verify the upload
107
+ if [ $? -eq 0 ]; then
108
+ echo "Uploading to Huggingface Hub successful"
109
+ else
110
+ echo "Failed to upload to Huggingface Hub"
111
+ fi
llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/config.yaml ADDED
@@ -0,0 +1,90 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ general:
2
+ project: bench_cluster
3
+ seed: 42
4
+ model:
5
+ ddp_bucket_cap_mb: 25
6
+ dtype: bfloat16
7
+ init_method:
8
+ std: 0.025
9
+ make_vocab_size_divisible_by: 1
10
+ model_config:
11
+ bos_token_id: 1
12
+ eos_token_id: 2
13
+ hidden_act: silu
14
+ hidden_size: 2048
15
+ initializer_range: 0.02
16
+ intermediate_size: 4096
17
+ is_llama_config: true
18
+ max_position_embeddings: 4096
19
+ num_attention_heads: 32
20
+ num_hidden_layers: 24
21
+ num_key_value_heads: 32
22
+ pad_token_id: null
23
+ pretraining_tp: 1
24
+ rms_norm_eps: 1.0e-05
25
+ rope_scaling: null
26
+ rope_theta: 10000.0
27
+ tie_word_embeddings: true
28
+ use_cache: true
29
+ vocab_size: 50257
30
+ optimizer:
31
+ accumulate_grad_in_fp32: true
32
+ clip_grad: 1.0
33
+ learning_rate_scheduler:
34
+ learning_rate: 0.0001
35
+ lr_decay_style: linear
36
+ lr_warmup_style: linear
37
+ lr_warmup_steps: 1
38
+ min_decay_lr: 1.0e-05
39
+ optimizer_factory:
40
+ adam_beta1: 0.9
41
+ adam_beta2: 0.95
42
+ adam_eps: 1.0e-08
43
+ name: adamW
44
+ torch_adam_is_fused: true
45
+ weight_decay: 0.01
46
+ zero_stage: 1
47
+ parallelism:
48
+ dp: 2
49
+ expert_parallel_size: 1
50
+ pp: 4
51
+ pp_engine: 1f1b
52
+ tp: 2
53
+ tp_linear_async_communication: false
54
+ tp_mode: REDUCE_SCATTER
55
+ profiler:
56
+ profiler_export_path: /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8
57
+ tokenizer:
58
+ tokenizer_max_length: null
59
+ tokenizer_name_or_path: openai-community/gpt2
60
+ tokenizer_revision: null
61
+ data_stages:
62
+ - name: Training Stage
63
+ start_training_step: 1
64
+ data:
65
+ dataset:
66
+ dataset_overwrite_cache: false
67
+ dataset_processing_num_proc_per_process: 64
68
+ hf_dataset_config_name: null
69
+ hf_dataset_or_datasets: roneneldan/TinyStories
70
+ hf_dataset_splits: train
71
+ text_column_name: text
72
+ num_loading_workers: 32
73
+ seed: 42
74
+ lighteval: null
75
+ tokens:
76
+ train_steps: 20
77
+ val_check_interval: -1
78
+ batch_accumulation_per_replica: 64
79
+ limit_test_batches: 0
80
+ limit_val_batches: 0
81
+ micro_batch_size: 8
82
+ sequence_length: 4096
83
+ logging:
84
+ iteration_step_info_interval: 1
85
+ log_level: info
86
+ log_level_replica: info
87
+ checkpoints:
88
+ checkpoint_interval: 100000
89
+ checkpoints_path: /dev/null
90
+ resume_checkpoint_path: null
llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/log.out ADDED
@@ -0,0 +1,333 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ========================
2
+ START TIME: Tue Jul 2 18:48:40 UTC 2024
3
+ python3 version = Python 3.10.14
4
+ ========================
5
+ The token has not been saved to the git credentials helper. Pass `add_to_git_credential=True` in this function directly or `--add-to-git-credential` if using via `huggingface-cli` if you want to set the git credential as well.
6
+ Token is valid (permission: write).
7
+ Your token has been saved to /admin/home/ferdinand_mom/.cache/huggingface/token
8
+ Login successful
9
+ Already on 'bench_cluster'
10
+ M examples/config_tiny_llama.py
11
+ M examples/config_tiny_llama.yaml
12
+ M examples/train_tiny_llama.sh
13
+ M src/nanotron/models/llama.py
14
+ M src/nanotron/trainer.py
15
+ Your branch is up to date with 'origin/bench_cluster'.
16
+ Job status: RUNNING
17
+ W0702 18:48:43.508000 139657118627648 torch/distributed/run.py:757]
18
+ W0702 18:48:43.508000 139657118627648 torch/distributed/run.py:757] *****************************************
19
+ W0702 18:48:43.508000 139657118627648 torch/distributed/run.py:757] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
20
+ W0702 18:48:43.508000 139657118627648 torch/distributed/run.py:757] *****************************************
21
+ W0702 18:48:43.543000 140183401318208 torch/distributed/run.py:757]
22
+ W0702 18:48:43.543000 140183401318208 torch/distributed/run.py:757] *****************************************
23
+ W0702 18:48:43.543000 140183401318208 torch/distributed/run.py:757] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
24
+ W0702 18:48:43.543000 140183401318208 torch/distributed/run.py:757] *****************************************
25
+ [default0]:07/02/2024 18:49:01 [WARNING|DP=0|PP=0|TP=0|ip-26-0-171-62]: [Vocab Size Padding] Padded vocab (size: 50257) with 1 dummy tokens (new size: 50258)
26
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Config:
27
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Config(general=GeneralArgs(project='bench_cluster',
28
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: run='%date_%jobid',
29
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: seed=42,
30
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: step=None,
31
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: consumed_train_samples=None,
32
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: benchmark_csv_path=None,
33
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: ignore_sanity_checks=True),
34
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: parallelism=ParallelismArgs(dp=2,
35
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pp=4,
36
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tp=2,
37
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f34257a8910>,
38
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
39
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tp_linear_async_communication=False,
40
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: expert_parallel_size=1),
41
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
42
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: eos_token_id=2,
43
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hidden_act='silu',
44
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hidden_size=2048,
45
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: initializer_range=0.02,
46
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: intermediate_size=4096,
47
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: is_llama_config=True,
48
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: max_position_embeddings=4096,
49
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_attention_heads=32,
50
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_hidden_layers=24,
51
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_key_value_heads=32,
52
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pad_token_id=None,
53
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pretraining_tp=1,
54
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rms_norm_eps=1e-05,
55
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rope_scaling=None,
56
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rope_theta=10000.0,
57
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tie_word_embeddings=True,
58
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: use_cache=True,
59
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: vocab_size=50258),
60
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: init_method=RandomInit(std=0.025),
61
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: dtype=torch.bfloat16,
62
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: make_vocab_size_divisible_by=1,
63
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: ddp_bucket_cap_mb=25),
64
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
65
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tokenizer_revision=None,
66
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tokenizer_max_length=None),
67
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
68
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: checkpoint_interval=100000,
69
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: save_initial_state=False,
70
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: resume_checkpoint_path=None,
71
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: checkpoints_path_is_shared_file_system=False),
72
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: logging=LoggingArgs(log_level='info',
73
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: log_level_replica='info',
74
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration_step_info_interval=1),
75
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tokens=TokensArgs(sequence_length=4096,
76
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: train_steps=20,
77
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: micro_batch_size=8,
78
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: batch_accumulation_per_replica=64,
79
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: val_check_interval=-1,
80
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: limit_val_batches=0,
81
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: limit_test_batches=0),
82
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
83
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: adam_beta1=0.9,
84
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: adam_beta2=0.95,
85
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: torch_adam_is_fused=True,
86
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: name='adamW'),
87
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: zero_stage=1,
88
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: weight_decay=0.01,
89
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: clip_grad=1.0,
90
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: accumulate_grad_in_fp32=True,
91
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
92
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lr_warmup_steps=1,
93
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lr_warmup_style='linear',
94
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lr_decay_style='linear',
95
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lr_decay_steps=19,
96
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lr_decay_starting_step=None,
97
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: min_decay_lr=1e-05)),
98
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: data_stages=[DatasetStageArgs(name='Training Stage',
99
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: start_training_step=1,
100
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
101
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hf_dataset_splits='train',
102
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hf_dataset_config_name=None,
103
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: dataset_processing_num_proc_per_process=64,
104
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: dataset_overwrite_cache=False,
105
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: text_column_name='text'),
106
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: seed=42,
107
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_loading_workers=32))],
108
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8')),
109
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lighteval=None)
110
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Model Config:
111
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: LlamaConfig(bos_token_id=1,
112
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: eos_token_id=2,
113
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hidden_act='silu',
114
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hidden_size=2048,
115
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: initializer_range=0.02,
116
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: intermediate_size=4096,
117
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: is_llama_config=True,
118
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: max_position_embeddings=4096,
119
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_attention_heads=32,
120
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_hidden_layers=24,
121
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_key_value_heads=32,
122
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pad_token_id=None,
123
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pretraining_tp=1,
124
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rms_norm_eps=1e-05,
125
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rope_scaling=None,
126
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rope_theta=10000.0,
127
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tie_word_embeddings=True,
128
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: use_cache=True,
129
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: vocab_size=50258)
130
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Building model..
131
+ [default0]:07/02/2024 18:49:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Setting PP block ranks...
132
+ [default1]:07/02/2024 18:49:13 [INFO|DP=0|PP=0|TP=1|ip-26-0-171-62]: Local number of parameters: 198M (378.21MiB)
133
+ [default1]:07/02/2024 18:49:13 [INFO|DP=0|PP=0|TP=1|ip-26-0-171-62]: [After model building] Memory usage: 385.23MiB. Peak allocated: 387.26MiB Peak reserved: 402.00MiB
134
+ [default1]:07/02/2024 18:49:13 [INFO|DP=0|PP=0|TP=1|ip-26-0-171-62]: No checkpoint path provided.
135
+ [default0]:07/02/2024 18:49:13 [INFO|DP=0|PP=2|TP=0|ip-26-0-171-88]: Local number of parameters: 126M (240.05MiB)
136
+ [default0]:07/02/2024 18:49:13 [INFO|DP=0|PP=2|TP=0|ip-26-0-171-88]: [After model building] Memory usage: 246.06MiB. Peak allocated: 248.09MiB Peak reserved: 262.00MiB
137
+ [default0]:07/02/2024 18:49:13 [INFO|DP=0|PP=2|TP=0|ip-26-0-171-88]: No checkpoint path provided.
138
+ [default0]:07/02/2024 18:49:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Total number of parameters: 1.21G (2313.02MiB)
139
+ [default0]:07/02/2024 18:49:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Local number of parameters: 198M (378.21MiB)
140
+ [default0]:07/02/2024 18:49:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [After model building] Memory usage: 385.23MiB. Peak allocated: 387.26MiB Peak reserved: 402.00MiB
141
+ [default0]:07/02/2024 18:49:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: No checkpoint path provided.
142
+ [default0]:07/02/2024 18:49:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Parametrizing model parameters using StandardParametrizator
143
+ [default5]:07/02/2024 18:49:13 [INFO|DP=0|PP=3|TP=1|ip-26-0-171-88]: Local number of parameters: 135M (258.20MiB)
144
+ [default5]:07/02/2024 18:49:13 [INFO|DP=0|PP=3|TP=1|ip-26-0-171-88]: [After model building] Memory usage: 262.21MiB. Peak allocated: 264.24MiB Peak reserved: 280.00MiB
145
+ [default5]:07/02/2024 18:49:13 [INFO|DP=0|PP=3|TP=1|ip-26-0-171-88]: No checkpoint path provided.
146
+ [default1]:07/02/2024 18:49:13 [INFO|DP=0|PP=2|TP=1|ip-26-0-171-88]: Local number of parameters: 126M (240.05MiB)
147
+ [default1]:07/02/2024 18:49:13 [INFO|DP=0|PP=2|TP=1|ip-26-0-171-88]: [After model building] Memory usage: 246.06MiB. Peak allocated: 248.09MiB Peak reserved: 262.00MiB
148
+ [default1]:07/02/2024 18:49:13 [INFO|DP=0|PP=2|TP=1|ip-26-0-171-88]: No checkpoint path provided.
149
+ [default5]:07/02/2024 18:49:13 [INFO|DP=0|PP=1|TP=1|ip-26-0-171-62]: Local number of parameters: 147M (280.05MiB)
150
+ [default5]:07/02/2024 18:49:13 [INFO|DP=0|PP=1|TP=1|ip-26-0-171-62]: [After model building] Memory usage: 287.07MiB. Peak allocated: 289.10MiB Peak reserved: 302.00MiB
151
+ [default5]:07/02/2024 18:49:13 [INFO|DP=0|PP=1|TP=1|ip-26-0-171-62]: No checkpoint path provided.
152
+ [default4]:07/02/2024 18:49:13 [INFO|DP=0|PP=1|TP=0|ip-26-0-171-62]: Local number of parameters: 147M (280.05MiB)
153
+ [default4]:07/02/2024 18:49:13 [INFO|DP=0|PP=1|TP=0|ip-26-0-171-62]: [After model building] Memory usage: 287.07MiB. Peak allocated: 289.10MiB Peak reserved: 302.00MiB
154
+ [default4]:07/02/2024 18:49:13 [INFO|DP=0|PP=1|TP=0|ip-26-0-171-62]: No checkpoint path provided.
155
+ [default4]:07/02/2024 18:49:13 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: Local number of parameters: 135M (258.20MiB)
156
+ [default4]:07/02/2024 18:49:13 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: [After model building] Memory usage: 262.21MiB. Peak allocated: 264.24MiB Peak reserved: 280.00MiB
157
+ [default4]:07/02/2024 18:49:13 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: No checkpoint path provided.
158
+ [default2]:07/02/2024 18:49:14 [INFO|DP=1|PP=0|TP=0|ip-26-0-171-62]: No checkpoint path provided.
159
+ [default3]:07/02/2024 18:49:14 [INFO|DP=1|PP=0|TP=1|ip-26-0-171-62]: No checkpoint path provided.
160
+ [default3]:07/02/2024 18:49:14 [INFO|DP=1|PP=2|TP=1|ip-26-0-171-88]: No checkpoint path provided.
161
+ [default7]:07/02/2024 18:49:14 [INFO|DP=1|PP=3|TP=1|ip-26-0-171-88]: No checkpoint path provided.
162
+ [default2]:07/02/2024 18:49:14 [INFO|DP=1|PP=2|TP=0|ip-26-0-171-88]: No checkpoint path provided.
163
+ [default6]:07/02/2024 18:49:14 [INFO|DP=1|PP=1|TP=0|ip-26-0-171-62]: No checkpoint path provided.
164
+ [default7]:07/02/2024 18:49:14 [INFO|DP=1|PP=1|TP=1|ip-26-0-171-62]: No checkpoint path provided.
165
+ [default6]:07/02/2024 18:49:14 [INFO|DP=1|PP=3|TP=0|ip-26-0-171-88]: No checkpoint path provided.
166
+ [default0]:07/02/2024 18:49:17 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [Optimizer Building] Using LearningRateForSP as learning rate
167
+ [default0]:07/02/2024 18:49:17 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] Size of optimizer params per rank:
168
+ [default0]:07/02/2024 18:49:17 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 0 has 99.1M out of 198M (50.00%) params' optimizer states
169
+ [default0]:07/02/2024 18:49:17 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 1 has 99.1M out of 198M (50.00%) params' optimizer states
170
+ [default0]:07/02/2024 18:49:17 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
171
+ [default0]:07/02/2024 18:49:17 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Using `datasets` library
172
+ [default0]:07/02/2024 18:49:17 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
173
+ [default0]:07/02/2024 18:49:18 [WARNING|DP=0|PP=0|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
174
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
175
+ [default0]:07/02/2024 18:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [Training Plan] There are 1 training stages
176
+ [default0]:07/02/2024 18:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [Stage Training Stage] start from step 1
177
+ [default0]:07/02/2024 18:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]:
178
+ [default0]:07/02/2024 18:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [Start training] datetime: 2024-07-02 18:49:18.786831 | mbs: 8 | grad_accum: 64 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
179
+ [default0]:07/02/2024 18:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
180
+ [default0]:07/02/2024 18:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 1519.87MiB. Peak allocated 1519.87MiB. Peak reserved: 1540.00MiB
181
+ [default3]:07/02/2024 18:49:18 [WARNING|DP=1|PP=2|TP=1|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
182
+ [default3]:07/02/2024 18:49:18 [WARNING|DP=1|PP=0|TP=1|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
183
+ [default2]:07/02/2024 18:49:18 [WARNING|DP=1|PP=0|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
184
+ [default1]:07/02/2024 18:49:18 [WARNING|DP=0|PP=2|TP=1|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
185
+ [default7]:07/02/2024 18:49:18 [WARNING|DP=1|PP=3|TP=1|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
186
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
187
+ [default4]:07/02/2024 18:49:18 [WARNING|DP=0|PP=1|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
188
+ [default6]:07/02/2024 18:49:18 [WARNING|DP=1|PP=1|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
189
+ [default5]:07/02/2024 18:49:18 [WARNING|DP=0|PP=1|TP=1|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
190
+ [default4]:07/02/2024 18:49:18 [WARNING|DP=0|PP=3|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
191
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
192
+ [default6]:07/02/2024 18:49:18 [WARNING|DP=1|PP=3|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
193
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
194
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
195
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
196
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
197
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
198
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
199
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
200
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
201
+ [default0]:07/02/2024 18:49:18 [WARNING|DP=0|PP=2|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
202
+ [default1]:07/02/2024 18:49:19 [WARNING|DP=0|PP=0|TP=1|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
203
+ [default5]:07/02/2024 18:49:19 [WARNING|DP=0|PP=3|TP=1|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
204
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
205
+ [default2]:07/02/2024 18:49:18 [WARNING|DP=1|PP=2|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
206
+ [default7]:07/02/2024 18:49:18 [WARNING|DP=1|PP=1|TP=1|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
207
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
208
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
209
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
210
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
211
+ [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.)
212
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
213
+ [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.)
214
+ [default7]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
215
+ [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.)
216
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
217
+ [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.)
218
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
219
+ [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.)
220
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
221
+ [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.)
222
+ [default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
223
+ [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.)
224
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
225
+ [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.)
226
+ [default1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
227
+ [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.)
228
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
229
+ [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.)
230
+ [default7]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
231
+ [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.)
232
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
233
+ [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.)
234
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
235
+ [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.)
236
+ [default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
237
+ [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.)
238
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
239
+ [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.)
240
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
241
+ [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.)
242
+ [default1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
243
+ [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
244
+ [default5]: warnings.warn(
245
+ [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
246
+ [default4]: warnings.warn(
247
+ [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
248
+ [default1]: warnings.warn(
249
+ [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
250
+ [default0]: warnings.warn(
251
+ [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
252
+ [default4]: warnings.warn(
253
+ [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
254
+ [default5]: warnings.warn(
255
+ [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
256
+ [default1]: warnings.warn(
257
+ [default0]:07/02/2024 18:50:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 1587.65MiB. Peak allocated 39608.35MiB. Peak reserved: 39938.00MiB
258
+ [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
259
+ [default0]: warnings.warn(
260
+ [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
261
+ [default6]: warnings.warn(
262
+ [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
263
+ [default7]: warnings.warn(
264
+ [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
265
+ [default2]: warnings.warn(
266
+ [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
267
+ [default3]: warnings.warn(
268
+ [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
269
+ [default6]: warnings.warn(
270
+ [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
271
+ [default7]: warnings.warn(
272
+ [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
273
+ [default2]: warnings.warn(
274
+ [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
275
+ [default3]: warnings.warn(
276
+ [default0]:07/02/2024 18:50:11 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 3289.64MiB. Peak reserved: 40318.00MiB
277
+ [default4]:07/02/2024 18:50:11 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 1 / 20 | consumed_tokens: 4.19M | elapsed_time_per_iteration_ms: 47K | tokens_per_sec: 89.3K | tokens_per_sec_per_gpu: 5.58K | global_batch_size: 1.02K | lm_loss: 11.2 | lr: 0.0001 | model_tflops_per_gpu: 50.6 | hardware_tflops_per_gpu: 50.6 | grad_norm: 14.8 | cuda_memory_allocated: 1.7G | cuda_max_memory_reserved: 15.4G | hd_total_memory_tb: 312G | hd_used_memory_tb: 67.8G | hd_free_memory_tb: 244G
278
+ [default0]:07/02/2024 18:50:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
279
+ [default0]:07/02/2024 18:50:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 3289.64MiB. Peak reserved: 41214.00MiB
280
+ [default4]:07/02/2024 18:50:35 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 2 / 20 | consumed_tokens: 8.39M | elapsed_time_per_iteration_ms: 24.2K | tokens_per_sec: 174K | tokens_per_sec_per_gpu: 10.8K | global_batch_size: 1.02K | lm_loss: 11.2 | lr: 9.53e-05 | model_tflops_per_gpu: 98.4 | hardware_tflops_per_gpu: 98.4 | grad_norm: 14.9 | cuda_memory_allocated: 1.7G | cuda_max_memory_reserved: 15.4G | hd_total_memory_tb: 312G | hd_used_memory_tb: 67.8G | hd_free_memory_tb: 244G
281
+ [default0]:07/02/2024 18:50:59 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
282
+ [default0]:STAGE:2024-07-02 18:51:04 3731127:3731127 ActivityProfilerController.cpp:314] Completed Stage: Warm Up
283
+ [default0]:07/02/2024 18:51:04 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 3289.64MiB. Peak reserved: 41214.00MiB
284
+ [default4]:07/02/2024 18:51:04 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 3 / 20 | consumed_tokens: 12.6M | elapsed_time_per_iteration_ms: 28.6K | tokens_per_sec: 147K | tokens_per_sec_per_gpu: 9.18K | global_batch_size: 1.02K | lm_loss: 9.53 | lr: 9.05e-05 | model_tflops_per_gpu: 83.3 | hardware_tflops_per_gpu: 83.3 | grad_norm: 35.8 | cuda_memory_allocated: 1.7G | cuda_max_memory_reserved: 15.4G | hd_total_memory_tb: 312G | hd_used_memory_tb: 67.8G | hd_free_memory_tb: 244G
285
+ [default0]:07/02/2024 18:51:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
286
+ [default4]:07/02/2024 18:51:32 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 4 / 20 | consumed_tokens: 16.8M | elapsed_time_per_iteration_ms: 27.8K | tokens_per_sec: 151K | tokens_per_sec_per_gpu: 9.44K | global_batch_size: 1.02K | lm_loss: 12.3 | lr: 8.58e-05 | model_tflops_per_gpu: 85.7 | hardware_tflops_per_gpu: 85.7 | grad_norm: 37.4 | cuda_memory_allocated: 1.7G | cuda_max_memory_reserved: 15.4G | hd_total_memory_tb: 312G | hd_used_memory_tb: 67.8G | hd_free_memory_tb: 244G
287
+ [default0]:07/02/2024 18:51:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 3289.64MiB. Peak reserved: 41214.00MiB
288
+ [default4]:07/02/2024 18:51:57 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 5 / 20 | consumed_tokens: 21M | elapsed_time_per_iteration_ms: 25.2K | tokens_per_sec: 166K | tokens_per_sec_per_gpu: 10.4K | global_batch_size: 1.02K | lm_loss: 9.94 | lr: 8.11e-05 | model_tflops_per_gpu: 94.2 | hardware_tflops_per_gpu: 94.2 | grad_norm: 14.1
289
+ [default0]:07/02/2024 18:51:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
290
+ [default4]:07/02/2024 18:52:23 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 6 / 20 | consumed_tokens: 25.2M | elapsed_time_per_iteration_ms: 26.3K | tokens_per_sec: 159K | tokens_per_sec_per_gpu: 9.95K | global_batch_size: 1.02K | lm_loss: 9.44 | lr: 7.63e-05 | model_tflops_per_gpu: 90.3 | hardware_tflops_per_gpu: 90.3 | grad_norm: 8.13
291
+ [default0]:STAGE:2024-07-02 18:52:34 3731127:3731127 ActivityProfilerController.cpp:320] Completed Stage: Collection
292
+ [default0]:STAGE:2024-07-02 18:52:35 3731127:3731127 ActivityProfilerController.cpp:324] Completed Stage: Post Processing
293
+ [default0]:07/02/2024 18:54:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
294
+ [default0]:07/02/2024 18:54:24 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
295
+ [default4]:07/02/2024 18:54:24 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 7 / 20 | consumed_tokens: 29.4M | elapsed_time_per_iteration_ms: 121K | tokens_per_sec: 34.7K | tokens_per_sec_per_gpu: 2.17K | global_batch_size: 1.02K | lm_loss: 8.73 | lr: 7.16e-05 | model_tflops_per_gpu: 19.7 | hardware_tflops_per_gpu: 19.7 | grad_norm: 6.04
296
+ [default0]:07/02/2024 18:54:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
297
+ [default4]:07/02/2024 18:54:50 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 8 / 20 | consumed_tokens: 33.6M | elapsed_time_per_iteration_ms: 26K | tokens_per_sec: 161K | tokens_per_sec_per_gpu: 10.1K | global_batch_size: 1.02K | lm_loss: 9.17 | lr: 6.68e-05 | model_tflops_per_gpu: 91.4 | hardware_tflops_per_gpu: 91.4 | grad_norm: 28
298
+ [default0]:07/02/2024 18:55:17 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
299
+ [default4]:07/02/2024 18:55:17 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 9 / 20 | consumed_tokens: 37.7M | elapsed_time_per_iteration_ms: 27.1K | tokens_per_sec: 155K | tokens_per_sec_per_gpu: 9.66K | global_batch_size: 1.02K | lm_loss: 8.33 | lr: 6.21e-05 | model_tflops_per_gpu: 87.6 | hardware_tflops_per_gpu: 87.6 | grad_norm: 9.42
300
+ [default0]:07/02/2024 18:55:44 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
301
+ [default4]:07/02/2024 18:55:44 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 10 / 20 | consumed_tokens: 41.9M | elapsed_time_per_iteration_ms: 26.6K | tokens_per_sec: 157K | tokens_per_sec_per_gpu: 9.84K | global_batch_size: 1.02K | lm_loss: 8.02 | lr: 5.74e-05 | model_tflops_per_gpu: 89.3 | hardware_tflops_per_gpu: 89.3 | grad_norm: 5.24
302
+ [default0]:07/02/2024 18:56:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
303
+ [default4]:07/02/2024 18:56:10 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 11 / 20 | consumed_tokens: 46.1M | elapsed_time_per_iteration_ms: 26.5K | tokens_per_sec: 158K | tokens_per_sec_per_gpu: 9.88K | global_batch_size: 1.02K | lm_loss: 7.85 | lr: 5.26e-05 | model_tflops_per_gpu: 89.6 | hardware_tflops_per_gpu: 89.6 | grad_norm: 4.81
304
+ [default4]:07/02/2024 18:56:38 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 12 / 20 | consumed_tokens: 50.3M | elapsed_time_per_iteration_ms: 27.9K | tokens_per_sec: 151K | tokens_per_sec_per_gpu: 9.41K | global_batch_size: 1.02K | lm_loss: 7.68 | lr: 4.79e-05 | model_tflops_per_gpu: 85.4 | hardware_tflops_per_gpu: 85.4 | grad_norm: 4.49
305
+ [default0]:07/02/2024 18:56:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
306
+ [default4]:07/02/2024 18:57:04 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 13 / 20 | consumed_tokens: 54.5M | elapsed_time_per_iteration_ms: 25.9K | tokens_per_sec: 162K | tokens_per_sec_per_gpu: 10.1K | global_batch_size: 1.02K | lm_loss: 7.53 | lr: 4.32e-05 | model_tflops_per_gpu: 91.7 | hardware_tflops_per_gpu: 91.7 | grad_norm: 4.15
307
+ [default0]:07/02/2024 18:57:04 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
308
+ [default0]:07/02/2024 18:57:30 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
309
+ [default4]:07/02/2024 18:57:30 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 14 / 20 | consumed_tokens: 58.7M | elapsed_time_per_iteration_ms: 25.7K | tokens_per_sec: 163K | tokens_per_sec_per_gpu: 10.2K | global_batch_size: 1.02K | lm_loss: 7.4 | lr: 3.84e-05 | model_tflops_per_gpu: 92.6 | hardware_tflops_per_gpu: 92.6 | grad_norm: 4.07
310
+ [default4]:07/02/2024 18:57:56 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 15 / 20 | consumed_tokens: 62.9M | elapsed_time_per_iteration_ms: 25.8K | tokens_per_sec: 163K | tokens_per_sec_per_gpu: 10.2K | global_batch_size: 1.02K | lm_loss: 7.26 | lr: 3.37e-05 | model_tflops_per_gpu: 92.3 | hardware_tflops_per_gpu: 92.3 | grad_norm: 3.25
311
+ [default0]:07/02/2024 18:57:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
312
+ [default0]:07/02/2024 18:58:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
313
+ [default4]:07/02/2024 18:58:20 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 16 / 20 | consumed_tokens: 67.1M | elapsed_time_per_iteration_ms: 24.1K | tokens_per_sec: 174K | tokens_per_sec_per_gpu: 10.9K | global_batch_size: 1.02K | lm_loss: 7.17 | lr: 2.89e-05 | model_tflops_per_gpu: 98.6 | hardware_tflops_per_gpu: 98.6 | grad_norm: 2.44
314
+ [default0]:07/02/2024 18:58:45 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
315
+ [default4]:07/02/2024 18:58:45 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 17 / 20 | consumed_tokens: 71.3M | elapsed_time_per_iteration_ms: 25.6K | tokens_per_sec: 164K | tokens_per_sec_per_gpu: 10.2K | global_batch_size: 1.02K | lm_loss: 7.1 | lr: 2.42e-05 | model_tflops_per_gpu: 92.9 | hardware_tflops_per_gpu: 92.9 | grad_norm: 2.88
316
+ [default0]:07/02/2024 18:59:11 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
317
+ [default4]:07/02/2024 18:59:11 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 18 / 20 | consumed_tokens: 75.5M | elapsed_time_per_iteration_ms: 25.4K | tokens_per_sec: 165K | tokens_per_sec_per_gpu: 10.3K | global_batch_size: 1.02K | lm_loss: 7.03 | lr: 1.95e-05 | model_tflops_per_gpu: 93.5 | hardware_tflops_per_gpu: 93.5 | grad_norm: 2.75
318
+ [default0]:07/02/2024 18:59:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 2344.10MiB. Peak allocated 40364.80MiB. Peak reserved: 41214.00MiB
319
+ [default4]:07/02/2024 18:59:35 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 19 / 20 | consumed_tokens: 79.7M | elapsed_time_per_iteration_ms: 24.3K | tokens_per_sec: 173K | tokens_per_sec_per_gpu: 10.8K | global_batch_size: 1.02K | lm_loss: 6.96 | lr: 1.47e-05 | model_tflops_per_gpu: 97.9 | hardware_tflops_per_gpu: 97.9 | grad_norm: 2.65
320
+ [default4]:07/02/2024 19:00:03 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-88]: iteration: 20 / 20 | consumed_tokens: 83.9M | elapsed_time_per_iteration_ms: 28.3K | tokens_per_sec: 148K | tokens_per_sec_per_gpu: 9.27K | global_batch_size: 1.02K | lm_loss: 6.91 | lr: 1e-05 | model_tflops_per_gpu: 84.1 | hardware_tflops_per_gpu: 84.1 | grad_norm: 2.47
321
+ W0702 19:00:25.632000 140177734498048 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1252] The node 'ip-26-0-171-88.ec2.internal_719045_0' has failed to send a keep-alive heartbeat to the rendezvous 'none' due to an error of type RendezvousTimeoutError.
322
+ W0702 19:00:25.717000 140183401318208 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1203] The node 'ip-26-0-171-88.ec2.internal_719045_0' has failed to shutdown the rendezvous 'none' due to an error of type RendezvousConnectionError.
323
+ W0702 19:00:25.721000 140183401318208 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1203] The node 'ip-26-0-171-88.ec2.internal_719045_0' has failed to shutdown the rendezvous 'none' due to an error of type RendezvousConnectionError.
324
+ Traceback (most recent call last):
325
+ File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py", line 4, in <module>
326
+ from bench_cluster.submit_jobs import submit_jobs, check_status
327
+ ImportError: cannot import name 'check_status' from 'bench_cluster.submit_jobs' (/fsx/ferdinandmom/ferdinand-hf/bench_cluster/bench_cluster/submit_jobs.py)
328
+ Traceback (most recent call last):
329
+ File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py", line 4, in <module>
330
+ from bench_cluster.submit_jobs import submit_jobs, check_status
331
+ ImportError: cannot import name 'check_status' from 'bench_cluster.submit_jobs' (/fsx/ferdinandmom/ferdinand-hf/bench_cluster/bench_cluster/submit_jobs.py)
332
+ 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.
333
+
llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/profiler/ip-26-0-171-62_3731127.1719946421954651053.pt.trace.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e3db7bd8500c8f3c039774ae4382976ca7c00c66f6c044c791c6564ac70c072c
3
+ size 2706802429
llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/status.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ completed