Hugging Face's logo

nanotron
/
old_bench

Model card Files Community
3outeille HF staff commited on
Commit
a0651f8
•
1 Parent(s): 7293135

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

Browse files
Files changed (6) hide show
  1. .gitattributes +1 -0
  2. llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/bench.slurm +111 -0
  3. llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/config.yaml +90 -0
  4. llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/log.out +297 -0
  5. llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/profiler/ip-26-0-171-62_3711718.1719945894658523576.pt.trace.json +3 -0
  6. llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/status.txt +1 -0
.gitattributes CHANGED
@@ -44,3 +44,4 @@ llama-1B/16_GPUS/dp-8_tp-2_pp-1_mbz-1/profiler/ip-26-0-163-43_666485.17199359321
44
  llama-1B/16_GPUS/dp-2_tp-1_pp-8_mbz-4/profiler/ip-26-0-163-43_687610.1719936674744161495.pt.trace.json filter=lfs diff=lfs merge=lfs -text
45
  llama-1B/16_GPUS/dp-1_tp-4_pp-4_mbz-4/profiler/ip-26-0-169-132_2343928.1719937262116152276.pt.trace.json filter=lfs diff=lfs merge=lfs -text
46
  llama-1B/16_GPUS/dp-2_tp-8_pp-1_mbz-32/profiler/ip-26-0-171-62_3649479.1719945163285636123.pt.trace.json filter=lfs diff=lfs merge=lfs -text
 
 
44
  llama-1B/16_GPUS/dp-2_tp-1_pp-8_mbz-4/profiler/ip-26-0-163-43_687610.1719936674744161495.pt.trace.json filter=lfs diff=lfs merge=lfs -text
45
  llama-1B/16_GPUS/dp-1_tp-4_pp-4_mbz-4/profiler/ip-26-0-169-132_2343928.1719937262116152276.pt.trace.json filter=lfs diff=lfs merge=lfs -text
46
  llama-1B/16_GPUS/dp-2_tp-8_pp-1_mbz-32/profiler/ip-26-0-171-62_3649479.1719945163285636123.pt.trace.json filter=lfs diff=lfs merge=lfs -text
47
+ llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/profiler/ip-26-0-171-62_3711718.1719945894658523576.pt.trace.json filter=lfs diff=lfs merge=lfs -text
llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/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-16_tp-1_pp-1_mbz-4/log.out
13
+ #SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/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-16_tp-1_pp-1_mbz-4/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-16_tp-1_pp-1_mbz-4/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-16_tp-1_pp-1_mbz-4/status.txt
84
+ else
85
+ if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/log.out; then
86
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/status.txt
87
+ elif grep -q " CUDA error: an illegal memory access" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/log.out; then
88
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/status.txt
89
+ elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/log.out; then
90
+ printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/status.txt
91
+ else
92
+ printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/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-16_tp-1_pp-1_mbz-4 --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-16_tp-1_pp-1_mbz-4 --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-16_tp-1_pp-1_mbz-4 llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4 --commit-message "Upload llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4"
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-16_tp-1_pp-1_mbz-4/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: 16
49
+ expert_parallel_size: 1
50
+ pp: 1
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/16_GPUS/dp-16_tp-1_pp-1_mbz-4
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: 16
79
+ limit_test_batches: 0
80
+ limit_val_batches: 0
81
+ micro_batch_size: 4
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-16_tp-1_pp-1_mbz-4/log.out ADDED
@@ -0,0 +1,297 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ========================
2
+ START TIME: Tue Jul 2 18:42:06 UTC 2024
3
+ python3 version = Python 3.10.14
4
+ ========================
5
+ The token has not been saved to the git credentials helper. Pass `add_to_git_credential=True` in this function directly or `--add-to-git-credential` if using via `huggingface-cli` if you want to set the git credential as well.
6
+ Token is valid (permission: write).
7
+ Your token has been saved to /admin/home/ferdinand_mom/.cache/huggingface/token
8
+ Login successful
9
+ Already on 'bench_cluster'
10
+ M examples/config_tiny_llama.py
11
+ M examples/config_tiny_llama.yaml
12
+ M examples/train_tiny_llama.sh
13
+ M src/nanotron/models/llama.py
14
+ M src/nanotron/trainer.py
15
+ Your branch is up to date with 'origin/bench_cluster'.
16
+ Job status: RUNNING
17
+ W0702 18:42:08.900000 140119937877824 torch/distributed/run.py:757]
18
+ W0702 18:42:08.900000 140119937877824 torch/distributed/run.py:757] *****************************************
19
+ W0702 18:42:08.900000 140119937877824 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 18:42:08.900000 140119937877824 torch/distributed/run.py:757] *****************************************
21
+ W0702 18:42:08.941000 140389991290688 torch/distributed/run.py:757]
22
+ W0702 18:42:08.941000 140389991290688 torch/distributed/run.py:757] *****************************************
23
+ W0702 18:42:08.941000 140389991290688 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 18:42:08.941000 140389991290688 torch/distributed/run.py:757] *****************************************
25
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Config:
26
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Config(general=GeneralArgs(project='bench_cluster',
27
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: run='%date_%jobid',
28
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: seed=42,
29
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: step=None,
30
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: consumed_train_samples=None,
31
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: benchmark_csv_path=None,
32
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: ignore_sanity_checks=True),
33
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: parallelism=ParallelismArgs(dp=16,
34
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pp=1,
35
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tp=1,
36
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f066b358910>,
37
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
38
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tp_linear_async_communication=False,
39
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: expert_parallel_size=1),
40
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
41
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: eos_token_id=2,
42
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hidden_act='silu',
43
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hidden_size=2048,
44
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: initializer_range=0.02,
45
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: intermediate_size=4096,
46
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: is_llama_config=True,
47
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: max_position_embeddings=4096,
48
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_attention_heads=32,
49
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_hidden_layers=24,
50
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_key_value_heads=32,
51
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pad_token_id=None,
52
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pretraining_tp=1,
53
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rms_norm_eps=1e-05,
54
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rope_scaling=None,
55
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rope_theta=10000.0,
56
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tie_word_embeddings=True,
57
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: use_cache=True,
58
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: vocab_size=50257),
59
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: init_method=RandomInit(std=0.025),
60
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: dtype=torch.bfloat16,
61
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: make_vocab_size_divisible_by=1,
62
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: ddp_bucket_cap_mb=25),
63
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
64
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tokenizer_revision=None,
65
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tokenizer_max_length=None),
66
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
67
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: checkpoint_interval=100000,
68
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: save_initial_state=False,
69
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: resume_checkpoint_path=None,
70
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: checkpoints_path_is_shared_file_system=False),
71
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: logging=LoggingArgs(log_level='info',
72
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: log_level_replica='info',
73
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration_step_info_interval=1),
74
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tokens=TokensArgs(sequence_length=4096,
75
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: train_steps=20,
76
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: micro_batch_size=4,
77
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: batch_accumulation_per_replica=16,
78
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: val_check_interval=-1,
79
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: limit_val_batches=0,
80
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: limit_test_batches=0),
81
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
82
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: adam_beta1=0.9,
83
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: adam_beta2=0.95,
84
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: torch_adam_is_fused=True,
85
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: name='adamW'),
86
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: zero_stage=1,
87
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: weight_decay=0.01,
88
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: clip_grad=1.0,
89
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: accumulate_grad_in_fp32=True,
90
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
91
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lr_warmup_steps=1,
92
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lr_warmup_style='linear',
93
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lr_decay_style='linear',
94
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lr_decay_steps=19,
95
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lr_decay_starting_step=None,
96
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: min_decay_lr=1e-05)),
97
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: data_stages=[DatasetStageArgs(name='Training Stage',
98
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: start_training_step=1,
99
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
100
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hf_dataset_splits='train',
101
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hf_dataset_config_name=None,
102
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: dataset_processing_num_proc_per_process=64,
103
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: dataset_overwrite_cache=False,
104
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: text_column_name='text'),
105
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: seed=42,
106
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_loading_workers=32))],
107
+ [default0]:07/02/2024 18:42:26 [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-16_tp-1_pp-1_mbz-4')),
108
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: lighteval=None)
109
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Model Config:
110
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: LlamaConfig(bos_token_id=1,
111
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: eos_token_id=2,
112
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hidden_act='silu',
113
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: hidden_size=2048,
114
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: initializer_range=0.02,
115
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: intermediate_size=4096,
116
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: is_llama_config=True,
117
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: max_position_embeddings=4096,
118
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_attention_heads=32,
119
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_hidden_layers=24,
120
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: num_key_value_heads=32,
121
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pad_token_id=None,
122
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: pretraining_tp=1,
123
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rms_norm_eps=1e-05,
124
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rope_scaling=None,
125
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: rope_theta=10000.0,
126
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: tie_word_embeddings=True,
127
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: use_cache=True,
128
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: vocab_size=50257)
129
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Building model..
130
+ [default0]:07/02/2024 18:42:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Setting PP block ranks...
131
+ [default7]:07/02/2024 18:42:35 [INFO|DP=7|PP=0|TP=0|ip-26-0-171-62]: No checkpoint path provided.
132
+ [default0]:07/02/2024 18:42:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Total number of parameters: 1.11G (2116.51MiB)
133
+ [default0]:07/02/2024 18:42:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Local number of parameters: 1.11G (2116.51MiB)
134
+ [default0]:07/02/2024 18:42:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [After model building] Memory usage: 2140.53MiB. Peak allocated: 2338.88MiB Peak reserved: 2392.00MiB
135
+ [default0]:07/02/2024 18:42:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: No checkpoint path provided.
136
+ [default0]:07/02/2024 18:42:35 [INFO|DP=8|PP=0|TP=0|ip-26-0-171-88]: No checkpoint path provided.
137
+ [default1]:07/02/2024 18:42:35 [INFO|DP=1|PP=0|TP=0|ip-26-0-171-62]: No checkpoint path provided.
138
+ [default4]:07/02/2024 18:42:35 [INFO|DP=4|PP=0|TP=0|ip-26-0-171-62]: No checkpoint path provided.
139
+ [default6]:07/02/2024 18:42:35 [INFO|DP=14|PP=0|TP=0|ip-26-0-171-88]: No checkpoint path provided.
140
+ [default2]:07/02/2024 18:42:35 [INFO|DP=2|PP=0|TP=0|ip-26-0-171-62]: No checkpoint path provided.
141
+ [default5]:07/02/2024 18:42:35 [INFO|DP=5|PP=0|TP=0|ip-26-0-171-62]: No checkpoint path provided.
142
+ [default3]:07/02/2024 18:42:35 [INFO|DP=3|PP=0|TP=0|ip-26-0-171-62]: No checkpoint path provided.
143
+ [default6]:07/02/2024 18:42:35 [INFO|DP=6|PP=0|TP=0|ip-26-0-171-62]: No checkpoint path provided.
144
+ [default1]:07/02/2024 18:42:35 [INFO|DP=9|PP=0|TP=0|ip-26-0-171-88]: No checkpoint path provided.
145
+ [default2]:07/02/2024 18:42:35 [INFO|DP=10|PP=0|TP=0|ip-26-0-171-88]: No checkpoint path provided.
146
+ [default5]:07/02/2024 18:42:35 [INFO|DP=13|PP=0|TP=0|ip-26-0-171-88]: No checkpoint path provided.
147
+ [default7]:07/02/2024 18:42:35 [INFO|DP=15|PP=0|TP=0|ip-26-0-171-88]: No checkpoint path provided.
148
+ [default0]:07/02/2024 18:42:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Parametrizing model parameters using StandardParametrizator
149
+ [default3]:07/02/2024 18:42:35 [INFO|DP=11|PP=0|TP=0|ip-26-0-171-88]: No checkpoint path provided.
150
+ [default4]:07/02/2024 18:42:35 [INFO|DP=12|PP=0|TP=0|ip-26-0-171-88]: No checkpoint path provided.
151
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [Optimizer Building] Using LearningRateForSP as learning rate
152
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] Size of optimizer params per rank:
153
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 0 has 69.4M out of 1.11G (6.25%) params' optimizer states
154
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 1 has 69.4M out of 1.11G (6.25%) params' optimizer states
155
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 2 has 69.4M out of 1.11G (6.25%) params' optimizer states
156
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 3 has 69.4M out of 1.11G (6.25%) params' optimizer states
157
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 4 has 69.4M out of 1.11G (6.25%) params' optimizer states
158
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 5 has 69.4M out of 1.11G (6.25%) params' optimizer states
159
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 6 has 69.4M out of 1.11G (6.25%) params' optimizer states
160
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 7 has 69.4M out of 1.11G (6.25%) params' optimizer states
161
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 8 has 69.4M out of 1.11G (6.25%) params' optimizer states
162
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 9 has 69.4M out of 1.11G (6.25%) params' optimizer states
163
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 10 has 69.4M out of 1.11G (6.25%) params' optimizer states
164
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 11 has 69.4M out of 1.11G (6.25%) params' optimizer states
165
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 12 has 69.4M out of 1.11G (6.25%) params' optimizer states
166
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 13 has 69.4M out of 1.11G (6.25%) params' optimizer states
167
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 14 has 69.4M out of 1.11G (6.25%) params' optimizer states
168
+ [default0]:07/02/2024 18:42:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [ZeRO sharding] DP Rank 15 has 69.4M out of 1.11G (6.25%) params' optimizer states
169
+ [default0]:07/02/2024 18:42:45 [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
170
+ [default0]:07/02/2024 18:42:45 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Using `datasets` library
171
+ [default0]:07/02/2024 18:42:45 [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')
172
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
173
+ [default0]:07/02/2024 18:42:46 [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]:07/02/2024 18:42:46 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [Training Plan] There are 1 training stages
175
+ [default0]:07/02/2024 18:42:46 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [Stage Training Stage] start from step 1
176
+ [default0]:07/02/2024 18:42:46 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]:
177
+ [default0]:07/02/2024 18:42:46 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: [Start training] datetime: 2024-07-02 18:42:46.787537 | mbs: 4 | grad_accum: 16 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
178
+ [default0]:07/02/2024 18:42:46 [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
179
+ [default0]:07/02/2024 18:42:46 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 6639.09MiB. Peak allocated 6639.09MiB. Peak reserved: 6892.00MiB
180
+ [default4]:07/02/2024 18:42:46 [WARNING|DP=4|PP=0|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
181
+ [default0]:07/02/2024 18:42:46 [WARNING|DP=8|PP=0|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
182
+ [default6]:07/02/2024 18:42:46 [WARNING|DP=14|PP=0|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
183
+ [default3]:07/02/2024 18:42:46 [WARNING|DP=3|PP=0|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
184
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
185
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
186
+ [default2]:07/02/2024 18:42:46 [WARNING|DP=2|PP=0|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
187
+ [default6]:07/02/2024 18:42:46 [WARNING|DP=6|PP=0|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
188
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
189
+ [default7]:07/02/2024 18:42:46 [WARNING|DP=7|PP=0|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
190
+ [default1]:07/02/2024 18:42:46 [WARNING|DP=9|PP=0|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
191
+ [default2]:07/02/2024 18:42:46 [WARNING|DP=10|PP=0|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
192
+ [default5]:07/02/2024 18:42:46 [WARNING|DP=13|PP=0|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
193
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
194
+ [default7]:07/02/2024 18:42:46 [WARNING|DP=15|PP=0|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
195
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
196
+ [default3]:07/02/2024 18:42:46 [WARNING|DP=11|PP=0|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
197
+ [default2]: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
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
200
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
201
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
202
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
203
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
204
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
205
+ [default5]:07/02/2024 18:42:47 [WARNING|DP=5|PP=0|TP=0|ip-26-0-171-62]: Repo card metadata block was not found. Setting CardData to empty.
206
+ [default1]:07/02/2024 18:42:47 [WARNING|DP=1|PP=0|TP=0|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
+ [default4]:07/02/2024 18:42:47 [WARNING|DP=12|PP=0|TP=0|ip-26-0-171-88]: Repo card metadata block was not found. Setting CardData to empty.
209
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
210
+ [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.)
211
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
212
+ [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.)
213
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
214
+ [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.)
215
+ [default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
216
+ [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.)
217
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
218
+ [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.)
219
+ [default7]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
220
+ [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.)
221
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
222
+ [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.)
223
+ [default1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
224
+ [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.)
225
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
226
+ [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.)
227
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
228
+ [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.)
229
+ [default1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
230
+ [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.)
231
+ [default7]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
232
+ [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.)
233
+ [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.)
234
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
235
+ [default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
236
+ [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.)
237
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
238
+ [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.)
239
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
240
+ [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.)
241
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
242
+ [default0]:07/02/2024 18:42:58 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 6715.38MiB. Peak allocated 42363.18MiB. Peak reserved: 43994.00MiB
243
+ [default0]:07/02/2024 18:43:07 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 1 / 20 | consumed_tokens: 4.19M | elapsed_time_per_iteration_ms: 20.5K | tokens_per_sec: 205K | tokens_per_sec_per_gpu: 12.8K | global_batch_size: 1.02K | lm_loss: 11.3 | lr: 0.0001 | model_tflops_per_gpu: 116 | hardware_tflops_per_gpu: 116 | grad_norm: 33.1 | cuda_memory_allocated: 7.6G | cuda_max_memory_reserved: 46.2G | hd_total_memory_tb: 312G | hd_used_memory_tb: 69.4G | hd_free_memory_tb: 243G
244
+ [default0]:07/02/2024 18:43:07 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 11617.88MiB. Peak reserved: 44042.00MiB
245
+ [default0]:07/02/2024 18:43:14 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.59MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
246
+ [default0]:07/02/2024 18:43:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 2 / 20 | consumed_tokens: 8.39M | elapsed_time_per_iteration_ms: 8.8K | tokens_per_sec: 476K | tokens_per_sec_per_gpu: 29.8K | global_batch_size: 1.02K | lm_loss: 11.3 | lr: 9.53e-05 | model_tflops_per_gpu: 270 | hardware_tflops_per_gpu: 270 | grad_norm: 33.3 | cuda_memory_allocated: 7.6G | cuda_max_memory_reserved: 46.2G | hd_total_memory_tb: 312G | hd_used_memory_tb: 69.4G | hd_free_memory_tb: 243G
247
+ [default0]:07/02/2024 18:43:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 11617.89MiB. Peak reserved: 44042.00MiB
248
+ [default0]:07/02/2024 18:43:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.59MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
249
+ [default0]:STAGE:2024-07-02 18:43:24 3711718:3711718 ActivityProfilerController.cpp:314] Completed Stage: Warm Up
250
+ [default0]:07/02/2024 18:43:24 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 3 / 20 | consumed_tokens: 12.6M | elapsed_time_per_iteration_ms: 8.77K | tokens_per_sec: 478K | tokens_per_sec_per_gpu: 29.9K | global_batch_size: 1.02K | lm_loss: 16 | lr: 9.05e-05 | model_tflops_per_gpu: 271 | hardware_tflops_per_gpu: 271 | grad_norm: 249 | cuda_memory_allocated: 7.6G | cuda_max_memory_reserved: 46.2G | hd_total_memory_tb: 312G | hd_used_memory_tb: 69.4G | hd_free_memory_tb: 243G
251
+ [default0]:07/02/2024 18:43:24 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 11617.89MiB. Peak reserved: 44042.00MiB
252
+ [default0]:07/02/2024 18:43:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.59MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
253
+ [default0]:07/02/2024 18:43:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 4 / 20 | consumed_tokens: 16.8M | elapsed_time_per_iteration_ms: 8.87K | tokens_per_sec: 473K | tokens_per_sec_per_gpu: 29.6K | global_batch_size: 1.02K | lm_loss: 15.1 | lr: 8.58e-05 | model_tflops_per_gpu: 268 | hardware_tflops_per_gpu: 268 | grad_norm: 41.6 | cuda_memory_allocated: 7.6G | cuda_max_memory_reserved: 46.2G | hd_total_memory_tb: 312G | hd_used_memory_tb: 69.4G | hd_free_memory_tb: 243G
254
+ [default0]:07/02/2024 18:43:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 11617.89MiB. Peak reserved: 44042.00MiB
255
+ [default0]:07/02/2024 18:43:42 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 5 / 20 | consumed_tokens: 21M | elapsed_time_per_iteration_ms: 8.87K | tokens_per_sec: 473K | tokens_per_sec_per_gpu: 29.6K | global_batch_size: 1.02K | lm_loss: 10.8 | lr: 8.11e-05 | model_tflops_per_gpu: 268 | hardware_tflops_per_gpu: 268 | grad_norm: 26
256
+ [default0]:07/02/2024 18:43:42 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
257
+ [default0]:07/02/2024 18:43:51 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 6 / 20 | consumed_tokens: 25.2M | elapsed_time_per_iteration_ms: 8.82K | tokens_per_sec: 475K | tokens_per_sec_per_gpu: 29.7K | global_batch_size: 1.02K | lm_loss: 10.8 | lr: 7.63e-05 | model_tflops_per_gpu: 270 | hardware_tflops_per_gpu: 270 | grad_norm: 18.9
258
+ [default0]:STAGE:2024-07-02 18:44:01 3711718:3711718 ActivityProfilerController.cpp:320] Completed Stage: Collection
259
+ [default0]:STAGE:2024-07-02 18:44:01 3711718:3711718 ActivityProfilerController.cpp:324] Completed Stage: Post Processing
260
+ [default0]:07/02/2024 18:45:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
261
+ [default0]:07/02/2024 18:45:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 7 / 20 | consumed_tokens: 29.4M | elapsed_time_per_iteration_ms: 8.5K | tokens_per_sec: 494K | tokens_per_sec_per_gpu: 30.8K | global_batch_size: 1.02K | lm_loss: 10.2 | lr: 7.16e-05 | model_tflops_per_gpu: 280 | hardware_tflops_per_gpu: 280 | grad_norm: 7.97
262
+ [default0]:07/02/2024 18:45:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
263
+ [default0]:07/02/2024 18:45:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 8 / 20 | consumed_tokens: 33.6M | elapsed_time_per_iteration_ms: 8.58K | tokens_per_sec: 489K | tokens_per_sec_per_gpu: 30.5K | global_batch_size: 1.02K | lm_loss: 9.15 | lr: 6.68e-05 | model_tflops_per_gpu: 277 | hardware_tflops_per_gpu: 277 | grad_norm: 6.46
264
+ [default0]:07/02/2024 18:45:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
265
+ [default0]:07/02/2024 18:45:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 9 / 20 | consumed_tokens: 37.7M | elapsed_time_per_iteration_ms: 8.65K | tokens_per_sec: 485K | tokens_per_sec_per_gpu: 30.3K | global_batch_size: 1.02K | lm_loss: 11.2 | lr: 6.21e-05 | model_tflops_per_gpu: 275 | hardware_tflops_per_gpu: 275 | grad_norm: 59.8
266
+ [default0]:07/02/2024 18:45:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
267
+ [default0]:07/02/2024 18:45:44 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 10 / 20 | consumed_tokens: 41.9M | elapsed_time_per_iteration_ms: 8.79K | tokens_per_sec: 477K | tokens_per_sec_per_gpu: 29.8K | global_batch_size: 1.02K | lm_loss: 9.6 | lr: 5.74e-05 | model_tflops_per_gpu: 271 | hardware_tflops_per_gpu: 271 | grad_norm: 44.2
268
+ [default0]:07/02/2024 18:45:44 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
269
+ [default0]:07/02/2024 18:45:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 11 / 20 | consumed_tokens: 46.1M | elapsed_time_per_iteration_ms: 8.6K | tokens_per_sec: 488K | tokens_per_sec_per_gpu: 30.5K | global_batch_size: 1.02K | lm_loss: 8.08 | lr: 5.26e-05 | model_tflops_per_gpu: 277 | hardware_tflops_per_gpu: 277 | grad_norm: 8.6
270
+ [default0]:07/02/2024 18:45:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
271
+ [default0]:07/02/2024 18:46:02 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 12 / 20 | consumed_tokens: 50.3M | elapsed_time_per_iteration_ms: 8.66K | tokens_per_sec: 484K | tokens_per_sec_per_gpu: 30.3K | global_batch_size: 1.02K | lm_loss: 7.86 | lr: 4.79e-05 | model_tflops_per_gpu: 275 | hardware_tflops_per_gpu: 275 | grad_norm: 5.09
272
+ [default0]:07/02/2024 18:46:02 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
273
+ [default0]:07/02/2024 18:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 13 / 20 | consumed_tokens: 54.5M | elapsed_time_per_iteration_ms: 8.64K | tokens_per_sec: 485K | tokens_per_sec_per_gpu: 30.3K | global_batch_size: 1.02K | lm_loss: 7.7 | lr: 4.32e-05 | model_tflops_per_gpu: 275 | hardware_tflops_per_gpu: 275 | grad_norm: 4.73
274
+ [default0]:07/02/2024 18:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
275
+ [default0]:07/02/2024 18:46:19 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 14 / 20 | consumed_tokens: 58.7M | elapsed_time_per_iteration_ms: 8.82K | tokens_per_sec: 476K | tokens_per_sec_per_gpu: 29.7K | global_batch_size: 1.02K | lm_loss: 7.56 | lr: 3.84e-05 | model_tflops_per_gpu: 270 | hardware_tflops_per_gpu: 270 | grad_norm: 5.1
276
+ [default0]:07/02/2024 18:46:19 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
277
+ [default0]:07/02/2024 18:46:28 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 15 / 20 | consumed_tokens: 62.9M | elapsed_time_per_iteration_ms: 8.68K | tokens_per_sec: 483K | tokens_per_sec_per_gpu: 30.2K | global_batch_size: 1.02K | lm_loss: 7.4 | lr: 3.37e-05 | model_tflops_per_gpu: 274 | hardware_tflops_per_gpu: 274 | grad_norm: 5.17
278
+ [default0]:07/02/2024 18:46:28 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
279
+ [default0]:07/02/2024 18:46:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 16 / 20 | consumed_tokens: 67.1M | elapsed_time_per_iteration_ms: 8.66K | tokens_per_sec: 485K | tokens_per_sec_per_gpu: 30.3K | global_batch_size: 1.02K | lm_loss: 7.3 | lr: 2.89e-05 | model_tflops_per_gpu: 275 | hardware_tflops_per_gpu: 275 | grad_norm: 5.17
280
+ [default0]:07/02/2024 18:46:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
281
+ [default0]:07/02/2024 18:46:45 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 17 / 20 | consumed_tokens: 71.3M | elapsed_time_per_iteration_ms: 8.73K | tokens_per_sec: 480K | tokens_per_sec_per_gpu: 30K | global_batch_size: 1.02K | lm_loss: 7.22 | lr: 2.42e-05 | model_tflops_per_gpu: 272 | hardware_tflops_per_gpu: 272 | grad_norm: 5.13
282
+ [default0]:07/02/2024 18:46:45 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
283
+ [default0]:07/02/2024 18:46:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 18 / 20 | consumed_tokens: 75.5M | elapsed_time_per_iteration_ms: 8.63K | tokens_per_sec: 486K | tokens_per_sec_per_gpu: 30.4K | global_batch_size: 1.02K | lm_loss: 7.15 | lr: 1.95e-05 | model_tflops_per_gpu: 276 | hardware_tflops_per_gpu: 276 | grad_norm: 5.04
284
+ [default0]:07/02/2024 18:46:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
285
+ [default0]:07/02/2024 18:47:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 19 / 20 | consumed_tokens: 79.7M | elapsed_time_per_iteration_ms: 8.71K | tokens_per_sec: 481K | tokens_per_sec_per_gpu: 30.1K | global_batch_size: 1.02K | lm_loss: 7.08 | lr: 1.47e-05 | model_tflops_per_gpu: 273 | hardware_tflops_per_gpu: 273 | grad_norm: 3.87
286
+ [default0]:07/02/2024 18:47:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: Memory usage: 7252.58MiB. Peak allocated 42900.39MiB. Peak reserved: 44042.00MiB
287
+ [default0]:07/02/2024 18:47:11 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-62]: iteration: 20 / 20 | consumed_tokens: 83.9M | elapsed_time_per_iteration_ms: 8.79K | tokens_per_sec: 477K | tokens_per_sec_per_gpu: 29.8K | global_batch_size: 1.02K | lm_loss: 7.03 | lr: 1e-05 | model_tflops_per_gpu: 271 | hardware_tflops_per_gpu: 271 | grad_norm: 2.93
288
+ Traceback (most recent call last):
289
+ File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py", line 4, in <module>
290
+ from bench_cluster.submit_jobs import submit_jobs, check_status
291
+ ImportError: cannot import name 'check_status' from 'bench_cluster.submit_jobs' (/fsx/ferdinandmom/ferdinand-hf/bench_cluster/bench_cluster/submit_jobs.py)
292
+ Traceback (most recent call last):
293
+ File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py", line 4, in <module>
294
+ from bench_cluster.submit_jobs import submit_jobs, check_status
295
+ ImportError: cannot import name 'check_status' from 'bench_cluster.submit_jobs' (/fsx/ferdinandmom/ferdinand-hf/bench_cluster/bench_cluster/submit_jobs.py)
296
+ 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.
297
+
llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/profiler/ip-26-0-171-62_3711718.1719945894658523576.pt.trace.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:11290d86efff0c9e92bb027c45555b4b8923ab33d0650fd87673f089348c52e4
3
+ size 2207745511
llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-4/status.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ completed