Hugging Face's logo

nanotron
/
old_bench

Model card Files Community
3outeille HF staff commited on
Commit
11d8243
•
1 Parent(s): be7ea13

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

Browse files
Files changed (8) hide show
  1. .gitattributes +1 -0
  2. llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/bench.slurm +111 -0
  3. llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/config.yaml +90 -0
  4. llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/log.out +340 -0
  5. llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/log_metrics.csv +21 -0
  6. llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/profiler.csv +2 -0
  7. llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/profiler/ip-26-0-170-31_2724547.1719930375401132771.pt.trace.json +3 -0
  8. llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/status.txt +1 -0
.gitattributes CHANGED
@@ -34,3 +34,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
  llama-1B/16_GPUS/dp-4_tp-4_pp-1_mbz-16/profiler/ip-26-0-161-178_137134.1719929907506010984.pt.trace.json filter=lfs diff=lfs merge=lfs -text
 
 
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
  llama-1B/16_GPUS/dp-4_tp-4_pp-1_mbz-16/profiler/ip-26-0-161-178_137134.1719929907506010984.pt.trace.json filter=lfs diff=lfs merge=lfs -text
37
+ llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/profiler/ip-26-0-170-31_2724547.1719930375401132771.pt.trace.json filter=lfs diff=lfs merge=lfs -text
llama-1B/16_GPUS/dp-1_tp-2_pp-8_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-1_tp-2_pp-8_mbz-8/log.out
13
+ #SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_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-1_tp-2_pp-8_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-1_tp-2_pp-8_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-1_tp-2_pp-8_mbz-8/status.txt
84
+ else
85
+ if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/log.out; then
86
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_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-1_tp-2_pp-8_mbz-8/log.out; then
88
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/status.txt
89
+ elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/log.out; then
90
+ printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/status.txt
91
+ else
92
+ printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_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-1_tp-2_pp-8_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-1_tp-2_pp-8_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-1_tp-2_pp-8_mbz-8 llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8 --commit-message "Upload llama-1B/16_GPUS/dp-1_tp-2_pp-8_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-1_tp-2_pp-8_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: 1
49
+ expert_parallel_size: 1
50
+ pp: 8
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-1_tp-2_pp-8_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: 128
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-1_tp-2_pp-8_mbz-8/log.out ADDED
@@ -0,0 +1,340 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ========================
2
+ START TIME: Tue Jul 2 14:17:32 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 14:17:34.530000 140419512616768 torch/distributed/run.py:757]
18
+ W0702 14:17:34.530000 140419512616768 torch/distributed/run.py:757] *****************************************
19
+ W0702 14:17:34.530000 140419512616768 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 14:17:34.530000 140419512616768 torch/distributed/run.py:757] *****************************************
21
+ W0702 14:17:34.535000 140122487379776 torch/distributed/run.py:757]
22
+ W0702 14:17:34.535000 140122487379776 torch/distributed/run.py:757] *****************************************
23
+ W0702 14:17:34.535000 140122487379776 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 14:17:34.535000 140122487379776 torch/distributed/run.py:757] *****************************************
25
+ [default0]:07/02/2024 14:17:53 [WARNING|DP=0|PP=0|TP=0|ip-26-0-170-31]: [Vocab Size Padding] Padded vocab (size: 50257) with 1 dummy tokens (new size: 50258)
26
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Config:
27
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Config(general=GeneralArgs(project='bench_cluster',
28
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: run='%date_%jobid',
29
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: seed=42,
30
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: step=None,
31
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: consumed_train_samples=None,
32
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: benchmark_csv_path=None,
33
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: ignore_sanity_checks=True),
34
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: parallelism=ParallelismArgs(dp=1,
35
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: pp=8,
36
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tp=2,
37
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f3ab5a30910>,
38
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
39
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tp_linear_async_communication=False,
40
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: expert_parallel_size=1),
41
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
42
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: eos_token_id=2,
43
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: hidden_act='silu',
44
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: hidden_size=2048,
45
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: initializer_range=0.02,
46
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: intermediate_size=4096,
47
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: is_llama_config=True,
48
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: max_position_embeddings=4096,
49
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: num_attention_heads=32,
50
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: num_hidden_layers=24,
51
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: num_key_value_heads=32,
52
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: pad_token_id=None,
53
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: pretraining_tp=1,
54
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: rms_norm_eps=1e-05,
55
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: rope_scaling=None,
56
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: rope_theta=10000.0,
57
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tie_word_embeddings=True,
58
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: use_cache=True,
59
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: vocab_size=50258),
60
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: init_method=RandomInit(std=0.025),
61
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: dtype=torch.bfloat16,
62
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: make_vocab_size_divisible_by=1,
63
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: ddp_bucket_cap_mb=25),
64
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
65
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tokenizer_revision=None,
66
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tokenizer_max_length=None),
67
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
68
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: checkpoint_interval=100000,
69
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: save_initial_state=False,
70
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: resume_checkpoint_path=None,
71
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: checkpoints_path_is_shared_file_system=False),
72
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: logging=LoggingArgs(log_level='info',
73
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: log_level_replica='info',
74
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: iteration_step_info_interval=1),
75
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tokens=TokensArgs(sequence_length=4096,
76
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: train_steps=20,
77
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: micro_batch_size=8,
78
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: batch_accumulation_per_replica=128,
79
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: val_check_interval=-1,
80
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: limit_val_batches=0,
81
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: limit_test_batches=0),
82
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
83
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: adam_beta1=0.9,
84
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: adam_beta2=0.95,
85
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: torch_adam_is_fused=True,
86
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: name='adamW'),
87
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: zero_stage=1,
88
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: weight_decay=0.01,
89
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: clip_grad=1.0,
90
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: accumulate_grad_in_fp32=True,
91
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
92
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: lr_warmup_steps=1,
93
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: lr_warmup_style='linear',
94
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: lr_decay_style='linear',
95
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: lr_decay_steps=19,
96
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: lr_decay_starting_step=None,
97
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: min_decay_lr=1e-05)),
98
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: data_stages=[DatasetStageArgs(name='Training Stage',
99
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: start_training_step=1,
100
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
101
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: hf_dataset_splits='train',
102
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: hf_dataset_config_name=None,
103
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: dataset_processing_num_proc_per_process=64,
104
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: dataset_overwrite_cache=False,
105
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: text_column_name='text'),
106
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: seed=42,
107
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: num_loading_workers=32))],
108
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8')),
109
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: lighteval=None)
110
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Model Config:
111
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: LlamaConfig(bos_token_id=1,
112
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: eos_token_id=2,
113
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: hidden_act='silu',
114
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: hidden_size=2048,
115
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: initializer_range=0.02,
116
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: intermediate_size=4096,
117
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: is_llama_config=True,
118
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: max_position_embeddings=4096,
119
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: num_attention_heads=32,
120
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: num_hidden_layers=24,
121
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: num_key_value_heads=32,
122
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: pad_token_id=None,
123
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: pretraining_tp=1,
124
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: rms_norm_eps=1e-05,
125
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: rope_scaling=None,
126
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: rope_theta=10000.0,
127
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tie_word_embeddings=True,
128
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: use_cache=True,
129
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: vocab_size=50258)
130
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Building model..
131
+ [default0]:07/02/2024 14:17:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Setting PP block ranks...
132
+ [default5]:07/02/2024 14:18:09 [INFO|DP=0|PP=6|TP=1|ip-26-0-171-56]: Local number of parameters: 83.9M (160.03MiB)
133
+ [default0]:07/02/2024 14:18:09 [INFO|DP=0|PP=4|TP=0|ip-26-0-171-56]: Local number of parameters: 62.9M (120.02MiB)
134
+ [default0]:07/02/2024 14:18:09 [INFO|DP=0|PP=4|TP=0|ip-26-0-171-56]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
135
+ [default0]:07/02/2024 14:18:09 [INFO|DP=0|PP=4|TP=0|ip-26-0-171-56]: No checkpoint path provided.
136
+ [default5]:07/02/2024 14:18:09 [INFO|DP=0|PP=6|TP=1|ip-26-0-171-56]: [After model building] Memory usage: 164.04MiB. Peak allocated: 166.07MiB Peak reserved: 180.00MiB
137
+ [default5]:07/02/2024 14:18:09 [INFO|DP=0|PP=6|TP=1|ip-26-0-171-56]: No checkpoint path provided.
138
+ [default3]:07/02/2024 14:18:09 [INFO|DP=0|PP=5|TP=1|ip-26-0-171-56]: Local number of parameters: 62.9M (120.02MiB)
139
+ [default1]:07/02/2024 14:18:09 [INFO|DP=0|PP=4|TP=1|ip-26-0-171-56]: Local number of parameters: 62.9M (120.02MiB)
140
+ [default6]:07/02/2024 14:18:09 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: Local number of parameters: 51.5M (98.16MiB)
141
+ [default6]:07/02/2024 14:18:09 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: [After model building] Memory usage: 98.17MiB. Peak allocated: 98.18MiB Peak reserved: 102.00MiB
142
+ [default1]:07/02/2024 14:18:09 [INFO|DP=0|PP=4|TP=1|ip-26-0-171-56]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
143
+ [default1]:07/02/2024 14:18:09 [INFO|DP=0|PP=4|TP=1|ip-26-0-171-56]: No checkpoint path provided.
144
+ [default2]:07/02/2024 14:18:09 [INFO|DP=0|PP=5|TP=0|ip-26-0-171-56]: Local number of parameters: 62.9M (120.02MiB)
145
+ [default2]:07/02/2024 14:18:09 [INFO|DP=0|PP=5|TP=0|ip-26-0-171-56]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
146
+ [default3]:07/02/2024 14:18:09 [INFO|DP=0|PP=5|TP=1|ip-26-0-171-56]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
147
+ [default3]:07/02/2024 14:18:09 [INFO|DP=0|PP=5|TP=1|ip-26-0-171-56]: No checkpoint path provided.
148
+ [default6]:07/02/2024 14:18:09 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: No checkpoint path provided.
149
+ [default2]:07/02/2024 14:18:09 [INFO|DP=0|PP=5|TP=0|ip-26-0-171-56]: No checkpoint path provided.
150
+ [default4]:07/02/2024 14:18:09 [INFO|DP=0|PP=6|TP=0|ip-26-0-171-56]: Local number of parameters: 83.9M (160.03MiB)
151
+ [default4]:07/02/2024 14:18:09 [INFO|DP=0|PP=6|TP=0|ip-26-0-171-56]: [After model building] Memory usage: 164.04MiB. Peak allocated: 166.07MiB Peak reserved: 180.00MiB
152
+ [default4]:07/02/2024 14:18:09 [INFO|DP=0|PP=6|TP=0|ip-26-0-171-56]: No checkpoint path provided.
153
+ [default7]:07/02/2024 14:18:09 [INFO|DP=0|PP=7|TP=1|ip-26-0-171-56]: Local number of parameters: 51.5M (98.16MiB)
154
+ [default7]:07/02/2024 14:18:09 [INFO|DP=0|PP=7|TP=1|ip-26-0-171-56]: [After model building] Memory usage: 98.17MiB. Peak allocated: 98.18MiB Peak reserved: 102.00MiB
155
+ [default7]:07/02/2024 14:18:09 [INFO|DP=0|PP=7|TP=1|ip-26-0-171-56]: No checkpoint path provided.
156
+ [default0]:07/02/2024 14:18:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Total number of parameters: 1.21G (2313.02MiB)
157
+ [default0]:07/02/2024 14:18:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Local number of parameters: 135M (258.19MiB)
158
+ [default0]:07/02/2024 14:18:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [After model building] Memory usage: 262.20MiB. Peak allocated: 264.23MiB Peak reserved: 280.00MiB
159
+ [default0]:07/02/2024 14:18:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: No checkpoint path provided.
160
+ [default3]:07/02/2024 14:18:09 [INFO|DP=0|PP=1|TP=1|ip-26-0-170-31]: Local number of parameters: 62.9M (120.02MiB)
161
+ [default2]:07/02/2024 14:18:09 [INFO|DP=0|PP=1|TP=0|ip-26-0-170-31]: Local number of parameters: 62.9M (120.02MiB)
162
+ [default2]:07/02/2024 14:18:09 [INFO|DP=0|PP=1|TP=0|ip-26-0-170-31]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
163
+ [default0]:07/02/2024 14:18:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Parametrizing model parameters using StandardParametrizator
164
+ [default3]:07/02/2024 14:18:09 [INFO|DP=0|PP=1|TP=1|ip-26-0-170-31]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
165
+ [default2]:07/02/2024 14:18:09 [INFO|DP=0|PP=1|TP=0|ip-26-0-170-31]: No checkpoint path provided.
166
+ [default3]:07/02/2024 14:18:09 [INFO|DP=0|PP=1|TP=1|ip-26-0-170-31]: No checkpoint path provided.
167
+ [default6]:07/02/2024 14:18:09 [INFO|DP=0|PP=3|TP=0|ip-26-0-170-31]: Local number of parameters: 83.9M (160.03MiB)
168
+ [default1]:07/02/2024 14:18:09 [INFO|DP=0|PP=0|TP=1|ip-26-0-170-31]: Local number of parameters: 135M (258.19MiB)
169
+ [default1]:07/02/2024 14:18:09 [INFO|DP=0|PP=0|TP=1|ip-26-0-170-31]: [After model building] Memory usage: 262.20MiB. Peak allocated: 264.23MiB Peak reserved: 280.00MiB
170
+ [default6]:07/02/2024 14:18:09 [INFO|DP=0|PP=3|TP=0|ip-26-0-170-31]: [After model building] Memory usage: 164.04MiB. Peak allocated: 166.07MiB Peak reserved: 180.00MiB
171
+ [default6]:07/02/2024 14:18:09 [INFO|DP=0|PP=3|TP=0|ip-26-0-170-31]: No checkpoint path provided.
172
+ [default5]:07/02/2024 14:18:09 [INFO|DP=0|PP=2|TP=1|ip-26-0-170-31]: Local number of parameters: 62.9M (120.02MiB)
173
+ [default5]:07/02/2024 14:18:09 [INFO|DP=0|PP=2|TP=1|ip-26-0-170-31]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
174
+ [default5]:07/02/2024 14:18:09 [INFO|DP=0|PP=2|TP=1|ip-26-0-170-31]: No checkpoint path provided.
175
+ [default1]:07/02/2024 14:18:09 [INFO|DP=0|PP=0|TP=1|ip-26-0-170-31]: No checkpoint path provided.
176
+ [default4]:07/02/2024 14:18:09 [INFO|DP=0|PP=2|TP=0|ip-26-0-170-31]: Local number of parameters: 62.9M (120.02MiB)
177
+ [default4]:07/02/2024 14:18:09 [INFO|DP=0|PP=2|TP=0|ip-26-0-170-31]: [After model building] Memory usage: 123.03MiB. Peak allocated: 125.06MiB Peak reserved: 138.00MiB
178
+ [default4]:07/02/2024 14:18:09 [INFO|DP=0|PP=2|TP=0|ip-26-0-170-31]: No checkpoint path provided.
179
+ [default7]:07/02/2024 14:18:09 [INFO|DP=0|PP=3|TP=1|ip-26-0-170-31]: Local number of parameters: 83.9M (160.03MiB)
180
+ [default7]:07/02/2024 14:18:09 [INFO|DP=0|PP=3|TP=1|ip-26-0-170-31]: [After model building] Memory usage: 164.04MiB. Peak allocated: 166.07MiB Peak reserved: 180.00MiB
181
+ [default7]:07/02/2024 14:18:09 [INFO|DP=0|PP=3|TP=1|ip-26-0-170-31]: No checkpoint path provided.
182
+ [default0]:07/02/2024 14:18:11 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [Optimizer Building] Using LearningRateForSP as learning rate
183
+ [default0]:07/02/2024 14:18:11 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [ZeRO sharding] Size of optimizer params per rank:
184
+ [default0]:07/02/2024 14:18:11 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [ZeRO sharding] DP Rank 0 has 135M out of 135M (100.00%) params' optimizer states
185
+ [default0]:07/02/2024 14:18:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
186
+ [default0]:07/02/2024 14:18:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Using `datasets` library
187
+ [default0]:07/02/2024 14:18:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
188
+ [default0]:07/02/2024 14:18:12 [WARNING|DP=0|PP=0|TP=0|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
189
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
190
+ [default0]:07/02/2024 14:18:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [Training Plan] There are 1 training stages
191
+ [default0]:07/02/2024 14:18:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [Stage Training Stage] start from step 1
192
+ [default0]:07/02/2024 14:18:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]:
193
+ [default0]:07/02/2024 14:18:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [Start training] datetime: 2024-07-02 14:18:12.970630 | mbs: 8 | grad_accum: 128 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
194
+ [default0]:07/02/2024 14:18:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
195
+ [default0]:07/02/2024 14:18:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 1294.97MiB. Peak allocated 1294.97MiB. Peak reserved: 1316.00MiB
196
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
197
+ [default1]:07/02/2024 14:18:13 [WARNING|DP=0|PP=4|TP=1|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
198
+ [default3]:07/02/2024 14:18:13 [WARNING|DP=0|PP=5|TP=1|ip-26-0-171-56]: 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
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
201
+ [default4]:07/02/2024 14:18:13 [WARNING|DP=0|PP=6|TP=0|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
202
+ [default0]:07/02/2024 14:18:13 [WARNING|DP=0|PP=4|TP=0|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
203
+ [default6]:07/02/2024 14:18:13 [WARNING|DP=0|PP=7|TP=0|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
204
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
205
+ [default5]:07/02/2024 14:18:13 [WARNING|DP=0|PP=6|TP=1|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
206
+ [default2]:07/02/2024 14:18:13 [WARNING|DP=0|PP=5|TP=0|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
207
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
208
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
209
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
210
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
211
+ [default2]:07/02/2024 14:18:13 [WARNING|DP=0|PP=1|TP=0|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
212
+ [default3]:07/02/2024 14:18:13 [WARNING|DP=0|PP=1|TP=1|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
213
+ [default7]:07/02/2024 14:18:13 [WARNING|DP=0|PP=3|TP=1|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
214
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
215
+ [default6]:07/02/2024 14:18:13 [WARNING|DP=0|PP=3|TP=0|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
216
+ [default1]:07/02/2024 14:18:13 [WARNING|DP=0|PP=0|TP=1|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
217
+ [default4]:07/02/2024 14:18:13 [WARNING|DP=0|PP=2|TP=0|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
218
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
219
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
220
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
221
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
222
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
223
+ [default7]:07/02/2024 14:18:13 [WARNING|DP=0|PP=7|TP=1|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
224
+ [default5]:07/02/2024 14:18:13 [WARNING|DP=0|PP=2|TP=1|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
225
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
226
+ [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.)
227
+ [default7]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
228
+ [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.)
229
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
230
+ [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.)
231
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
232
+ [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.)
233
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
234
+ [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.)
235
+ [default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
236
+ [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.)
237
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
238
+ [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.)
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]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
242
+ [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.)
243
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
244
+ [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.)
245
+ [default7]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
246
+ [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.)
247
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
248
+ [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.)
249
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
250
+ [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.)
251
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
252
+ [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.)
253
+ [default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
254
+ [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.)
255
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
256
+ [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.)
257
+ [default1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
258
+ [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
259
+ [default6]: warnings.warn(
260
+ [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
261
+ [default7]: warnings.warn(
262
+ [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
263
+ [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
264
+ [default4]: warnings.warn(
265
+ [default5]: 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
+ [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
269
+ [default2]: warnings.warn(
270
+ [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
271
+ [default1]: warnings.warn(
272
+ [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
273
+ [default0]: warnings.warn(
274
+ [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
275
+ [default6]: warnings.warn(
276
+ [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
277
+ [default4]: warnings.warn(
278
+ [default0]:07/02/2024 14:19:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 1361.25MiB. Peak allocated 44767.58MiB. Peak reserved: 45090.00MiB
279
+ [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
280
+ [default7]: warnings.warn(
281
+ [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
282
+ [default1]: warnings.warn(
283
+ [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
284
+ [default0]: warnings.warn(
285
+ [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
286
+ [default5]: warnings.warn(
287
+ [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
288
+ [default2]: warnings.warn(
289
+ [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
290
+ [default3]: warnings.warn(
291
+ [default6]:07/02/2024 14:19:59 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 1 / 20 | consumed_tokens: 4.19M | elapsed_time_per_iteration_ms: 105K | tokens_per_sec: 40K | tokens_per_sec_per_gpu: 2.5K | global_batch_size: 1.02K | lm_loss: 11.2 | lr: 0.0001 | model_tflops_per_gpu: 22.7 | hardware_tflops_per_gpu: 22.7 | grad_norm: 17.8 | cuda_memory_allocated: 994M | cuda_max_memory_reserved: 9.15G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.6G | hd_free_memory_tb: 247G
292
+ [default0]:07/02/2024 14:19:59 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 2394.03MiB. Peak reserved: 45090.00MiB
293
+ [default0]:07/02/2024 14:20:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46370.00MiB
294
+ [default6]:07/02/2024 14:20:57 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 2 / 20 | consumed_tokens: 8.39M | elapsed_time_per_iteration_ms: 58.2K | tokens_per_sec: 72.1K | tokens_per_sec_per_gpu: 4.5K | global_batch_size: 1.02K | lm_loss: 11.2 | lr: 9.53e-05 | model_tflops_per_gpu: 40.9 | hardware_tflops_per_gpu: 40.9 | grad_norm: 17.8 | cuda_memory_allocated: 994M | cuda_max_memory_reserved: 9.15G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.6G | hd_free_memory_tb: 247G
295
+ [default0]:07/02/2024 14:20:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 2394.04MiB. Peak reserved: 46370.00MiB
296
+ [default0]:07/02/2024 14:21:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
297
+ [default6]:07/02/2024 14:21:54 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 3 / 20 | consumed_tokens: 12.6M | elapsed_time_per_iteration_ms: 56.5K | tokens_per_sec: 74.3K | tokens_per_sec_per_gpu: 4.64K | global_batch_size: 1.02K | lm_loss: 9.62 | lr: 9.05e-05 | model_tflops_per_gpu: 42.1 | hardware_tflops_per_gpu: 42.1 | grad_norm: 21.7 | cuda_memory_allocated: 994M | cuda_max_memory_reserved: 9.15G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.6G | hd_free_memory_tb: 247G
298
+ [default0]:STAGE:2024-07-02 14:21:54 2724547:2724547 ActivityProfilerController.cpp:314] Completed Stage: Warm Up
299
+ [default0]:07/02/2024 14:21:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 2394.04MiB. Peak reserved: 46498.00MiB
300
+ [default0]:07/02/2024 14:22:49 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
301
+ [default6]:07/02/2024 14:22:49 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 4 / 20 | consumed_tokens: 16.8M | elapsed_time_per_iteration_ms: 55.6K | tokens_per_sec: 75.5K | tokens_per_sec_per_gpu: 4.72K | global_batch_size: 1.02K | lm_loss: 10.4 | lr: 8.58e-05 | model_tflops_per_gpu: 42.8 | hardware_tflops_per_gpu: 42.8 | grad_norm: 45.6 | cuda_memory_allocated: 994M | cuda_max_memory_reserved: 9.15G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.6G | hd_free_memory_tb: 247G
302
+ [default0]:07/02/2024 14:22:49 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 2394.04MiB. Peak reserved: 46498.00MiB
303
+ [default0]:07/02/2024 14:23:44 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
304
+ [default6]:07/02/2024 14:23:44 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 5 / 20 | consumed_tokens: 21M | elapsed_time_per_iteration_ms: 54.6K | tokens_per_sec: 76.8K | tokens_per_sec_per_gpu: 4.8K | global_batch_size: 1.02K | lm_loss: 9.43 | lr: 8.11e-05 | model_tflops_per_gpu: 43.5 | hardware_tflops_per_gpu: 43.5 | grad_norm: 11.2
305
+ [default6]:07/02/2024 14:24:40 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 6 / 20 | consumed_tokens: 25.2M | elapsed_time_per_iteration_ms: 55.9K | tokens_per_sec: 75.1K | tokens_per_sec_per_gpu: 4.69K | global_batch_size: 1.02K | lm_loss: 9.37 | lr: 7.63e-05 | model_tflops_per_gpu: 42.6 | hardware_tflops_per_gpu: 42.6 | grad_norm: 7.68
306
+ [default0]:STAGE:2024-07-02 14:24:53 2724547:2724547 ActivityProfilerController.cpp:320] Completed Stage: Collection
307
+ [default0]:STAGE:2024-07-02 14:24:55 2724547:2724547 ActivityProfilerController.cpp:324] Completed Stage: Post Processing
308
+ [default0]:07/02/2024 14:26:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
309
+ [default6]:07/02/2024 14:27:34 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 7 / 20 | consumed_tokens: 29.4M | elapsed_time_per_iteration_ms: 174K | tokens_per_sec: 24K | tokens_per_sec_per_gpu: 1.5K | global_batch_size: 1.02K | lm_loss: 8.97 | lr: 7.16e-05 | model_tflops_per_gpu: 13.6 | hardware_tflops_per_gpu: 13.6 | grad_norm: 5.69
310
+ [default0]:07/02/2024 14:27:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
311
+ [default0]:07/02/2024 14:28:30 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
312
+ [default6]:07/02/2024 14:28:30 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 8 / 20 | consumed_tokens: 33.6M | elapsed_time_per_iteration_ms: 55.6K | tokens_per_sec: 75.4K | tokens_per_sec_per_gpu: 4.71K | global_batch_size: 1.02K | lm_loss: 8.47 | lr: 6.68e-05 | model_tflops_per_gpu: 42.7 | hardware_tflops_per_gpu: 42.7 | grad_norm: 5.25
313
+ [default0]:07/02/2024 14:29:24 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
314
+ [default6]:07/02/2024 14:29:24 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 9 / 20 | consumed_tokens: 37.7M | elapsed_time_per_iteration_ms: 54.2K | tokens_per_sec: 77.4K | tokens_per_sec_per_gpu: 4.84K | global_batch_size: 1.02K | lm_loss: 8.01 | lr: 6.21e-05 | model_tflops_per_gpu: 43.9 | hardware_tflops_per_gpu: 43.9 | grad_norm: 4.65
315
+ [default0]:07/02/2024 14:30:19 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
316
+ [default6]:07/02/2024 14:30:19 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 10 / 20 | consumed_tokens: 41.9M | elapsed_time_per_iteration_ms: 55K | tokens_per_sec: 76.3K | tokens_per_sec_per_gpu: 4.77K | global_batch_size: 1.02K | lm_loss: 7.75 | lr: 5.74e-05 | model_tflops_per_gpu: 43.2 | hardware_tflops_per_gpu: 43.2 | grad_norm: 3.83
317
+ [default0]:07/02/2024 14:31:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
318
+ [default6]:07/02/2024 14:31:16 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 11 / 20 | consumed_tokens: 46.1M | elapsed_time_per_iteration_ms: 57.1K | tokens_per_sec: 73.5K | tokens_per_sec_per_gpu: 4.59K | global_batch_size: 1.02K | lm_loss: 7.62 | lr: 5.26e-05 | model_tflops_per_gpu: 41.7 | hardware_tflops_per_gpu: 41.7 | grad_norm: 4.98
319
+ [default0]:07/02/2024 14:32:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
320
+ [default6]:07/02/2024 14:32:10 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 12 / 20 | consumed_tokens: 50.3M | elapsed_time_per_iteration_ms: 53.8K | tokens_per_sec: 77.9K | tokens_per_sec_per_gpu: 4.87K | global_batch_size: 1.02K | lm_loss: 7.46 | lr: 4.79e-05 | model_tflops_per_gpu: 44.2 | hardware_tflops_per_gpu: 44.2 | grad_norm: 3.44
321
+ [default0]:07/02/2024 14:33:04 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
322
+ [default6]:07/02/2024 14:33:04 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 13 / 20 | consumed_tokens: 54.5M | elapsed_time_per_iteration_ms: 53.9K | tokens_per_sec: 77.9K | tokens_per_sec_per_gpu: 4.87K | global_batch_size: 1.02K | lm_loss: 7.34 | lr: 4.32e-05 | model_tflops_per_gpu: 44.1 | hardware_tflops_per_gpu: 44.1 | grad_norm: 3.45
323
+ [default0]:07/02/2024 14:33:58 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
324
+ [default6]:07/02/2024 14:33:58 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 14 / 20 | consumed_tokens: 58.7M | elapsed_time_per_iteration_ms: 54.7K | tokens_per_sec: 76.7K | tokens_per_sec_per_gpu: 4.8K | global_batch_size: 1.02K | lm_loss: 7.22 | lr: 3.84e-05 | model_tflops_per_gpu: 43.5 | hardware_tflops_per_gpu: 43.5 | grad_norm: 3.25
325
+ [default6]:07/02/2024 14:34:52 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 15 / 20 | consumed_tokens: 62.9M | elapsed_time_per_iteration_ms: 53.9K | tokens_per_sec: 77.8K | tokens_per_sec_per_gpu: 4.86K | global_batch_size: 1.02K | lm_loss: 7.1 | lr: 3.37e-05 | model_tflops_per_gpu: 44.1 | hardware_tflops_per_gpu: 44.1 | grad_norm: 2.89
326
+ [default0]:07/02/2024 14:34:52 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
327
+ [default6]:07/02/2024 14:35:46 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 16 / 20 | consumed_tokens: 67.1M | elapsed_time_per_iteration_ms: 53.2K | tokens_per_sec: 78.9K | tokens_per_sec_per_gpu: 4.93K | global_batch_size: 1.02K | lm_loss: 7.01 | lr: 2.89e-05 | model_tflops_per_gpu: 44.7 | hardware_tflops_per_gpu: 44.7 | grad_norm: 2.58
328
+ [default0]:07/02/2024 14:35:46 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
329
+ [default6]:07/02/2024 14:36:38 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 17 / 20 | consumed_tokens: 71.3M | elapsed_time_per_iteration_ms: 52.9K | tokens_per_sec: 79.3K | tokens_per_sec_per_gpu: 4.96K | global_batch_size: 1.02K | lm_loss: 6.94 | lr: 2.42e-05 | model_tflops_per_gpu: 45 | hardware_tflops_per_gpu: 45 | grad_norm: 2.48
330
+ [default0]:07/02/2024 14:36:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
331
+ [default6]:07/02/2024 14:37:33 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 18 / 20 | consumed_tokens: 75.5M | elapsed_time_per_iteration_ms: 55.1K | tokens_per_sec: 76.1K | tokens_per_sec_per_gpu: 4.76K | global_batch_size: 1.02K | lm_loss: 6.88 | lr: 1.95e-05 | model_tflops_per_gpu: 43.2 | hardware_tflops_per_gpu: 43.2 | grad_norm: 2.49
332
+ [default0]:07/02/2024 14:37:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
333
+ [default0]:07/02/2024 14:38:28 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 2394.03MiB. Peak allocated 45800.36MiB. Peak reserved: 46498.00MiB
334
+ [default6]:07/02/2024 14:38:28 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 19 / 20 | consumed_tokens: 79.7M | elapsed_time_per_iteration_ms: 54.5K | tokens_per_sec: 76.9K | tokens_per_sec_per_gpu: 4.81K | global_batch_size: 1.02K | lm_loss: 6.83 | lr: 1.47e-05 | model_tflops_per_gpu: 43.6 | hardware_tflops_per_gpu: 43.6 | grad_norm: 2.43
335
+ [default6]:07/02/2024 14:39:22 [INFO|DP=0|PP=7|TP=0|ip-26-0-171-56]: iteration: 20 / 20 | consumed_tokens: 83.9M | elapsed_time_per_iteration_ms: 53.7K | tokens_per_sec: 78.1K | tokens_per_sec_per_gpu: 4.88K | global_batch_size: 1.02K | lm_loss: 6.78 | lr: 1e-05 | model_tflops_per_gpu: 44.3 | hardware_tflops_per_gpu: 44.3 | grad_norm: 2.26
336
+ Saved 1 csv files over 1 completed logs
337
+ Processing file: /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/profiler/ip-26-0-170-31_2724547.1719930375401132771.pt.trace.json
338
+ Results written to /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/profiler.csv
339
+ 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.
340
+
llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/log_metrics.csv ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ iteration,consumed_tokens,elapsed_time_per_iteration_ms,tokens_per_sec,tokens_per_sec_per_gpu,global_batch_size,lm_loss,lr,model_tflops_per_gpu,hardware_tflops_per_gpu,grad_norm,memory_usage_MiB,peak_allocated_MiB,peak_reserved_MiB
2
+ 1,4190000.0000000005,105000.0,40000.0,2500.0,1020.0,11.2,0.0001,22.7,22.7,17.8,2394.03,45800.36,46370.0
3
+ 2,8390000.0,58200.0,72100.0,4500.0,1020.0,11.2,9.53e-05,40.9,40.9,17.8,2394.03,45800.36,46498.0
4
+ 3,12600000.0,56500.0,74300.0,4640.0,1020.0,9.62,9.05e-05,42.1,42.1,21.7,2394.03,45800.36,46498.0
5
+ 4,16800000.0,55600.0,75500.0,4720.0,1020.0,10.4,8.58e-05,42.8,42.8,45.6,2394.03,45800.36,46498.0
6
+ 5,21000000.0,54600.0,76800.0,4800.0,1020.0,9.43,8.11e-05,43.5,43.5,11.2,,,
7
+ 6,25200000.0,55900.0,75100.0,4690.0,1020.0,9.37,7.63e-05,42.6,42.6,7.68,2394.03,45800.36,46498.0
8
+ 7,29400000.0,174000.0,24000.0,1500.0,1020.0,8.97,7.16e-05,13.6,13.6,5.69,2394.03,45800.36,46498.0
9
+ 8,33600000.0,55600.0,75400.0,4710.0,1020.0,8.47,6.68e-05,42.7,42.7,5.25,2394.03,45800.36,46498.0
10
+ 9,37700000.0,54200.0,77400.0,4840.0,1020.0,8.01,6.21e-05,43.9,43.9,4.65,2394.03,45800.36,46498.0
11
+ 10,41900000.0,55000.0,76300.0,4770.0,1020.0,7.75,5.74e-05,43.2,43.2,3.83,2394.03,45800.36,46498.0
12
+ 11,46100000.0,57100.0,73500.0,4590.0,1020.0,7.62,5.26e-05,41.7,41.7,4.98,2394.03,45800.36,46498.0
13
+ 12,50300000.0,53800.0,77900.0,4870.0,1020.0,7.46,4.79e-05,44.2,44.2,3.44,2394.03,45800.36,46498.0
14
+ 13,54500000.0,53900.0,77900.0,4870.0,1020.0,7.34,4.32e-05,44.1,44.1,3.45,2394.03,45800.36,46498.0
15
+ 14,58700000.0,54700.0,76700.0,4800.0,1020.0,7.22,3.84e-05,43.5,43.5,3.25,,,
16
+ 15,62900000.0,53900.0,77800.0,4860.0,1020.0,7.1,3.37e-05,44.1,44.1,2.89,2394.03,45800.36,46498.0
17
+ 16,67099999.99999999,53200.0,78900.0,4930.0,1020.0,7.01,2.89e-05,44.7,44.7,2.58,2394.03,45800.36,46498.0
18
+ 17,71300000.0,52900.0,79300.0,4960.0,1020.0,6.94,2.42e-05,45.0,45.0,2.48,2394.03,45800.36,46498.0
19
+ 18,75500000.0,55100.0,76100.0,4760.0,1020.0,6.88,1.95e-05,43.2,43.2,2.49,2394.03,45800.36,46498.0
20
+ 19,79700000.0,54500.0,76900.0,4810.0,1020.0,6.83,1.47e-05,43.6,43.6,2.43,,,
21
+ 20,83900000.0,53700.0,78100.0,4880.0,1020.0,6.78,1e-05,44.3,44.3,2.26,,,
llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/profiler.csv ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ forward,backward
2
+ 0ms 980μs,7ms 495μs
llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/profiler/ip-26-0-170-31_2724547.1719930375401132771.pt.trace.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:54bdb04457a1cdfde8d1c05c8d2cf3105ded0de3a032e341e097c6ae7217f7ad
3
+ size 3220325852
llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/status.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ completed