Hugging Face's logo

nanotron
/
old_bench

Model card Files Community
3outeille HF staff commited on
Commit
ccb19f2
•
1 Parent(s): 2227859

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

Browse files
Files changed (6) hide show
  1. .gitattributes +1 -0
  2. llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-4/bench.slurm +111 -0
  3. llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-4/config.yaml +90 -0
  4. llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-4/log.out +312 -0
  5. llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-4/profiler/ip-26-0-163-147_663432.1719949200647344699.pt.trace.json +3 -0
  6. llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-4/status.txt +1 -0
.gitattributes CHANGED
@@ -59,3 +59,4 @@ llama-1B/16_GPUS/dp-1_tp-8_pp-2_mbz-32/profiler/ip-26-0-163-226_3030049.17199485
59
  llama-1B/16_GPUS/dp-2_tp-4_pp-2_mbz-16/profiler/ip-26-0-163-147_624442.1719948565344187651.pt.trace.json filter=lfs diff=lfs merge=lfs -text
60
  llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-1/profiler/ip-26-0-171-21_2559077.1719948801791737708.pt.trace.json filter=lfs diff=lfs merge=lfs -text
61
  llama-1B/16_GPUS/dp-2_tp-8_pp-1_mbz-16/profiler/ip-26-0-165-24_762231.1719949067497881377.pt.trace.json filter=lfs diff=lfs merge=lfs -text
 
 
59
  llama-1B/16_GPUS/dp-2_tp-4_pp-2_mbz-16/profiler/ip-26-0-163-147_624442.1719948565344187651.pt.trace.json filter=lfs diff=lfs merge=lfs -text
60
  llama-1B/16_GPUS/dp-16_tp-1_pp-1_mbz-1/profiler/ip-26-0-171-21_2559077.1719948801791737708.pt.trace.json filter=lfs diff=lfs merge=lfs -text
61
  llama-1B/16_GPUS/dp-2_tp-8_pp-1_mbz-16/profiler/ip-26-0-165-24_762231.1719949067497881377.pt.trace.json filter=lfs diff=lfs merge=lfs -text
62
+ llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-4/profiler/ip-26-0-163-147_663432.1719949200647344699.pt.trace.json filter=lfs diff=lfs merge=lfs -text
llama-1B/16_GPUS/dp-4_tp-1_pp-4_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-4_tp-1_pp-4_mbz-4/log.out
13
+ #SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_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-4_tp-1_pp-4_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-4_tp-1_pp-4_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-4_tp-1_pp-4_mbz-4/status.txt
84
+ else
85
+ if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-4/log.out; then
86
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_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-4_tp-1_pp-4_mbz-4/log.out; then
88
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-4/status.txt
89
+ elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-4/log.out; then
90
+ printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-4/status.txt
91
+ else
92
+ printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_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-4_tp-1_pp-4_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-4_tp-1_pp-4_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-4_tp-1_pp-4_mbz-4 llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-4 --commit-message "Upload llama-1B/16_GPUS/dp-4_tp-1_pp-4_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-4_tp-1_pp-4_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: 4
49
+ expert_parallel_size: 1
50
+ pp: 4
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-4_tp-1_pp-4_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: 64
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-4_tp-1_pp-4_mbz-4/log.out ADDED
@@ -0,0 +1,312 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ========================
2
+ START TIME: Tue Jul 2 19:35:55 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 19:35:57.633000 140372239103808 torch/distributed/run.py:757]
18
+ W0702 19:35:57.633000 140372239103808 torch/distributed/run.py:757] *****************************************
19
+ W0702 19:35:57.633000 140372239103808 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 19:35:57.633000 140372239103808 torch/distributed/run.py:757] *****************************************
21
+ W0702 19:35:57.637000 140056483870528 torch/distributed/run.py:757]
22
+ W0702 19:35:57.637000 140056483870528 torch/distributed/run.py:757] *****************************************
23
+ W0702 19:35:57.637000 140056483870528 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 19:35:57.637000 140056483870528 torch/distributed/run.py:757] *****************************************
25
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Config:
26
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Config(general=GeneralArgs(project='bench_cluster',
27
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: run='%date_%jobid',
28
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: seed=42,
29
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: step=None,
30
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: consumed_train_samples=None,
31
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: benchmark_csv_path=None,
32
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: ignore_sanity_checks=True),
33
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: parallelism=ParallelismArgs(dp=4,
34
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: pp=4,
35
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: tp=1,
36
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7fdf86128910>,
37
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
38
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: tp_linear_async_communication=False,
39
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: expert_parallel_size=1),
40
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
41
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: eos_token_id=2,
42
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: hidden_act='silu',
43
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: hidden_size=2048,
44
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: initializer_range=0.02,
45
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: intermediate_size=4096,
46
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: is_llama_config=True,
47
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: max_position_embeddings=4096,
48
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: num_attention_heads=32,
49
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: num_hidden_layers=24,
50
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: num_key_value_heads=32,
51
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: pad_token_id=None,
52
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: pretraining_tp=1,
53
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: rms_norm_eps=1e-05,
54
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: rope_scaling=None,
55
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: rope_theta=10000.0,
56
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: tie_word_embeddings=True,
57
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: use_cache=True,
58
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: vocab_size=50257),
59
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: init_method=RandomInit(std=0.025),
60
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: dtype=torch.bfloat16,
61
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: make_vocab_size_divisible_by=1,
62
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: ddp_bucket_cap_mb=25),
63
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
64
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: tokenizer_revision=None,
65
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: tokenizer_max_length=None),
66
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
67
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: checkpoint_interval=100000,
68
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: save_initial_state=False,
69
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: resume_checkpoint_path=None,
70
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: checkpoints_path_is_shared_file_system=False),
71
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: logging=LoggingArgs(log_level='info',
72
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: log_level_replica='info',
73
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: iteration_step_info_interval=1),
74
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: tokens=TokensArgs(sequence_length=4096,
75
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: train_steps=20,
76
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: micro_batch_size=4,
77
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: batch_accumulation_per_replica=64,
78
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: val_check_interval=-1,
79
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: limit_val_batches=0,
80
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: limit_test_batches=0),
81
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
82
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: adam_beta1=0.9,
83
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: adam_beta2=0.95,
84
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: torch_adam_is_fused=True,
85
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: name='adamW'),
86
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: zero_stage=1,
87
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: weight_decay=0.01,
88
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: clip_grad=1.0,
89
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: accumulate_grad_in_fp32=True,
90
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
91
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: lr_warmup_steps=1,
92
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: lr_warmup_style='linear',
93
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: lr_decay_style='linear',
94
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: lr_decay_steps=19,
95
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: lr_decay_starting_step=None,
96
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: min_decay_lr=1e-05)),
97
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: data_stages=[DatasetStageArgs(name='Training Stage',
98
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: start_training_step=1,
99
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
100
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: hf_dataset_splits='train',
101
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: hf_dataset_config_name=None,
102
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: dataset_processing_num_proc_per_process=64,
103
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: dataset_overwrite_cache=False,
104
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: text_column_name='text'),
105
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: seed=42,
106
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: num_loading_workers=32))],
107
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-4')),
108
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: lighteval=None)
109
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Model Config:
110
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: LlamaConfig(bos_token_id=1,
111
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: eos_token_id=2,
112
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: hidden_act='silu',
113
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: hidden_size=2048,
114
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: initializer_range=0.02,
115
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: intermediate_size=4096,
116
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: is_llama_config=True,
117
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: max_position_embeddings=4096,
118
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: num_attention_heads=32,
119
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: num_hidden_layers=24,
120
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: num_key_value_heads=32,
121
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: pad_token_id=None,
122
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: pretraining_tp=1,
123
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: rms_norm_eps=1e-05,
124
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: rope_scaling=None,
125
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: rope_theta=10000.0,
126
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: tie_word_embeddings=True,
127
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: use_cache=True,
128
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: vocab_size=50257)
129
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Building model..
130
+ [default0]:07/02/2024 19:36:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Setting PP block ranks...
131
+ [default4]:07/02/2024 19:36:26 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: Local number of parameters: 271M (516.35MiB)
132
+ [default4]:07/02/2024 19:36:26 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: [After model building] Memory usage: 520.36MiB. Peak allocated: 522.39MiB Peak reserved: 534.00MiB
133
+ [default4]:07/02/2024 19:36:26 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: No checkpoint path provided.
134
+ [default0]:07/02/2024 19:36:26 [INFO|DP=0|PP=2|TP=0|ip-26-0-163-226]: Local number of parameters: 252M (480.05MiB)
135
+ [default0]:07/02/2024 19:36:26 [INFO|DP=0|PP=2|TP=0|ip-26-0-163-226]: [After model building] Memory usage: 486.06MiB. Peak allocated: 488.09MiB Peak reserved: 502.00MiB
136
+ [default0]:07/02/2024 19:36:26 [INFO|DP=0|PP=2|TP=0|ip-26-0-163-226]: No checkpoint path provided.
137
+ [default0]:07/02/2024 19:36:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Total number of parameters: 1.21G (2312.82MiB)
138
+ [default0]:07/02/2024 19:36:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Local number of parameters: 397M (756.37MiB)
139
+ [default0]:07/02/2024 19:36:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: [After model building] Memory usage: 763.38MiB. Peak allocated: 765.41MiB Peak reserved: 792.00MiB
140
+ [default0]:07/02/2024 19:36:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: No checkpoint path provided.
141
+ [default0]:07/02/2024 19:36:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Parametrizing model parameters using StandardParametrizator
142
+ [default4]:07/02/2024 19:36:26 [INFO|DP=0|PP=1|TP=0|ip-26-0-163-147]: Local number of parameters: 294M (560.05MiB)
143
+ [default4]:07/02/2024 19:36:26 [INFO|DP=0|PP=1|TP=0|ip-26-0-163-147]: [After model building] Memory usage: 567.07MiB. Peak allocated: 569.10MiB Peak reserved: 594.00MiB
144
+ [default4]:07/02/2024 19:36:26 [INFO|DP=0|PP=1|TP=0|ip-26-0-163-147]: No checkpoint path provided.
145
+ [default3]:07/02/2024 19:36:26 [INFO|DP=3|PP=2|TP=0|ip-26-0-163-226]: No checkpoint path provided.
146
+ [default7]:07/02/2024 19:36:26 [INFO|DP=3|PP=3|TP=0|ip-26-0-163-226]: No checkpoint path provided.
147
+ [default7]:07/02/2024 19:36:26 [INFO|DP=3|PP=1|TP=0|ip-26-0-163-147]: No checkpoint path provided.
148
+ [default3]:07/02/2024 19:36:26 [INFO|DP=3|PP=0|TP=0|ip-26-0-163-147]: No checkpoint path provided.
149
+ [default1]:07/02/2024 19:36:26 [INFO|DP=1|PP=2|TP=0|ip-26-0-163-226]: No checkpoint path provided.
150
+ [default5]:07/02/2024 19:36:26 [INFO|DP=1|PP=3|TP=0|ip-26-0-163-226]: No checkpoint path provided.
151
+ [default2]:07/02/2024 19:36:26 [INFO|DP=2|PP=2|TP=0|ip-26-0-163-226]: No checkpoint path provided.
152
+ [default2]:07/02/2024 19:36:26 [INFO|DP=2|PP=0|TP=0|ip-26-0-163-147]: No checkpoint path provided.
153
+ [default1]:07/02/2024 19:36:26 [INFO|DP=1|PP=0|TP=0|ip-26-0-163-147]: No checkpoint path provided.
154
+ [default6]:07/02/2024 19:36:26 [INFO|DP=2|PP=3|TP=0|ip-26-0-163-226]: No checkpoint path provided.
155
+ [default6]:07/02/2024 19:36:26 [INFO|DP=2|PP=1|TP=0|ip-26-0-163-147]: No checkpoint path provided.
156
+ [default5]:07/02/2024 19:36:26 [INFO|DP=1|PP=1|TP=0|ip-26-0-163-147]: No checkpoint path provided.
157
+ [default0]:07/02/2024 19:36:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: [Optimizer Building] Using LearningRateForSP as learning rate
158
+ [default0]:07/02/2024 19:36:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: [ZeRO sharding] Size of optimizer params per rank:
159
+ [default0]:07/02/2024 19:36:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: [ZeRO sharding] DP Rank 0 has 99.1M out of 397M (25.00%) params' optimizer states
160
+ [default0]:07/02/2024 19:36:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: [ZeRO sharding] DP Rank 1 has 99.1M out of 397M (25.00%) params' optimizer states
161
+ [default0]:07/02/2024 19:36:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: [ZeRO sharding] DP Rank 2 has 99.1M out of 397M (25.00%) params' optimizer states
162
+ [default0]:07/02/2024 19:36:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: [ZeRO sharding] DP Rank 3 has 99.1M out of 397M (25.00%) params' optimizer states
163
+ [default0]:07/02/2024 19:36:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
164
+ [default0]:07/02/2024 19:36:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Using `datasets` library
165
+ [default0]:07/02/2024 19:36:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
166
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
167
+ [default0]:07/02/2024 19:36:32 [WARNING|DP=0|PP=0|TP=0|ip-26-0-163-147]: Repo card metadata block was not found. Setting CardData to empty.
168
+ [default0]:07/02/2024 19:36:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: [Training Plan] There are 1 training stages
169
+ [default0]:07/02/2024 19:36:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: [Stage Training Stage] start from step 1
170
+ [default0]:07/02/2024 19:36:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]:
171
+ [default0]:07/02/2024 19:36:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: [Start training] datetime: 2024-07-02 19:36:33.475408 | mbs: 4 | grad_accum: 64 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
172
+ [default0]:07/02/2024 19:36:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
173
+ [default0]:07/02/2024 19:36:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 2654.31MiB. Peak allocated 2654.31MiB. Peak reserved: 2686.00MiB
174
+ [default1]:07/02/2024 19:36:33 [WARNING|DP=1|PP=0|TP=0|ip-26-0-163-147]: Repo card metadata block was not found. Setting CardData to empty.
175
+ [default3]:07/02/2024 19:36:33 [WARNING|DP=3|PP=0|TP=0|ip-26-0-163-147]: Repo card metadata block was not found. Setting CardData to empty.
176
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
177
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
178
+ [default3]:07/02/2024 19:36:33 [WARNING|DP=3|PP=2|TP=0|ip-26-0-163-226]: Repo card metadata block was not found. Setting CardData to empty.
179
+ [default1]:07/02/2024 19:36:33 [WARNING|DP=1|PP=2|TP=0|ip-26-0-163-226]: Repo card metadata block was not found. Setting CardData to empty.
180
+ [default7]:07/02/2024 19:36:33 [WARNING|DP=3|PP=3|TP=0|ip-26-0-163-226]: Repo card metadata block was not found. Setting CardData to empty.
181
+ [default5]:07/02/2024 19:36:33 [WARNING|DP=1|PP=3|TP=0|ip-26-0-163-226]: Repo card metadata block was not found. Setting CardData to empty.
182
+ [default4]:07/02/2024 19:36:33 [WARNING|DP=0|PP=3|TP=0|ip-26-0-163-226]: Repo card metadata block was not found. Setting CardData to empty.
183
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
184
+ [default2]:07/02/2024 19:36:33 [WARNING|DP=2|PP=0|TP=0|ip-26-0-163-147]: Repo card metadata block was not found. Setting CardData to empty.
185
+ [default2]:07/02/2024 19:36:33 [WARNING|DP=2|PP=2|TP=0|ip-26-0-163-226]: Repo card metadata block was not found. Setting CardData to empty.
186
+ [default0]:07/02/2024 19:36:33 [WARNING|DP=0|PP=2|TP=0|ip-26-0-163-226]: Repo card metadata block was not found. Setting CardData to empty.
187
+ [default6]: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]:Repo card metadata block was not found. Setting CardData to empty.
190
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
191
+ [default6]:07/02/2024 19:36:33 [WARNING|DP=2|PP=3|TP=0|ip-26-0-163-226]: Repo card metadata block was not found. Setting CardData to empty.
192
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
193
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
194
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
195
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
196
+ [default5]:07/02/2024 19:36:33 [WARNING|DP=1|PP=1|TP=0|ip-26-0-163-147]: Repo card metadata block was not found. Setting CardData to empty.
197
+ [default6]:07/02/2024 19:36:33 [WARNING|DP=2|PP=1|TP=0|ip-26-0-163-147]: Repo card metadata block was not found. Setting CardData to empty.
198
+ [default7]:07/02/2024 19:36:33 [WARNING|DP=3|PP=1|TP=0|ip-26-0-163-147]: 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
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
201
+ [default4]:07/02/2024 19:36:33 [WARNING|DP=0|PP=1|TP=0|ip-26-0-163-147]: Repo card metadata block was not found. Setting CardData to empty.
202
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
203
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
204
+ [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.)
205
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
206
+ [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.)
207
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
208
+ [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.)
209
+ [default1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
210
+ [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.)
211
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
212
+ [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.)
213
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
214
+ [default0]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: Attempting to run cuBLAS, but there was no current CUDA context! Attempting to set the primary context... (Triggered internally at ../aten/src/ATen/cuda/CublasHandlePool.cpp:135.)
215
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
216
+ [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.)
217
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
218
+ [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.)
219
+ [default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
220
+ [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.)
221
+ [default7]: 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
+ [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.)
225
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
226
+ [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.)
227
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
228
+ [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.)
229
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
230
+ [default0]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: Attempting to run cuBLAS, but there was no current CUDA context! Attempting to set the primary context... (Triggered internally at ../aten/src/ATen/cuda/CublasHandlePool.cpp:135.)
231
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
232
+ [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.)
233
+ [default0]: 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
+ [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.)
237
+ [default7]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
238
+ [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.)
239
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
240
+ [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
241
+ [default5]: warnings.warn(
242
+ [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
243
+ [default1]: warnings.warn(
244
+ [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
245
+ [default4]: warnings.warn(
246
+ [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
247
+ [default0]: warnings.warn(
248
+ [default0]:07/02/2024 19:37:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 2721.95MiB. Peak allocated 37165.88MiB. Peak reserved: 37434.00MiB
249
+ [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
250
+ [default6]: warnings.warn(
251
+ [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
252
+ [default2]: warnings.warn(
253
+ [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
254
+ [default7]: warnings.warn(
255
+ [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
256
+ [default3]: warnings.warn(
257
+ [default4]:07/02/2024 19:37:13 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 1 / 20 | consumed_tokens: 4.19M | elapsed_time_per_iteration_ms: 38.8K | tokens_per_sec: 108K | tokens_per_sec_per_gpu: 6.75K | global_batch_size: 1.02K | lm_loss: 11.1 | lr: 0.0001 | model_tflops_per_gpu: 61.3 | hardware_tflops_per_gpu: 61.3 | grad_norm: 25.1 | cuda_memory_allocated: 2.51G | cuda_max_memory_reserved: 15.5G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.8G | hd_free_memory_tb: 246G
258
+ [default0]:07/02/2024 19:37:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 5180.18MiB. Peak reserved: 38192.00MiB
259
+ [default0]:07/02/2024 19:37:30 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
260
+ [default4]:07/02/2024 19:37:32 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 2 / 20 | consumed_tokens: 8.39M | elapsed_time_per_iteration_ms: 19.1K | tokens_per_sec: 219K | tokens_per_sec_per_gpu: 13.7K | global_batch_size: 1.02K | lm_loss: 11.1 | lr: 9.53e-05 | model_tflops_per_gpu: 124 | hardware_tflops_per_gpu: 124 | grad_norm: 25.2 | cuda_memory_allocated: 2.51G | cuda_max_memory_reserved: 15.5G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.8G | hd_free_memory_tb: 246G
261
+ [default0]:07/02/2024 19:37:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 5180.18MiB. Peak reserved: 38320.00MiB
262
+ [default0]:07/02/2024 19:37:49 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
263
+ [default0]:STAGE:2024-07-02 19:37:51 663432:663432 ActivityProfilerController.cpp:314] Completed Stage: Warm Up
264
+ [default4]:07/02/2024 19:37:51 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 3 / 20 | consumed_tokens: 12.6M | elapsed_time_per_iteration_ms: 19K | tokens_per_sec: 221K | tokens_per_sec_per_gpu: 13.8K | global_batch_size: 1.02K | lm_loss: 11.4 | lr: 9.05e-05 | model_tflops_per_gpu: 125 | hardware_tflops_per_gpu: 125 | grad_norm: 217 | cuda_memory_allocated: 2.51G | cuda_max_memory_reserved: 15.5G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.8G | hd_free_memory_tb: 246G
265
+ [default0]:07/02/2024 19:37:51 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 5180.18MiB. Peak reserved: 38320.00MiB
266
+ [default0]:07/02/2024 19:38:07 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
267
+ [default4]:07/02/2024 19:38:11 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 4 / 20 | consumed_tokens: 16.8M | elapsed_time_per_iteration_ms: 19.8K | tokens_per_sec: 212K | tokens_per_sec_per_gpu: 13.3K | global_batch_size: 1.02K | lm_loss: 13.8 | lr: 8.58e-05 | model_tflops_per_gpu: 120 | hardware_tflops_per_gpu: 120 | grad_norm: 22.5 | cuda_memory_allocated: 2.51G | cuda_max_memory_reserved: 15.5G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.8G | hd_free_memory_tb: 246G
268
+ [default0]:07/02/2024 19:38:11 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 5180.18MiB. Peak reserved: 38320.00MiB
269
+ [default4]:07/02/2024 19:38:29 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 5 / 20 | consumed_tokens: 21M | elapsed_time_per_iteration_ms: 18.2K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 14.4K | global_batch_size: 1.02K | lm_loss: 9.98 | lr: 8.11e-05 | model_tflops_per_gpu: 131 | hardware_tflops_per_gpu: 131 | grad_norm: 16.4
270
+ [default0]:07/02/2024 19:38:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
271
+ [default4]:07/02/2024 19:38:47 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 6 / 20 | consumed_tokens: 25.2M | elapsed_time_per_iteration_ms: 18.6K | tokens_per_sec: 226K | tokens_per_sec_per_gpu: 14.1K | global_batch_size: 1.02K | lm_loss: 10.9 | lr: 7.63e-05 | model_tflops_per_gpu: 128 | hardware_tflops_per_gpu: 128 | grad_norm: 93.8
272
+ [default0]:STAGE:2024-07-02 19:38:59 663432:663432 ActivityProfilerController.cpp:320] Completed Stage: Collection
273
+ [default0]:STAGE:2024-07-02 19:39:00 663432:663432 ActivityProfilerController.cpp:324] Completed Stage: Post Processing
274
+ [default0]:07/02/2024 19:40:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
275
+ [default0]:07/02/2024 19:40:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
276
+ [default4]:07/02/2024 19:40:32 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 7 / 20 | consumed_tokens: 29.4M | elapsed_time_per_iteration_ms: 105K | tokens_per_sec: 40K | tokens_per_sec_per_gpu: 2.5K | global_batch_size: 1.02K | lm_loss: 9.16 | lr: 7.16e-05 | model_tflops_per_gpu: 22.7 | hardware_tflops_per_gpu: 22.7 | grad_norm: 19.8
277
+ [default4]:07/02/2024 19:40:50 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 8 / 20 | consumed_tokens: 33.6M | elapsed_time_per_iteration_ms: 18.2K | tokens_per_sec: 230K | tokens_per_sec_per_gpu: 14.4K | global_batch_size: 1.02K | lm_loss: 8.83 | lr: 6.68e-05 | model_tflops_per_gpu: 131 | hardware_tflops_per_gpu: 131 | grad_norm: 6.08
278
+ [default0]:07/02/2024 19:40:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
279
+ [default4]:07/02/2024 19:41:08 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 9 / 20 | consumed_tokens: 37.7M | elapsed_time_per_iteration_ms: 17.9K | tokens_per_sec: 235K | tokens_per_sec_per_gpu: 14.7K | global_batch_size: 1.02K | lm_loss: 8.47 | lr: 6.21e-05 | model_tflops_per_gpu: 133 | hardware_tflops_per_gpu: 133 | grad_norm: 5.23
280
+ [default0]:07/02/2024 19:41:08 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
281
+ [default0]:07/02/2024 19:41:25 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
282
+ [default4]:07/02/2024 19:41:25 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 10 / 20 | consumed_tokens: 41.9M | elapsed_time_per_iteration_ms: 17.2K | tokens_per_sec: 244K | tokens_per_sec_per_gpu: 15.2K | global_batch_size: 1.02K | lm_loss: 8.17 | lr: 5.74e-05 | model_tflops_per_gpu: 138 | hardware_tflops_per_gpu: 138 | grad_norm: 7.71
283
+ [default0]:07/02/2024 19:41:45 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
284
+ [default4]:07/02/2024 19:41:45 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 11 / 20 | consumed_tokens: 46.1M | elapsed_time_per_iteration_ms: 19.4K | tokens_per_sec: 216K | tokens_per_sec_per_gpu: 13.5K | global_batch_size: 1.02K | lm_loss: 7.93 | lr: 5.26e-05 | model_tflops_per_gpu: 123 | hardware_tflops_per_gpu: 123 | grad_norm: 5.53
285
+ [default0]:07/02/2024 19:42:04 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
286
+ [default4]:07/02/2024 19:42:04 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 12 / 20 | consumed_tokens: 50.3M | elapsed_time_per_iteration_ms: 18.8K | tokens_per_sec: 223K | tokens_per_sec_per_gpu: 14K | global_batch_size: 1.02K | lm_loss: 7.75 | lr: 4.79e-05 | model_tflops_per_gpu: 127 | hardware_tflops_per_gpu: 127 | grad_norm: 4.64
287
+ [default4]:07/02/2024 19:42:22 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 13 / 20 | consumed_tokens: 54.5M | elapsed_time_per_iteration_ms: 18.7K | tokens_per_sec: 224K | tokens_per_sec_per_gpu: 14K | global_batch_size: 1.02K | lm_loss: 7.58 | lr: 4.32e-05 | model_tflops_per_gpu: 127 | hardware_tflops_per_gpu: 127 | grad_norm: 2.9
288
+ [default0]:07/02/2024 19:42:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
289
+ [default0]:07/02/2024 19:42:39 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
290
+ [default4]:07/02/2024 19:42:39 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 14 / 20 | consumed_tokens: 58.7M | elapsed_time_per_iteration_ms: 16.2K | tokens_per_sec: 258K | tokens_per_sec_per_gpu: 16.2K | global_batch_size: 1.02K | lm_loss: 7.5 | lr: 3.84e-05 | model_tflops_per_gpu: 147 | hardware_tflops_per_gpu: 147 | grad_norm: 4.18
291
+ [default0]:07/02/2024 19:42:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
292
+ [default4]:07/02/2024 19:42:57 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 15 / 20 | consumed_tokens: 62.9M | elapsed_time_per_iteration_ms: 18.1K | tokens_per_sec: 232K | tokens_per_sec_per_gpu: 14.5K | global_batch_size: 1.02K | lm_loss: 7.4 | lr: 3.37e-05 | model_tflops_per_gpu: 131 | hardware_tflops_per_gpu: 131 | grad_norm: 3.86
293
+ [default0]:07/02/2024 19:43:14 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
294
+ [default4]:07/02/2024 19:43:14 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 16 / 20 | consumed_tokens: 67.1M | elapsed_time_per_iteration_ms: 17.7K | tokens_per_sec: 237K | tokens_per_sec_per_gpu: 14.8K | global_batch_size: 1.02K | lm_loss: 7.29 | lr: 2.89e-05 | model_tflops_per_gpu: 134 | hardware_tflops_per_gpu: 134 | grad_norm: 3.06
295
+ [default0]:07/02/2024 19:43:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
296
+ [default4]:07/02/2024 19:43:33 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 17 / 20 | consumed_tokens: 71.3M | elapsed_time_per_iteration_ms: 18.3K | tokens_per_sec: 229K | tokens_per_sec_per_gpu: 14.3K | global_batch_size: 1.02K | lm_loss: 7.19 | lr: 2.42e-05 | model_tflops_per_gpu: 130 | hardware_tflops_per_gpu: 130 | grad_norm: 2.39
297
+ [default4]:07/02/2024 19:43:53 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 18 / 20 | consumed_tokens: 75.5M | elapsed_time_per_iteration_ms: 20.3K | tokens_per_sec: 206K | tokens_per_sec_per_gpu: 12.9K | global_batch_size: 1.02K | lm_loss: 7.13 | lr: 1.95e-05 | model_tflops_per_gpu: 117 | hardware_tflops_per_gpu: 117 | grad_norm: 2.2
298
+ [default0]:07/02/2024 19:43:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
299
+ [default0]:07/02/2024 19:44:11 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-147]: Memory usage: 3478.34MiB. Peak allocated 37922.28MiB. Peak reserved: 38320.00MiB
300
+ [default4]:07/02/2024 19:44:11 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 19 / 20 | consumed_tokens: 79.7M | elapsed_time_per_iteration_ms: 17.8K | tokens_per_sec: 236K | tokens_per_sec_per_gpu: 14.8K | global_batch_size: 1.02K | lm_loss: 7.08 | lr: 1.47e-05 | model_tflops_per_gpu: 134 | hardware_tflops_per_gpu: 134 | grad_norm: 2.64
301
+ [default4]:07/02/2024 19:44:28 [INFO|DP=0|PP=3|TP=0|ip-26-0-163-226]: iteration: 20 / 20 | consumed_tokens: 83.9M | elapsed_time_per_iteration_ms: 17.2K | tokens_per_sec: 244K | tokens_per_sec_per_gpu: 15.2K | global_batch_size: 1.02K | lm_loss: 7.03 | lr: 1e-05 | model_tflops_per_gpu: 138 | hardware_tflops_per_gpu: 138 | grad_norm: 2.3
302
+ W0702 19:44:49.386000 140366572283648 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1252] The node 'ip-26-0-163-226.ec2.internal_3069344_0' has failed to send a keep-alive heartbeat to the rendezvous 'none' due to an error of type RendezvousTimeoutError.
303
+ Traceback (most recent call last):
304
+ File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py", line 4, in <module>
305
+ from bench_cluster.submit_jobs import submit_jobs, check_status
306
+ ImportError: cannot import name 'check_status' from 'bench_cluster.submit_jobs' (/fsx/ferdinandmom/ferdinand-hf/bench_cluster/bench_cluster/submit_jobs.py)
307
+ Traceback (most recent call last):
308
+ File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py", line 4, in <module>
309
+ from bench_cluster.submit_jobs import submit_jobs, check_status
310
+ ImportError: cannot import name 'check_status' from 'bench_cluster.submit_jobs' (/fsx/ferdinandmom/ferdinand-hf/bench_cluster/bench_cluster/submit_jobs.py)
311
+ 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.
312
+
llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-4/profiler/ip-26-0-163-147_663432.1719949200647344699.pt.trace.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ad4e5e2208bc00923c2bb704f4365d73ef37831bda34f91ac6ae1033fa1f27b4
3
+ size 2571198645
llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-4/status.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ completed