Hugging Face's logo

nanotron
/
old_bench

Model card Files Community
3outeille HF staff commited on
Commit
11a99a9
•
1 Parent(s): 8a76c11

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

Browse files
Files changed (6) hide show
  1. .gitattributes +1 -0
  2. llama-1B/16_GPUS/dp-8_tp-2_pp-1_mbz-8/bench.slurm +111 -0
  3. llama-1B/16_GPUS/dp-8_tp-2_pp-1_mbz-8/config.yaml +90 -0
  4. llama-1B/16_GPUS/dp-8_tp-2_pp-1_mbz-8/log.out +324 -0
  5. llama-1B/16_GPUS/dp-8_tp-2_pp-1_mbz-8/profiler/ip-26-0-165-24_682306.1719946948445078242.pt.trace.json +3 -0
  6. llama-1B/16_GPUS/dp-8_tp-2_pp-1_mbz-8/status.txt +1 -0
.gitattributes CHANGED
@@ -49,3 +49,4 @@ llama-1B/16_GPUS/dp-4_tp-2_pp-2_mbz-2/profiler/ip-26-0-171-102_3600759.171994548
49
  llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-2/profiler/ip-26-0-160-192_925006.1719945618152196843.pt.trace.json filter=lfs diff=lfs merge=lfs -text
50
  llama-1B/16_GPUS/dp-1_tp-8_pp-2_mbz-16/profiler/ip-26-0-165-24_655682.1719946219098357028.pt.trace.json filter=lfs diff=lfs merge=lfs -text
51
  llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/profiler/ip-26-0-171-62_3731127.1719946421954651053.pt.trace.json filter=lfs diff=lfs merge=lfs -text
 
 
49
  llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-2/profiler/ip-26-0-160-192_925006.1719945618152196843.pt.trace.json filter=lfs diff=lfs merge=lfs -text
50
  llama-1B/16_GPUS/dp-1_tp-8_pp-2_mbz-16/profiler/ip-26-0-165-24_655682.1719946219098357028.pt.trace.json filter=lfs diff=lfs merge=lfs -text
51
  llama-1B/16_GPUS/dp-2_tp-2_pp-4_mbz-8/profiler/ip-26-0-171-62_3731127.1719946421954651053.pt.trace.json filter=lfs diff=lfs merge=lfs -text
52
+ llama-1B/16_GPUS/dp-8_tp-2_pp-1_mbz-8/profiler/ip-26-0-165-24_682306.1719946948445078242.pt.trace.json filter=lfs diff=lfs merge=lfs -text
llama-1B/16_GPUS/dp-8_tp-2_pp-1_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-8_tp-2_pp-1_mbz-8/log.out
13
+ #SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-8_tp-2_pp-1_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-8_tp-2_pp-1_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-8_tp-2_pp-1_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-8_tp-2_pp-1_mbz-8/status.txt
84
+ else
85
+ if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-8_tp-2_pp-1_mbz-8/log.out; then
86
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-8_tp-2_pp-1_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-8_tp-2_pp-1_mbz-8/log.out; then
88
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-8_tp-2_pp-1_mbz-8/status.txt
89
+ elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-8_tp-2_pp-1_mbz-8/log.out; then
90
+ printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-8_tp-2_pp-1_mbz-8/status.txt
91
+ else
92
+ printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-8_tp-2_pp-1_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-8_tp-2_pp-1_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-8_tp-2_pp-1_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-8_tp-2_pp-1_mbz-8 llama-1B/16_GPUS/dp-8_tp-2_pp-1_mbz-8 --commit-message "Upload llama-1B/16_GPUS/dp-8_tp-2_pp-1_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-8_tp-2_pp-1_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: 8
49
+ expert_parallel_size: 1
50
+ pp: 1
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-8_tp-2_pp-1_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: 16
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-8_tp-2_pp-1_mbz-8/log.out ADDED
@@ -0,0 +1,324 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ========================
2
+ START TIME: Tue Jul 2 18:59:20 UTC 2024
3
+ python3 version = Python 3.10.14
4
+ ========================
5
+ The token has not been saved to the git credentials helper. Pass `add_to_git_credential=True` in this function directly or `--add-to-git-credential` if using via `huggingface-cli` if you want to set the git credential as well.
6
+ Token is valid (permission: write).
7
+ Your token has been saved to /admin/home/ferdinand_mom/.cache/huggingface/token
8
+ Login successful
9
+ Already on 'bench_cluster'
10
+ M examples/config_tiny_llama.py
11
+ M examples/config_tiny_llama.yaml
12
+ M examples/train_tiny_llama.sh
13
+ M src/nanotron/models/llama.py
14
+ M src/nanotron/trainer.py
15
+ Your branch is up to date with 'origin/bench_cluster'.
16
+ Job status: RUNNING
17
+ W0702 18:59:22.989000 140181104912192 torch/distributed/run.py:757]
18
+ W0702 18:59:22.989000 140181104912192 torch/distributed/run.py:757] *****************************************
19
+ W0702 18:59:22.989000 140181104912192 torch/distributed/run.py:757] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
20
+ W0702 18:59:22.989000 140181104912192 torch/distributed/run.py:757] *****************************************
21
+ W0702 18:59:22.992000 139700739450688 torch/distributed/run.py:757]
22
+ W0702 18:59:22.992000 139700739450688 torch/distributed/run.py:757] *****************************************
23
+ W0702 18:59:22.992000 139700739450688 torch/distributed/run.py:757] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
24
+ W0702 18:59:22.992000 139700739450688 torch/distributed/run.py:757] *****************************************
25
+ [default0]:07/02/2024 18:59:41 [WARNING|DP=0|PP=0|TP=0|ip-26-0-165-24]: [Vocab Size Padding] Padded vocab (size: 50257) with 1 dummy tokens (new size: 50258)
26
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Config:
27
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Config(general=GeneralArgs(project='bench_cluster',
28
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: run='%date_%jobid',
29
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: seed=42,
30
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: step=None,
31
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: consumed_train_samples=None,
32
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: benchmark_csv_path=None,
33
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: ignore_sanity_checks=True),
34
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: parallelism=ParallelismArgs(dp=8,
35
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: pp=1,
36
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tp=2,
37
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f3eb23e4910>,
38
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
39
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tp_linear_async_communication=False,
40
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: expert_parallel_size=1),
41
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
42
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: eos_token_id=2,
43
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: hidden_act='silu',
44
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: hidden_size=2048,
45
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: initializer_range=0.02,
46
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: intermediate_size=4096,
47
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: is_llama_config=True,
48
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: max_position_embeddings=4096,
49
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: num_attention_heads=32,
50
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: num_hidden_layers=24,
51
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: num_key_value_heads=32,
52
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: pad_token_id=None,
53
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: pretraining_tp=1,
54
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: rms_norm_eps=1e-05,
55
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: rope_scaling=None,
56
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: rope_theta=10000.0,
57
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tie_word_embeddings=True,
58
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: use_cache=True,
59
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: vocab_size=50258),
60
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: init_method=RandomInit(std=0.025),
61
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: dtype=torch.bfloat16,
62
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: make_vocab_size_divisible_by=1,
63
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: ddp_bucket_cap_mb=25),
64
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
65
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tokenizer_revision=None,
66
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tokenizer_max_length=None),
67
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
68
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: checkpoint_interval=100000,
69
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: save_initial_state=False,
70
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: resume_checkpoint_path=None,
71
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: checkpoints_path_is_shared_file_system=False),
72
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: logging=LoggingArgs(log_level='info',
73
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: log_level_replica='info',
74
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration_step_info_interval=1),
75
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tokens=TokensArgs(sequence_length=4096,
76
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: train_steps=20,
77
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: micro_batch_size=8,
78
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: batch_accumulation_per_replica=16,
79
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: val_check_interval=-1,
80
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: limit_val_batches=0,
81
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: limit_test_batches=0),
82
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
83
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: adam_beta1=0.9,
84
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: adam_beta2=0.95,
85
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: torch_adam_is_fused=True,
86
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: name='adamW'),
87
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: zero_stage=1,
88
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: weight_decay=0.01,
89
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: clip_grad=1.0,
90
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: accumulate_grad_in_fp32=True,
91
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
92
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: lr_warmup_steps=1,
93
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: lr_warmup_style='linear',
94
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: lr_decay_style='linear',
95
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: lr_decay_steps=19,
96
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: lr_decay_starting_step=None,
97
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: min_decay_lr=1e-05)),
98
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: data_stages=[DatasetStageArgs(name='Training Stage',
99
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: start_training_step=1,
100
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
101
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: hf_dataset_splits='train',
102
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: hf_dataset_config_name=None,
103
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: dataset_processing_num_proc_per_process=64,
104
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: dataset_overwrite_cache=False,
105
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: text_column_name='text'),
106
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: seed=42,
107
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: num_loading_workers=32))],
108
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-8_tp-2_pp-1_mbz-8')),
109
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: lighteval=None)
110
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Model Config:
111
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: LlamaConfig(bos_token_id=1,
112
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: eos_token_id=2,
113
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: hidden_act='silu',
114
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: hidden_size=2048,
115
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: initializer_range=0.02,
116
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: intermediate_size=4096,
117
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: is_llama_config=True,
118
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: max_position_embeddings=4096,
119
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: num_attention_heads=32,
120
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: num_hidden_layers=24,
121
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: num_key_value_heads=32,
122
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: pad_token_id=None,
123
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: pretraining_tp=1,
124
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: rms_norm_eps=1e-05,
125
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: rope_scaling=None,
126
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: rope_theta=10000.0,
127
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: tie_word_embeddings=True,
128
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: use_cache=True,
129
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: vocab_size=50258)
130
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Building model..
131
+ [default0]:07/02/2024 18:59:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Setting PP block ranks...
132
+ [default0]:07/02/2024 18:59:52 [INFO|DP=4|PP=0|TP=0|ip-26-0-170-160]: No checkpoint path provided.
133
+ [default1]:07/02/2024 18:59:52 [INFO|DP=4|PP=0|TP=1|ip-26-0-170-160]: No checkpoint path provided.
134
+ [default3]:07/02/2024 18:59:52 [INFO|DP=5|PP=0|TP=1|ip-26-0-170-160]: No checkpoint path provided.
135
+ [default2]:07/02/2024 18:59:52 [INFO|DP=5|PP=0|TP=0|ip-26-0-170-160]: No checkpoint path provided.
136
+ [default4]:07/02/2024 18:59:52 [INFO|DP=6|PP=0|TP=0|ip-26-0-170-160]: No checkpoint path provided.
137
+ [default5]:07/02/2024 18:59:52 [INFO|DP=6|PP=0|TP=1|ip-26-0-170-160]: No checkpoint path provided.
138
+ [default6]:07/02/2024 18:59:52 [INFO|DP=7|PP=0|TP=0|ip-26-0-170-160]: No checkpoint path provided.
139
+ [default7]:07/02/2024 18:59:52 [INFO|DP=7|PP=0|TP=1|ip-26-0-170-160]: No checkpoint path provided.
140
+ [default0]:07/02/2024 18:59:52 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Total number of parameters: 1.11G (2116.70MiB)
141
+ [default0]:07/02/2024 18:59:52 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Local number of parameters: 555M (1058.35MiB)
142
+ [default0]:07/02/2024 18:59:52 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [After model building] Memory usage: 1082.37MiB. Peak allocated: 1182.56MiB Peak reserved: 1200.00MiB
143
+ [default0]:07/02/2024 18:59:52 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: No checkpoint path provided.
144
+ [default0]:07/02/2024 18:59:52 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Parametrizing model parameters using StandardParametrizator
145
+ [default1]:07/02/2024 18:59:52 [INFO|DP=0|PP=0|TP=1|ip-26-0-165-24]: Local number of parameters: 555M (1058.35MiB)
146
+ [default1]:07/02/2024 18:59:52 [INFO|DP=0|PP=0|TP=1|ip-26-0-165-24]: [After model building] Memory usage: 1082.37MiB. Peak allocated: 1182.56MiB Peak reserved: 1200.00MiB
147
+ [default1]:07/02/2024 18:59:52 [INFO|DP=0|PP=0|TP=1|ip-26-0-165-24]: No checkpoint path provided.
148
+ [default4]:07/02/2024 18:59:52 [INFO|DP=2|PP=0|TP=0|ip-26-0-165-24]: No checkpoint path provided.
149
+ [default5]:07/02/2024 18:59:52 [INFO|DP=2|PP=0|TP=1|ip-26-0-165-24]: No checkpoint path provided.
150
+ [default2]:07/02/2024 18:59:52 [INFO|DP=1|PP=0|TP=0|ip-26-0-165-24]: No checkpoint path provided.
151
+ [default3]:07/02/2024 18:59:52 [INFO|DP=1|PP=0|TP=1|ip-26-0-165-24]: No checkpoint path provided.
152
+ [default6]:07/02/2024 18:59:52 [INFO|DP=3|PP=0|TP=0|ip-26-0-165-24]: No checkpoint path provided.
153
+ [default7]:07/02/2024 18:59:52 [INFO|DP=3|PP=0|TP=1|ip-26-0-165-24]: No checkpoint path provided.
154
+ [default0]:07/02/2024 18:59:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [Optimizer Building] Using LearningRateForSP as learning rate
155
+ [default0]:07/02/2024 18:59:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [ZeRO sharding] Size of optimizer params per rank:
156
+ [default0]:07/02/2024 18:59:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [ZeRO sharding] DP Rank 0 has 69.4M out of 555M (12.50%) params' optimizer states
157
+ [default0]:07/02/2024 18:59:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [ZeRO sharding] DP Rank 1 has 69.4M out of 555M (12.50%) params' optimizer states
158
+ [default0]:07/02/2024 18:59:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [ZeRO sharding] DP Rank 2 has 69.4M out of 555M (12.50%) params' optimizer states
159
+ [default0]:07/02/2024 18:59:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [ZeRO sharding] DP Rank 3 has 69.4M out of 555M (12.50%) params' optimizer states
160
+ [default0]:07/02/2024 18:59:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [ZeRO sharding] DP Rank 4 has 69.4M out of 555M (12.50%) params' optimizer states
161
+ [default0]:07/02/2024 18:59:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [ZeRO sharding] DP Rank 5 has 69.4M out of 555M (12.50%) params' optimizer states
162
+ [default0]:07/02/2024 18:59:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [ZeRO sharding] DP Rank 6 has 69.4M out of 555M (12.50%) params' optimizer states
163
+ [default0]:07/02/2024 18:59:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [ZeRO sharding] DP Rank 7 has 69.4M out of 555M (12.50%) params' optimizer states
164
+ [default0]:07/02/2024 18:59:59 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
165
+ [default0]:07/02/2024 18:59:59 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Using `datasets` library
166
+ [default0]:07/02/2024 18:59:59 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
167
+ [default0]:07/02/2024 18:59:59 [WARNING|DP=0|PP=0|TP=0|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
168
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
169
+ [default0]:07/02/2024 19:00:00 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [Training Plan] There are 1 training stages
170
+ [default0]:07/02/2024 19:00:00 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [Stage Training Stage] start from step 1
171
+ [default0]:07/02/2024 19:00:00 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]:
172
+ [default0]:07/02/2024 19:00:00 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: [Start training] datetime: 2024-07-02 19:00:00.508957 | mbs: 8 | grad_accum: 16 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
173
+ [default0]:07/02/2024 19:00:00 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
174
+ [default0]:07/02/2024 19:00:00 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 3463.66MiB. Peak allocated 3463.66MiB. Peak reserved: 3584.00MiB
175
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
176
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
177
+ [default0]:07/02/2024 19:00:00 [WARNING|DP=4|PP=0|TP=0|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
178
+ [default1]:07/02/2024 19:00:00 [WARNING|DP=4|PP=0|TP=1|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
179
+ [default4]:07/02/2024 19:00:00 [WARNING|DP=2|PP=0|TP=0|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
180
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
181
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
182
+ [default3]:07/02/2024 19:00:00 [WARNING|DP=5|PP=0|TP=1|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
183
+ [default7]:07/02/2024 19:00:00 [WARNING|DP=7|PP=0|TP=1|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
184
+ [default2]:07/02/2024 19:00:00 [WARNING|DP=5|PP=0|TP=0|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
185
+ [default4]:07/02/2024 19:00:00 [WARNING|DP=6|PP=0|TP=0|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
186
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
187
+ [default5]:07/02/2024 19:00:00 [WARNING|DP=6|PP=0|TP=1|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
188
+ [default4]: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
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
191
+ [default5]:07/02/2024 19:00:00 [WARNING|DP=2|PP=0|TP=1|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
192
+ [default2]:07/02/2024 19:00:00 [WARNING|DP=1|PP=0|TP=0|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
193
+ [default1]:07/02/2024 19:00:00 [WARNING|DP=0|PP=0|TP=1|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
194
+ [default3]:07/02/2024 19:00:00 [WARNING|DP=1|PP=0|TP=1|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
195
+ [default7]:07/02/2024 19:00:00 [WARNING|DP=3|PP=0|TP=1|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
196
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
197
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
198
+ [default6]:07/02/2024 19:00:00 [WARNING|DP=3|PP=0|TP=0|ip-26-0-165-24]: Repo card metadata block was not found. Setting CardData to empty.
199
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
200
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
201
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
202
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
203
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
204
+ [default6]:07/02/2024 19:00:00 [WARNING|DP=7|PP=0|TP=0|ip-26-0-170-160]: Repo card metadata block was not found. Setting CardData to empty.
205
+ [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.)
206
+ [default1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
207
+ [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.)
208
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
209
+ [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.)
210
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
211
+ [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.)
212
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
213
+ [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.)
214
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
215
+ [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.)
216
+ [default7]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
217
+ [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.)
218
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
219
+ [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.)
220
+ [default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
221
+ [default2]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: c10d::allreduce_: an autograd kernel was not registered to the Autograd key(s) but we are trying to backprop through it. This may lead to silently incorrect behavior. This behavior is deprecated and will be removed in a future version of PyTorch. If your operator is differentiable, please ensure you have registered an autograd kernel to the correct Autograd key (e.g. DispatchKey::Autograd, DispatchKey::CompositeImplicitAutograd). If your operator is not differentiable, or to squash this warning and use the previous behavior, please register torch::CppFunction::makeFallthrough() to DispatchKey::Autograd. (Triggered internally at ../torch/csrc/autograd/autograd_not_implemented_fallback.cpp:63.)
222
+ [default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
223
+ [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.)
224
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
225
+ [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.)
226
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
227
+ [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.)
228
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
229
+ [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.)
230
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
231
+ [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.)
232
+ [default7]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
233
+ [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.)
234
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
235
+ [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.)
236
+ [default1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
237
+ [default0]:07/02/2024 19:00:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 3540.23MiB. Peak allocated 42325.02MiB. Peak reserved: 44112.00MiB
238
+ [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
239
+ [default1]: warnings.warn(
240
+ [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
241
+ [default0]: warnings.warn(
242
+ [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
243
+ [default4]: warnings.warn(
244
+ [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
245
+ [default5]: warnings.warn(
246
+ [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
247
+ [default6]: warnings.warn(
248
+ [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
249
+ [default7]: warnings.warn(
250
+ [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
251
+ [default3]: warnings.warn(
252
+ [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
253
+ [default2]: warnings.warn(
254
+ [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
255
+ [default2]: warnings.warn(
256
+ [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
257
+ [default4]: warnings.warn(
258
+ [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
259
+ [default7]: warnings.warn(
260
+ [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
261
+ [default3]: warnings.warn(
262
+ [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
263
+ [default6]: warnings.warn(
264
+ [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
265
+ [default5]: warnings.warn(
266
+ [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
267
+ [default0]: warnings.warn(
268
+ [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
269
+ [default1]: warnings.warn(
270
+ [default0]:07/02/2024 19:00:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 1 / 20 | consumed_tokens: 4.19M | elapsed_time_per_iteration_ms: 22.2K | tokens_per_sec: 189K | tokens_per_sec_per_gpu: 11.8K | global_batch_size: 1.02K | lm_loss: 11.5 | lr: 0.0001 | model_tflops_per_gpu: 107 | hardware_tflops_per_gpu: 107 | grad_norm: 26.4 | cuda_memory_allocated: 4.27G | cuda_max_memory_reserved: 46.3G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.6G | hd_free_memory_tb: 247G
271
+ [default0]:07/02/2024 19:00:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 6324.40MiB. Peak reserved: 44160.00MiB
272
+ [default0]:07/02/2024 19:00:30 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.41MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
273
+ [default0]:07/02/2024 19:00:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 2 / 20 | consumed_tokens: 8.39M | elapsed_time_per_iteration_ms: 10.2K | tokens_per_sec: 413K | tokens_per_sec_per_gpu: 25.8K | global_batch_size: 1.02K | lm_loss: 11.5 | lr: 9.53e-05 | model_tflops_per_gpu: 234 | hardware_tflops_per_gpu: 234 | grad_norm: 26.6 | cuda_memory_allocated: 4.27G | cuda_max_memory_reserved: 46.3G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.6G | hd_free_memory_tb: 247G
274
+ [default0]:07/02/2024 19:00:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 6324.41MiB. Peak reserved: 44162.00MiB
275
+ [default0]:07/02/2024 19:00:40 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.41MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
276
+ [default0]:07/02/2024 19:00:42 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 3 / 20 | consumed_tokens: 12.6M | elapsed_time_per_iteration_ms: 9.82K | tokens_per_sec: 427K | tokens_per_sec_per_gpu: 26.7K | global_batch_size: 1.02K | lm_loss: 11.5 | lr: 9.05e-05 | model_tflops_per_gpu: 242 | hardware_tflops_per_gpu: 242 | grad_norm: 262 | cuda_memory_allocated: 4.27G | cuda_max_memory_reserved: 46.3G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.6G | hd_free_memory_tb: 247G
277
+ [default0]:07/02/2024 19:00:42 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 6324.41MiB. Peak reserved: 44162.00MiB
278
+ [default0]:STAGE:2024-07-02 19:00:42 682306:682306 ActivityProfilerController.cpp:314] Completed Stage: Warm Up
279
+ [default0]:07/02/2024 19:00:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.41MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
280
+ [default0]:07/02/2024 19:00:52 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 4 / 20 | consumed_tokens: 16.8M | elapsed_time_per_iteration_ms: 9.75K | tokens_per_sec: 430K | tokens_per_sec_per_gpu: 26.9K | global_batch_size: 1.02K | lm_loss: 14.6 | lr: 8.58e-05 | model_tflops_per_gpu: 244 | hardware_tflops_per_gpu: 244 | grad_norm: 29.1 | cuda_memory_allocated: 4.27G | cuda_max_memory_reserved: 46.3G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.6G | hd_free_memory_tb: 247G
281
+ [default0]:07/02/2024 19:00:52 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 6324.41MiB. Peak reserved: 44162.00MiB
282
+ [default0]:07/02/2024 19:01:02 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 5 / 20 | consumed_tokens: 21M | elapsed_time_per_iteration_ms: 9.97K | tokens_per_sec: 421K | tokens_per_sec_per_gpu: 26.3K | global_batch_size: 1.02K | lm_loss: 10.8 | lr: 8.11e-05 | model_tflops_per_gpu: 238 | hardware_tflops_per_gpu: 238 | grad_norm: 31
283
+ [default0]:07/02/2024 19:01:02 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
284
+ [default0]:07/02/2024 19:01:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 6 / 20 | consumed_tokens: 25.2M | elapsed_time_per_iteration_ms: 9.98K | tokens_per_sec: 420K | tokens_per_sec_per_gpu: 26.3K | global_batch_size: 1.02K | lm_loss: 10.6 | lr: 7.63e-05 | model_tflops_per_gpu: 238 | hardware_tflops_per_gpu: 238 | grad_norm: 27.4
285
+ [default0]:STAGE:2024-07-02 19:01:22 682306:682306 ActivityProfilerController.cpp:320] Completed Stage: Collection
286
+ [default0]:STAGE:2024-07-02 19:01:24 682306:682306 ActivityProfilerController.cpp:324] Completed Stage: Post Processing
287
+ [default0]:07/02/2024 19:02:44 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
288
+ [default0]:07/02/2024 19:02:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 7 / 20 | consumed_tokens: 29.4M | elapsed_time_per_iteration_ms: 9.83K | tokens_per_sec: 427K | tokens_per_sec_per_gpu: 26.7K | global_batch_size: 1.02K | lm_loss: 10.2 | lr: 7.16e-05 | model_tflops_per_gpu: 242 | hardware_tflops_per_gpu: 242 | grad_norm: 9.44
289
+ [default0]:07/02/2024 19:02:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
290
+ [default0]:07/02/2024 19:03:04 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 8 / 20 | consumed_tokens: 33.6M | elapsed_time_per_iteration_ms: 9.72K | tokens_per_sec: 432K | tokens_per_sec_per_gpu: 27K | global_batch_size: 1.02K | lm_loss: 13 | lr: 6.68e-05 | model_tflops_per_gpu: 245 | hardware_tflops_per_gpu: 245 | grad_norm: 78.4
291
+ [default0]:07/02/2024 19:03:04 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
292
+ [default0]:07/02/2024 19:03:14 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 9 / 20 | consumed_tokens: 37.7M | elapsed_time_per_iteration_ms: 9.79K | tokens_per_sec: 429K | tokens_per_sec_per_gpu: 26.8K | global_batch_size: 1.02K | lm_loss: 9.45 | lr: 6.21e-05 | model_tflops_per_gpu: 243 | hardware_tflops_per_gpu: 243 | grad_norm: 12.9
293
+ [default0]:07/02/2024 19:03:14 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
294
+ [default0]:07/02/2024 19:03:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 10 / 20 | consumed_tokens: 41.9M | elapsed_time_per_iteration_ms: 9.76K | tokens_per_sec: 430K | tokens_per_sec_per_gpu: 26.8K | global_batch_size: 1.02K | lm_loss: 9.22 | lr: 5.74e-05 | model_tflops_per_gpu: 244 | hardware_tflops_per_gpu: 244 | grad_norm: 6.81
295
+ [default0]:07/02/2024 19:03:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
296
+ [default0]:07/02/2024 19:03:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 11 / 20 | consumed_tokens: 46.1M | elapsed_time_per_iteration_ms: 9.98K | tokens_per_sec: 420K | tokens_per_sec_per_gpu: 26.3K | global_batch_size: 1.02K | lm_loss: 8.95 | lr: 5.26e-05 | model_tflops_per_gpu: 238 | hardware_tflops_per_gpu: 238 | grad_norm: 5.83
297
+ [default0]:07/02/2024 19:03:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
298
+ [default0]:07/02/2024 19:03:44 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 12 / 20 | consumed_tokens: 50.3M | elapsed_time_per_iteration_ms: 10.2K | tokens_per_sec: 413K | tokens_per_sec_per_gpu: 25.8K | global_batch_size: 1.02K | lm_loss: 8.59 | lr: 4.79e-05 | model_tflops_per_gpu: 234 | hardware_tflops_per_gpu: 234 | grad_norm: 6.29
299
+ [default0]:07/02/2024 19:03:44 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
300
+ [default0]:07/02/2024 19:03:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 13 / 20 | consumed_tokens: 54.5M | elapsed_time_per_iteration_ms: 9.79K | tokens_per_sec: 428K | tokens_per_sec_per_gpu: 26.8K | global_batch_size: 1.02K | lm_loss: 8.14 | lr: 4.32e-05 | model_tflops_per_gpu: 243 | hardware_tflops_per_gpu: 243 | grad_norm: 5.48
301
+ [default0]:07/02/2024 19:03:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
302
+ [default0]:07/02/2024 19:04:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 14 / 20 | consumed_tokens: 58.7M | elapsed_time_per_iteration_ms: 9.97K | tokens_per_sec: 421K | tokens_per_sec_per_gpu: 26.3K | global_batch_size: 1.02K | lm_loss: 7.72 | lr: 3.84e-05 | model_tflops_per_gpu: 239 | hardware_tflops_per_gpu: 239 | grad_norm: 4.86
303
+ [default0]:07/02/2024 19:04:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
304
+ [default0]:07/02/2024 19:04:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 15 / 20 | consumed_tokens: 62.9M | elapsed_time_per_iteration_ms: 9.79K | tokens_per_sec: 428K | tokens_per_sec_per_gpu: 26.8K | global_batch_size: 1.02K | lm_loss: 7.49 | lr: 3.37e-05 | model_tflops_per_gpu: 243 | hardware_tflops_per_gpu: 243 | grad_norm: 5.16
305
+ [default0]:07/02/2024 19:04:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
306
+ [default0]:07/02/2024 19:04:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 16 / 20 | consumed_tokens: 67.1M | elapsed_time_per_iteration_ms: 10K | tokens_per_sec: 418K | tokens_per_sec_per_gpu: 26.1K | global_batch_size: 1.02K | lm_loss: 7.39 | lr: 2.89e-05 | model_tflops_per_gpu: 237 | hardware_tflops_per_gpu: 237 | grad_norm: 6.94
307
+ [default0]:07/02/2024 19:04:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
308
+ [default0]:07/02/2024 19:04:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 17 / 20 | consumed_tokens: 71.3M | elapsed_time_per_iteration_ms: 9.77K | tokens_per_sec: 429K | tokens_per_sec_per_gpu: 26.8K | global_batch_size: 1.02K | lm_loss: 7.35 | lr: 2.42e-05 | model_tflops_per_gpu: 243 | hardware_tflops_per_gpu: 243 | grad_norm: 5.96
309
+ [default0]:07/02/2024 19:04:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
310
+ [default0]:07/02/2024 19:04:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 18 / 20 | consumed_tokens: 75.5M | elapsed_time_per_iteration_ms: 9.85K | tokens_per_sec: 426K | tokens_per_sec_per_gpu: 26.6K | global_batch_size: 1.02K | lm_loss: 7.32 | lr: 1.95e-05 | model_tflops_per_gpu: 242 | hardware_tflops_per_gpu: 242 | grad_norm: 6.89
311
+ [default0]:07/02/2024 19:04:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
312
+ [default0]:07/02/2024 19:04:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 19 / 20 | consumed_tokens: 79.7M | elapsed_time_per_iteration_ms: 9.79K | tokens_per_sec: 428K | tokens_per_sec_per_gpu: 26.8K | global_batch_size: 1.02K | lm_loss: 7.21 | lr: 1.47e-05 | model_tflops_per_gpu: 243 | hardware_tflops_per_gpu: 243 | grad_norm: 5.6
313
+ [default0]:07/02/2024 19:04:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: Memory usage: 4075.40MiB. Peak allocated 42861.19MiB. Peak reserved: 44162.00MiB
314
+ [default0]:07/02/2024 19:05:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-165-24]: iteration: 20 / 20 | consumed_tokens: 83.9M | elapsed_time_per_iteration_ms: 10K | tokens_per_sec: 418K | tokens_per_sec_per_gpu: 26.1K | global_batch_size: 1.02K | lm_loss: 7.11 | lr: 1e-05 | model_tflops_per_gpu: 237 | hardware_tflops_per_gpu: 237 | grad_norm: 4.32
315
+ Traceback (most recent call last):
316
+ File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py", line 4, in <module>
317
+ from bench_cluster.submit_jobs import submit_jobs, check_status
318
+ ImportError: cannot import name 'check_status' from 'bench_cluster.submit_jobs' (/fsx/ferdinandmom/ferdinand-hf/bench_cluster/bench_cluster/submit_jobs.py)
319
+ Traceback (most recent call last):
320
+ File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py", line 4, in <module>
321
+ from bench_cluster.submit_jobs import submit_jobs, check_status
322
+ ImportError: cannot import name 'check_status' from 'bench_cluster.submit_jobs' (/fsx/ferdinandmom/ferdinand-hf/bench_cluster/bench_cluster/submit_jobs.py)
323
+ 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.
324
+
llama-1B/16_GPUS/dp-8_tp-2_pp-1_mbz-8/profiler/ip-26-0-165-24_682306.1719946948445078242.pt.trace.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2803663212c04d5d791d5f7e5a0de1328497f85b21f5b0eb050b655338efe2bd
3
+ size 2282776027
llama-1B/16_GPUS/dp-8_tp-2_pp-1_mbz-8/status.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ completed