Hugging Face's logo

nanotron
/
old_bench

Model card Files Community
3outeille HF staff commited on
Commit
8a7bc97
•
1 Parent(s): 364526b

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

Browse files
Files changed (8) hide show
  1. .gitattributes +1 -0
  2. llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/bench.slurm +111 -0
  3. llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/config.yaml +90 -0
  4. llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/log.out +257 -0
  5. llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/log_metrics.csv +21 -0
  6. llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/profiler.csv +2 -0
  7. llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/profiler/ip-26-0-162-233_1835168.1720046865197689959.pt.trace.json +3 -0
  8. llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/status.txt +1 -0
.gitattributes CHANGED
@@ -109,3 +109,4 @@ llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-2/profiler/ip-26-0-174-36_223115.172004284691
109
  llama-1B/8_GPUS/dp-2_tp-2_pp-2_mbz-4/profiler/ip-26-0-164-207_594258.1720042744576660800.pt.trace.json filter=lfs diff=lfs merge=lfs -text
110
  llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-4/profiler/ip-26-0-160-225_339725.1720044373966110627.pt.trace.json filter=lfs diff=lfs merge=lfs -text
111
  llama-1B/8_GPUS/dp-2_tp-1_pp-4_mbz-4/profiler/ip-26-0-164-207_605035.1720046681997701534.pt.trace.json filter=lfs diff=lfs merge=lfs -text
 
 
109
  llama-1B/8_GPUS/dp-2_tp-2_pp-2_mbz-4/profiler/ip-26-0-164-207_594258.1720042744576660800.pt.trace.json filter=lfs diff=lfs merge=lfs -text
110
  llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-4/profiler/ip-26-0-160-225_339725.1720044373966110627.pt.trace.json filter=lfs diff=lfs merge=lfs -text
111
  llama-1B/8_GPUS/dp-2_tp-1_pp-4_mbz-4/profiler/ip-26-0-164-207_605035.1720046681997701534.pt.trace.json filter=lfs diff=lfs merge=lfs -text
112
+ llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/profiler/ip-26-0-162-233_1835168.1720046865197689959.pt.trace.json filter=lfs diff=lfs merge=lfs -text
llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/bench.slurm ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+
3
+ #SBATCH --job-name=bench_cluster
4
+ #SBATCH --time=02:00:00
5
+ #SBATCH --partition=hopper-prod
6
+ #SBATCH --nodes=1
7
+ #SBATCH --gres=gpu:8
8
+ #SBATCH --qos=normal
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/8_GPUS/dp-2_tp-4_pp-1_mbz-16/log.out
13
+ #SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/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/8_GPUS/dp-2_tp-4_pp-1_mbz-16/config.yaml"
57
+
58
+ LAUNCHER="torchrun \
59
+ --nproc_per_node 8 \
60
+ --nnodes 1 \
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/8_GPUS/dp-2_tp-4_pp-1_mbz-16/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/8_GPUS/dp-2_tp-4_pp-1_mbz-16/status.txt
84
+ else
85
+ if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/log.out; then
86
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/status.txt
87
+ elif grep -q " CUDA error: an illegal memory access" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/log.out; then
88
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/status.txt
89
+ elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/log.out; then
90
+ printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/status.txt
91
+ else
92
+ printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/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/8_GPUS/dp-2_tp-4_pp-1_mbz-16 --is_logs
99
+ python /fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py report --inp_dir /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16 --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/8_GPUS/dp-2_tp-4_pp-1_mbz-16 llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16 --commit-message "Upload llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16"
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/8_GPUS/dp-2_tp-4_pp-1_mbz-16/config.yaml ADDED
@@ -0,0 +1,90 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ general:
2
+ project: bench_cluster
3
+ seed: 42
4
+ model:
5
+ ddp_bucket_cap_mb: 25
6
+ dtype: bfloat16
7
+ init_method:
8
+ std: 0.025
9
+ make_vocab_size_divisible_by: 1
10
+ model_config:
11
+ bos_token_id: 1
12
+ eos_token_id: 2
13
+ hidden_act: silu
14
+ hidden_size: 2048
15
+ initializer_range: 0.02
16
+ intermediate_size: 4096
17
+ is_llama_config: true
18
+ max_position_embeddings: 4096
19
+ num_attention_heads: 32
20
+ num_hidden_layers: 24
21
+ num_key_value_heads: 32
22
+ pad_token_id: null
23
+ pretraining_tp: 1
24
+ rms_norm_eps: 1.0e-05
25
+ rope_scaling: null
26
+ rope_theta: 10000.0
27
+ tie_word_embeddings: true
28
+ use_cache: true
29
+ vocab_size: 50257
30
+ optimizer:
31
+ accumulate_grad_in_fp32: true
32
+ clip_grad: 1.0
33
+ learning_rate_scheduler:
34
+ learning_rate: 0.0001
35
+ lr_decay_style: linear
36
+ lr_warmup_style: linear
37
+ lr_warmup_steps: 1
38
+ min_decay_lr: 1.0e-05
39
+ optimizer_factory:
40
+ adam_beta1: 0.9
41
+ adam_beta2: 0.95
42
+ adam_eps: 1.0e-08
43
+ name: adamW
44
+ torch_adam_is_fused: true
45
+ weight_decay: 0.01
46
+ zero_stage: 1
47
+ parallelism:
48
+ dp: 2
49
+ expert_parallel_size: 1
50
+ pp: 1
51
+ pp_engine: 1f1b
52
+ tp: 4
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/8_GPUS/dp-2_tp-4_pp-1_mbz-16
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: 0
73
+ seed: 42
74
+ lighteval: null
75
+ tokens:
76
+ train_steps: 20
77
+ val_check_interval: -1
78
+ batch_accumulation_per_replica: 32
79
+ limit_test_batches: 0
80
+ limit_val_batches: 0
81
+ micro_batch_size: 16
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/8_GPUS/dp-2_tp-4_pp-1_mbz-16/log.out ADDED
@@ -0,0 +1,257 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ========================
2
+ START TIME: Wed Jul 3 22:42:51 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
+ W0703 22:42:58.927000 140687439050560 torch/distributed/run.py:757]
18
+ W0703 22:42:58.927000 140687439050560 torch/distributed/run.py:757] *****************************************
19
+ W0703 22:42:58.927000 140687439050560 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
+ W0703 22:42:58.927000 140687439050560 torch/distributed/run.py:757] *****************************************
21
+ [default0]:07/03/2024 22:43:20 [WARNING|DP=0|PP=0|TP=0|ip-26-0-162-233]: [Vocab Size Padding] Padded vocab (size: 50257) with 3 dummy tokens (new size: 50260)
22
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Config:
23
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Config(general=GeneralArgs(project='bench_cluster',
24
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: run='%date_%jobid',
25
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: seed=42,
26
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: step=None,
27
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: consumed_train_samples=None,
28
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: benchmark_csv_path=None,
29
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: ignore_sanity_checks=True),
30
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: parallelism=ParallelismArgs(dp=2,
31
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: pp=1,
32
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: tp=4,
33
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f4cf32dc670>,
34
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
35
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: tp_linear_async_communication=False,
36
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: expert_parallel_size=1),
37
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
38
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: eos_token_id=2,
39
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: hidden_act='silu',
40
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: hidden_size=2048,
41
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: initializer_range=0.02,
42
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: intermediate_size=4096,
43
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: is_llama_config=True,
44
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: max_position_embeddings=4096,
45
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: num_attention_heads=32,
46
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: num_hidden_layers=24,
47
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: num_key_value_heads=32,
48
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: pad_token_id=None,
49
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: pretraining_tp=1,
50
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: rms_norm_eps=1e-05,
51
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: rope_scaling=None,
52
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: rope_theta=10000.0,
53
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: tie_word_embeddings=True,
54
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: use_cache=True,
55
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: vocab_size=50260),
56
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: init_method=RandomInit(std=0.025),
57
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: dtype=torch.bfloat16,
58
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: make_vocab_size_divisible_by=1,
59
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: ddp_bucket_cap_mb=25),
60
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
61
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: tokenizer_revision=None,
62
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: tokenizer_max_length=None),
63
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
64
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: checkpoint_interval=100000,
65
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: save_initial_state=False,
66
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: resume_checkpoint_path=None,
67
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: checkpoints_path_is_shared_file_system=False),
68
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: logging=LoggingArgs(log_level='info',
69
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: log_level_replica='info',
70
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration_step_info_interval=1),
71
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: tokens=TokensArgs(sequence_length=4096,
72
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: train_steps=20,
73
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: micro_batch_size=16,
74
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: batch_accumulation_per_replica=32,
75
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: val_check_interval=-1,
76
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: limit_val_batches=0,
77
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: limit_test_batches=0),
78
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
79
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: adam_beta1=0.9,
80
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: adam_beta2=0.95,
81
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: torch_adam_is_fused=True,
82
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: name='adamW'),
83
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: zero_stage=1,
84
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: weight_decay=0.01,
85
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: clip_grad=1.0,
86
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: accumulate_grad_in_fp32=True,
87
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
88
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: lr_warmup_steps=1,
89
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: lr_warmup_style='linear',
90
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: lr_decay_style='linear',
91
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: lr_decay_steps=19,
92
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: lr_decay_starting_step=None,
93
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: min_decay_lr=1e-05)),
94
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: data_stages=[DatasetStageArgs(name='Training Stage',
95
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: start_training_step=1,
96
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
97
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: hf_dataset_splits='train',
98
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: hf_dataset_config_name=None,
99
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: dataset_processing_num_proc_per_process=64,
100
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: dataset_overwrite_cache=False,
101
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: text_column_name='text'),
102
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: seed=42,
103
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: num_loading_workers=0))],
104
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16')),
105
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: lighteval=None)
106
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Model Config:
107
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: LlamaConfig(bos_token_id=1,
108
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: eos_token_id=2,
109
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: hidden_act='silu',
110
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: hidden_size=2048,
111
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: initializer_range=0.02,
112
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: intermediate_size=4096,
113
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: is_llama_config=True,
114
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: max_position_embeddings=4096,
115
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: num_attention_heads=32,
116
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: num_hidden_layers=24,
117
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: num_key_value_heads=32,
118
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: pad_token_id=None,
119
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: pretraining_tp=1,
120
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: rms_norm_eps=1e-05,
121
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: rope_scaling=None,
122
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: rope_theta=10000.0,
123
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: tie_word_embeddings=True,
124
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: use_cache=True,
125
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: vocab_size=50260)
126
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Building model..
127
+ [default0]:07/03/2024 22:43:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Setting PP block ranks...
128
+ [default3]:07/03/2024 22:43:33 [INFO|DP=0|PP=0|TP=3|ip-26-0-162-233]: Local number of parameters: 277M (529.27MiB)
129
+ [default3]:07/03/2024 22:43:33 [INFO|DP=0|PP=0|TP=3|ip-26-0-162-233]: [After model building] Memory usage: 554.21MiB. Peak allocated: 606.24MiB Peak reserved: 608.00MiB
130
+ [default3]:07/03/2024 22:43:33 [INFO|DP=0|PP=0|TP=3|ip-26-0-162-233]: No checkpoint path provided.
131
+ [default0]:07/03/2024 22:43:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Total number of parameters: 1.11G (2117.09MiB)
132
+ [default0]:07/03/2024 22:43:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Local number of parameters: 277M (529.27MiB)
133
+ [default0]:07/03/2024 22:43:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: [After model building] Memory usage: 554.21MiB. Peak allocated: 606.24MiB Peak reserved: 608.00MiB
134
+ [default0]:07/03/2024 22:43:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: No checkpoint path provided.
135
+ [default0]:07/03/2024 22:43:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Parametrizing model parameters using StandardParametrizator
136
+ [default1]:07/03/2024 22:43:33 [INFO|DP=0|PP=0|TP=1|ip-26-0-162-233]: Local number of parameters: 277M (529.27MiB)
137
+ [default1]:07/03/2024 22:43:33 [INFO|DP=0|PP=0|TP=1|ip-26-0-162-233]: [After model building] Memory usage: 554.21MiB. Peak allocated: 606.24MiB Peak reserved: 608.00MiB
138
+ [default1]:07/03/2024 22:43:33 [INFO|DP=0|PP=0|TP=1|ip-26-0-162-233]: No checkpoint path provided.
139
+ [default2]:07/03/2024 22:43:33 [INFO|DP=0|PP=0|TP=2|ip-26-0-162-233]: Local number of parameters: 277M (529.27MiB)
140
+ [default2]:07/03/2024 22:43:33 [INFO|DP=0|PP=0|TP=2|ip-26-0-162-233]: [After model building] Memory usage: 554.21MiB. Peak allocated: 606.24MiB Peak reserved: 608.00MiB
141
+ [default2]:07/03/2024 22:43:33 [INFO|DP=0|PP=0|TP=2|ip-26-0-162-233]: No checkpoint path provided.
142
+ [default6]:07/03/2024 22:43:33 [INFO|DP=1|PP=0|TP=2|ip-26-0-162-233]: No checkpoint path provided.
143
+ [default4]:07/03/2024 22:43:33 [INFO|DP=1|PP=0|TP=0|ip-26-0-162-233]: No checkpoint path provided.
144
+ [default5]:07/03/2024 22:43:33 [INFO|DP=1|PP=0|TP=1|ip-26-0-162-233]: No checkpoint path provided.
145
+ [default7]:07/03/2024 22:43:33 [INFO|DP=1|PP=0|TP=3|ip-26-0-162-233]: No checkpoint path provided.
146
+ [default0]:07/03/2024 22:43:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: [Optimizer Building] Using LearningRateForSP as learning rate
147
+ [default0]:07/03/2024 22:43:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: [ZeRO sharding] Size of optimizer params per rank:
148
+ [default0]:07/03/2024 22:43:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: [ZeRO sharding] DP Rank 0 has 139M out of 277M (50.00%) params' optimizer states
149
+ [default0]:07/03/2024 22:43:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: [ZeRO sharding] DP Rank 1 has 139M out of 277M (50.00%) params' optimizer states
150
+ [default0]:07/03/2024 22:43:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
151
+ [default0]:07/03/2024 22:43:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Using `datasets` library
152
+ [default0]:07/03/2024 22:43:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
153
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
154
+ [default0]:07/03/2024 22:43:38 [WARNING|DP=0|PP=0|TP=0|ip-26-0-162-233]: Repo card metadata block was not found. Setting CardData to empty.
155
+ [default0]:07/03/2024 22:43:40 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: [Training Plan] There are 1 training stages
156
+ [default0]:07/03/2024 22:43:40 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: [Stage Training Stage] start from step 1
157
+ [default0]:07/03/2024 22:43:40 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]:
158
+ [default0]:07/03/2024 22:43:40 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: [Start training] datetime: 2024-07-03 22:43:40.213744 | mbs: 16 | grad_accum: 32 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
159
+ [default0]:07/03/2024 22:43:40 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
160
+ [default0]:07/03/2024 22:43:40 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 2142.76MiB. Peak allocated 2142.76MiB. Peak reserved: 2198.00MiB
161
+ [default4]:07/03/2024 22:43:40 [WARNING|DP=1|PP=0|TP=0|ip-26-0-162-233]: Repo card metadata block was not found. Setting CardData to empty.
162
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
163
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
164
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
165
+ [default5]:07/03/2024 22:43:40 [WARNING|DP=1|PP=0|TP=1|ip-26-0-162-233]: Repo card metadata block was not found. Setting CardData to empty.
166
+ [default2]:07/03/2024 22:43:40 [WARNING|DP=0|PP=0|TP=2|ip-26-0-162-233]: Repo card metadata block was not found. Setting CardData to empty.
167
+ [default7]:07/03/2024 22:43:40 [WARNING|DP=1|PP=0|TP=3|ip-26-0-162-233]: Repo card metadata block was not found. Setting CardData to empty.
168
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
169
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
170
+ [default6]:07/03/2024 22:43:40 [WARNING|DP=1|PP=0|TP=2|ip-26-0-162-233]: Repo card metadata block was not found. Setting CardData to empty.
171
+ [default3]:07/03/2024 22:43:40 [WARNING|DP=0|PP=0|TP=3|ip-26-0-162-233]: Repo card metadata block was not found. Setting CardData to empty.
172
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
173
+ [default1]:07/03/2024 22:43:40 [WARNING|DP=0|PP=0|TP=1|ip-26-0-162-233]: Repo card metadata block was not found. Setting CardData to empty.
174
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
175
+ [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.)
176
+ [default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
177
+ [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.)
178
+ [default1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
179
+ [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.)
180
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
181
+ [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.)
182
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
183
+ [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.)
184
+ [default7]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
185
+ [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.)
186
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
187
+ [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.)
188
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
189
+ [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.)
190
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
191
+ [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
192
+ [default5]: warnings.warn(
193
+ [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
194
+ [default2]: warnings.warn(
195
+ [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
196
+ [default0]:07/03/2024 22:44:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 2219.90MiB. Peak allocated 47278.37MiB. Peak reserved: 48996.00MiB
197
+ [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
198
+ [default7]: warnings.warn(
199
+ [default3]: warnings.warn(
200
+ [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
201
+ [default4]: warnings.warn(
202
+ [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
203
+ [default1]: warnings.warn(
204
+ [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
205
+ [default6]: warnings.warn(
206
+ [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
207
+ [default0]: warnings.warn(
208
+ [default0]:07/03/2024 22:44:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 1 / 20 | consumed_tokens: 4.19M | elapsed_time_per_iteration_ms: 23.8K | tokens_per_sec: 177K | tokens_per_sec_per_gpu: 22.1K | global_batch_size: 1.02K | lm_loss: 11.4 | lr: 0.0001 | model_tflops_per_gpu: 200 | hardware_tflops_per_gpu: 200 | grad_norm: 20.6 | cuda_memory_allocated: 3.44G | cuda_max_memory_reserved: 51.4G | hd_total_memory_tb: 312G | hd_used_memory_tb: 66.1G | hd_free_memory_tb: 246G
209
+ [default0]:07/03/2024 22:44:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 4603.54MiB. Peak reserved: 48996.00MiB
210
+ [default0]:07/03/2024 22:44:21 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.37MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
211
+ [default0]:07/03/2024 22:44:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 2 / 20 | consumed_tokens: 8.39M | elapsed_time_per_iteration_ms: 18.3K | tokens_per_sec: 230K | tokens_per_sec_per_gpu: 28.7K | global_batch_size: 1.02K | lm_loss: 11.4 | lr: 9.53e-05 | model_tflops_per_gpu: 261 | hardware_tflops_per_gpu: 261 | grad_norm: 20.7 | cuda_memory_allocated: 3.44G | cuda_max_memory_reserved: 51.4G | hd_total_memory_tb: 312G | hd_used_memory_tb: 66.1G | hd_free_memory_tb: 246G
212
+ [default0]:07/03/2024 22:44:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 4603.55MiB. Peak reserved: 48998.00MiB
213
+ [default0]:07/03/2024 22:44:40 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.37MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
214
+ [default0]:07/03/2024 22:44:40 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 3 / 20 | consumed_tokens: 12.6M | elapsed_time_per_iteration_ms: 18.2K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 28.8K | global_batch_size: 1.02K | lm_loss: 11.6 | lr: 9.05e-05 | model_tflops_per_gpu: 262 | hardware_tflops_per_gpu: 262 | grad_norm: 195 | cuda_memory_allocated: 3.44G | cuda_max_memory_reserved: 51.4G | hd_total_memory_tb: 312G | hd_used_memory_tb: 66.1G | hd_free_memory_tb: 246G
215
+ [default0]:07/03/2024 22:44:40 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 4603.55MiB. Peak reserved: 48998.00MiB
216
+ [default0]:STAGE:2024-07-03 22:44:40 1835168:1835168 ActivityProfilerController.cpp:314] Completed Stage: Warm Up
217
+ [default0]:07/03/2024 22:44:58 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.37MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
218
+ [default0]:07/03/2024 22:44:58 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 4 / 20 | consumed_tokens: 16.8M | elapsed_time_per_iteration_ms: 18.2K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 28.8K | global_batch_size: 1.02K | lm_loss: 13.6 | lr: 8.58e-05 | model_tflops_per_gpu: 262 | hardware_tflops_per_gpu: 262 | grad_norm: 28.1 | cuda_memory_allocated: 3.44G | cuda_max_memory_reserved: 51.4G | hd_total_memory_tb: 312G | hd_used_memory_tb: 66.1G | hd_free_memory_tb: 246G
219
+ [default0]:07/03/2024 22:44:58 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 4603.55MiB. Peak reserved: 48998.00MiB
220
+ [default0]:07/03/2024 22:45:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 5 / 20 | consumed_tokens: 21M | elapsed_time_per_iteration_ms: 18.1K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 28.9K | global_batch_size: 1.02K | lm_loss: 12 | lr: 8.11e-05 | model_tflops_per_gpu: 262 | hardware_tflops_per_gpu: 262 | grad_norm: 48.8
221
+ [default0]:07/03/2024 22:45:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
222
+ [default0]:07/03/2024 22:45:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 6 / 20 | consumed_tokens: 25.2M | elapsed_time_per_iteration_ms: 18.2K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 28.9K | global_batch_size: 1.02K | lm_loss: 10.9 | lr: 7.63e-05 | model_tflops_per_gpu: 262 | hardware_tflops_per_gpu: 262 | grad_norm: 19.7
223
+ [default0]:STAGE:2024-07-03 22:45:53 1835168:1835168 ActivityProfilerController.cpp:320] Completed Stage: Collection
224
+ [default0]:STAGE:2024-07-03 22:45:54 1835168:1835168 ActivityProfilerController.cpp:324] Completed Stage: Post Processing
225
+ [default0]:07/03/2024 22:48:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
226
+ [default0]:07/03/2024 22:48:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 7 / 20 | consumed_tokens: 29.4M | elapsed_time_per_iteration_ms: 18.1K | tokens_per_sec: 232K | tokens_per_sec_per_gpu: 29K | global_batch_size: 1.02K | lm_loss: 10.4 | lr: 7.16e-05 | model_tflops_per_gpu: 263 | hardware_tflops_per_gpu: 263 | grad_norm: 8.64
227
+ [default0]:07/03/2024 22:48:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
228
+ [default0]:07/03/2024 22:48:49 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 8 / 20 | consumed_tokens: 33.6M | elapsed_time_per_iteration_ms: 18.1K | tokens_per_sec: 232K | tokens_per_sec_per_gpu: 28.9K | global_batch_size: 1.02K | lm_loss: 9.66 | lr: 6.68e-05 | model_tflops_per_gpu: 263 | hardware_tflops_per_gpu: 263 | grad_norm: 6.86
229
+ [default0]:07/03/2024 22:48:49 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
230
+ [default0]:07/03/2024 22:49:07 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 9 / 20 | consumed_tokens: 37.7M | elapsed_time_per_iteration_ms: 18.1K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 28.9K | global_batch_size: 1.02K | lm_loss: 11.2 | lr: 6.21e-05 | model_tflops_per_gpu: 262 | hardware_tflops_per_gpu: 262 | grad_norm: 52.7
231
+ [default0]:07/03/2024 22:49:07 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
232
+ [default0]:07/03/2024 22:49:25 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 10 / 20 | consumed_tokens: 41.9M | elapsed_time_per_iteration_ms: 18.2K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 28.9K | global_batch_size: 1.02K | lm_loss: 9.08 | lr: 5.74e-05 | model_tflops_per_gpu: 262 | hardware_tflops_per_gpu: 262 | grad_norm: 15.1
233
+ [default0]:07/03/2024 22:49:25 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
234
+ [default0]:07/03/2024 22:49:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 11 / 20 | consumed_tokens: 46.1M | elapsed_time_per_iteration_ms: 18.2K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 28.9K | global_batch_size: 1.02K | lm_loss: 8.54 | lr: 5.26e-05 | model_tflops_per_gpu: 262 | hardware_tflops_per_gpu: 262 | grad_norm: 6.74
235
+ [default0]:07/03/2024 22:49:43 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
236
+ [default0]:07/03/2024 22:50:02 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 12 / 20 | consumed_tokens: 50.3M | elapsed_time_per_iteration_ms: 18.1K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 28.9K | global_batch_size: 1.02K | lm_loss: 8.34 | lr: 4.79e-05 | model_tflops_per_gpu: 262 | hardware_tflops_per_gpu: 262 | grad_norm: 5.79
237
+ [default0]:07/03/2024 22:50:02 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
238
+ [default0]:07/03/2024 22:50:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 13 / 20 | consumed_tokens: 54.5M | elapsed_time_per_iteration_ms: 18.2K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 28.9K | global_batch_size: 1.02K | lm_loss: 8.11 | lr: 4.32e-05 | model_tflops_per_gpu: 262 | hardware_tflops_per_gpu: 262 | grad_norm: 5.58
239
+ [default0]:07/03/2024 22:50:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
240
+ [default0]:07/03/2024 22:50:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 14 / 20 | consumed_tokens: 58.7M | elapsed_time_per_iteration_ms: 18.2K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 28.9K | global_batch_size: 1.02K | lm_loss: 7.86 | lr: 3.84e-05 | model_tflops_per_gpu: 262 | hardware_tflops_per_gpu: 262 | grad_norm: 5.35
241
+ [default0]:07/03/2024 22:50:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
242
+ [default0]:07/03/2024 22:50:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 15 / 20 | consumed_tokens: 62.9M | elapsed_time_per_iteration_ms: 18.2K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 28.9K | global_batch_size: 1.02K | lm_loss: 7.64 | lr: 3.37e-05 | model_tflops_per_gpu: 262 | hardware_tflops_per_gpu: 262 | grad_norm: 4.87
243
+ [default0]:07/03/2024 22:50:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
244
+ [default0]:07/03/2024 22:51:14 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 16 / 20 | consumed_tokens: 67.1M | elapsed_time_per_iteration_ms: 18.2K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 28.9K | global_batch_size: 1.02K | lm_loss: 7.53 | lr: 2.89e-05 | model_tflops_per_gpu: 262 | hardware_tflops_per_gpu: 262 | grad_norm: 5.02
245
+ [default0]:07/03/2024 22:51:14 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
246
+ [default0]:07/03/2024 22:51:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 17 / 20 | consumed_tokens: 71.3M | elapsed_time_per_iteration_ms: 18.2K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 28.9K | global_batch_size: 1.02K | lm_loss: 7.46 | lr: 2.42e-05 | model_tflops_per_gpu: 262 | hardware_tflops_per_gpu: 262 | grad_norm: 5.63
247
+ [default0]:07/03/2024 22:51:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
248
+ [default0]:07/03/2024 22:51:51 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 18 / 20 | consumed_tokens: 75.5M | elapsed_time_per_iteration_ms: 18.2K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 28.9K | global_batch_size: 1.02K | lm_loss: 7.33 | lr: 1.95e-05 | model_tflops_per_gpu: 262 | hardware_tflops_per_gpu: 262 | grad_norm: 5.41
249
+ [default0]:07/03/2024 22:51:51 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
250
+ [default0]:07/03/2024 22:52:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 19 / 20 | consumed_tokens: 79.7M | elapsed_time_per_iteration_ms: 18.2K | tokens_per_sec: 230K | tokens_per_sec_per_gpu: 28.7K | global_batch_size: 1.02K | lm_loss: 7.19 | lr: 1.47e-05 | model_tflops_per_gpu: 261 | hardware_tflops_per_gpu: 261 | grad_norm: 3.12
251
+ [default0]:07/03/2024 22:52:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: Memory usage: 3280.35MiB. Peak allocated 48338.84MiB. Peak reserved: 48998.00MiB
252
+ [default0]:07/03/2024 22:52:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-162-233]: iteration: 20 / 20 | consumed_tokens: 83.9M | elapsed_time_per_iteration_ms: 18.3K | tokens_per_sec: 229K | tokens_per_sec_per_gpu: 28.6K | global_batch_size: 1.02K | lm_loss: 7.14 | lr: 1e-05 | model_tflops_per_gpu: 260 | hardware_tflops_per_gpu: 260 | grad_norm: 3.66
253
+ Saved 1 csv files over 1 completed logs
254
+ Processing file: /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/profiler/ip-26-0-162-233_1835168.1720046865197689959.pt.trace.json
255
+ Results written to /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/profiler.csv
256
+ 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.
257
+
llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/log_metrics.csv ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ iteration,consumed_tokens,elapsed_time_per_iteration_ms,tokens_per_sec,tokens_per_sec_per_gpu,global_batch_size,lm_loss,lr,model_tflops_per_gpu,hardware_tflops_per_gpu,grad_norm,memory_usage_MiB,peak_allocated_MiB,peak_reserved_MiB
2
+ 1,4190000.0000000005,23800.0,177000.0,22100.0,1020.0,11.4,0.0001,200.0,200.0,20.6,3280.37,48338.84,48998.0
3
+ 2,8390000.0,18300.0,230000.0,28700.0,1020.0,11.4,9.53e-05,261.0,261.0,20.7,3280.37,48338.84,48998.0
4
+ 3,12600000.0,18200.0,231000.0,28800.0,1020.0,11.6,9.05e-05,262.0,262.0,195.0,3280.37,48338.84,48998.0
5
+ 4,16800000.0,18200.0,231000.0,28800.0,1020.0,13.6,8.58e-05,262.0,262.0,28.1,3280.35,4603.55,48998.0
6
+ 5,21000000.0,18100.0,231000.0,28900.0,1020.0,12.0,8.11e-05,262.0,262.0,48.8,3280.35,48338.84,48998.0
7
+ 6,25200000.0,18200.0,231000.0,28900.0,1020.0,10.9,7.63e-05,262.0,262.0,19.7,3280.35,48338.84,48998.0
8
+ 7,29400000.0,18100.0,232000.0,29000.0,1020.0,10.4,7.16e-05,263.0,263.0,8.64,3280.35,48338.84,48998.0
9
+ 8,33600000.0,18100.0,232000.0,28900.0,1020.0,9.66,6.68e-05,263.0,263.0,6.86,3280.35,48338.84,48998.0
10
+ 9,37700000.0,18100.0,231000.0,28900.0,1020.0,11.2,6.21e-05,262.0,262.0,52.7,3280.35,48338.84,48998.0
11
+ 10,41900000.0,18200.0,231000.0,28900.0,1020.0,9.08,5.74e-05,262.0,262.0,15.1,3280.35,48338.84,48998.0
12
+ 11,46100000.0,18200.0,231000.0,28900.0,1020.0,8.54,5.26e-05,262.0,262.0,6.74,3280.35,48338.84,48998.0
13
+ 12,50300000.0,18100.0,231000.0,28900.0,1020.0,8.34,4.79e-05,262.0,262.0,5.79,3280.35,48338.84,48998.0
14
+ 13,54500000.0,18200.0,231000.0,28900.0,1020.0,8.11,4.32e-05,262.0,262.0,5.58,3280.35,48338.84,48998.0
15
+ 14,58700000.0,18200.0,231000.0,28900.0,1020.0,7.86,3.84e-05,262.0,262.0,5.35,3280.35,48338.84,48998.0
16
+ 15,62900000.0,18200.0,231000.0,28900.0,1020.0,7.64,3.37e-05,262.0,262.0,4.87,3280.35,48338.84,48998.0
17
+ 16,67099999.99999999,18200.0,231000.0,28900.0,1020.0,7.53,2.89e-05,262.0,262.0,5.02,3280.35,48338.84,48998.0
18
+ 17,71300000.0,18200.0,231000.0,28900.0,1020.0,7.46,2.42e-05,262.0,262.0,5.63,3280.35,48338.84,48998.0
19
+ 18,75500000.0,18200.0,231000.0,28900.0,1020.0,7.33,1.95e-05,262.0,262.0,5.41,3280.35,48338.84,48998.0
20
+ 19,79700000.0,18200.0,230000.0,28700.0,1020.0,7.19,1.47e-05,261.0,261.0,3.12,3280.35,48338.84,48998.0
21
+ 20,83900000.0,18300.0,229000.0,28600.0,1020.0,7.14,1e-05,260.0,260.0,3.66,,,
llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/profiler.csv ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ forward,backward
2
+ 1ms 91μs,0ms 486μs
llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/profiler/ip-26-0-162-233_1835168.1720046865197689959.pt.trace.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fc55f1ebfea96bdde4b53832ea4f4ffd313f44c4b3ad68e2d743605f2a404317
3
+ size 4562131238
llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-16/status.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ completed