Hugging Face's logo

nanotron
/
old_bench

Model card Files Community
3outeille HF staff commited on
Commit
bb1ca7c
•
1 Parent(s): a921e7d

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

Browse files
Files changed (8) hide show
  1. .gitattributes +1 -0
  2. llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/bench.slurm +111 -0
  3. llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/config.yaml +90 -0
  4. llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/log.out +253 -0
  5. llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/log_metrics.csv +21 -0
  6. llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/profiler.csv +2 -0
  7. llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/profiler/ip-26-0-160-225_333932.1720042364135456939.pt.trace.json +3 -0
  8. llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/status.txt +1 -0
.gitattributes CHANGED
@@ -104,3 +104,4 @@ llama-1B/8_GPUS/dp-1_tp-2_pp-4_mbz-8/profiler/ip-26-0-164-207_582119.17200400690
104
  llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-8/profiler/ip-26-0-162-233_1808546.1720040095324285932.pt.trace.json filter=lfs diff=lfs merge=lfs -text
105
  llama-1B/8_GPUS/dp-1_tp-4_pp-2_mbz-8/profiler/ip-26-0-164-207_591039.1720041521624478261.pt.trace.json filter=lfs diff=lfs merge=lfs -text
106
  llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/profiler/ip-26-0-163-220_391612.1720042520675458251.pt.trace.json filter=lfs diff=lfs merge=lfs -text
 
 
104
  llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-8/profiler/ip-26-0-162-233_1808546.1720040095324285932.pt.trace.json filter=lfs diff=lfs merge=lfs -text
105
  llama-1B/8_GPUS/dp-1_tp-4_pp-2_mbz-8/profiler/ip-26-0-164-207_591039.1720041521624478261.pt.trace.json filter=lfs diff=lfs merge=lfs -text
106
  llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/profiler/ip-26-0-163-220_391612.1720042520675458251.pt.trace.json filter=lfs diff=lfs merge=lfs -text
107
+ llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/profiler/ip-26-0-160-225_333932.1720042364135456939.pt.trace.json filter=lfs diff=lfs merge=lfs -text
llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/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-1_tp-1_pp-8_mbz-4/log.out
13
+ #SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/log.out
14
+
15
+ # Function to update status based on squeue output
16
+ update_status() {
17
+ job_id=$1
18
+ status_file=$2
19
+ # For unknown reasons, it doenst update status for pending. It only works for running
20
+ while true; do
21
+ job_status=$(squeue --job $job_id --noheader --format=%T)
22
+ echo "Job status: $job_status"
23
+ if [ -z "$job_status" ]; then
24
+ # Job has finished or is not found
25
+ break
26
+ elif [ "$job_status" = "RUNNING" ]; then
27
+ printf "running" > $status_file
28
+ break
29
+ fi
30
+ sleep 10
31
+ done
32
+ }
33
+
34
+ # Misc initializations.
35
+ echo "========================"
36
+ echo "START TIME: $(date)"
37
+ source /fsx/ferdinandmom/miniforge3/etc/profile.d/conda.sh
38
+ conda activate /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster
39
+ echo python3 version = $(python3 --version)
40
+ echo "========================"
41
+
42
+ # Slurm stuff
43
+ export HOSTNAMES=$(scontrol show hostnames "$SLURM_JOB_NODELIST")
44
+ export MASTER_ADDR=$(scontrol show hostnames "$SLURM_JOB_NODELIST" | head -n 1)
45
+ export MASTER_PORT=$((1024 + RANDOM % 64511))
46
+
47
+ export TMPDIR=/scratch
48
+ export HF_DATASETS_CACHE="/admin/home/ferdinand_mom/.cache"
49
+ export CUBLAS_WORKSPACE_CONFIG=":4096:8"
50
+ export CUDA_DEVICE_MAX_CONNECTIONS="1"
51
+
52
+ huggingface-cli login --token $HUGGINGFACE_TOKEN
53
+
54
+
55
+ NANOTRON_REPO="/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron"
56
+ CMD="$NANOTRON_REPO/run_train.py --config-file /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/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-1_tp-1_pp-8_mbz-4/status.txt &
76
+
77
+ # Run the main command
78
+ srun -u $LAUNCHER $CMD
79
+ exit_status=$?
80
+
81
+ # Update status based on the exit status of `srun`
82
+ if [ $exit_status -eq 0 ]; then
83
+ printf "completed" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/status.txt
84
+ else
85
+ if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/log.out; then
86
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/status.txt
87
+ elif grep -q " CUDA error: an illegal memory access" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/log.out; then
88
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/status.txt
89
+ elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/log.out; then
90
+ printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/status.txt
91
+ else
92
+ printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/status.txt
93
+ fi
94
+ fi
95
+
96
+ # Run the report script if the job completed successfully
97
+ if [ $exit_status -eq 0 ]; then
98
+ python /fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py report --inp_dir /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4 --is_logs
99
+ python /fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py report --inp_dir /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4 --is_profiler
100
+ fi
101
+
102
+
103
+ # Push to hub the folder using huggingface_cli
104
+ huggingface-cli upload nanotron/bench_cluster /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4 llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4 --commit-message "Upload llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4"
105
+
106
+ # Verify the upload
107
+ if [ $? -eq 0 ]; then
108
+ echo "Uploading to Huggingface Hub successful"
109
+ else
110
+ echo "Failed to upload to Huggingface Hub"
111
+ fi
llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/config.yaml ADDED
@@ -0,0 +1,90 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ general:
2
+ project: bench_cluster
3
+ seed: 42
4
+ model:
5
+ ddp_bucket_cap_mb: 25
6
+ dtype: bfloat16
7
+ init_method:
8
+ std: 0.025
9
+ make_vocab_size_divisible_by: 1
10
+ model_config:
11
+ bos_token_id: 1
12
+ eos_token_id: 2
13
+ hidden_act: silu
14
+ hidden_size: 2048
15
+ initializer_range: 0.02
16
+ intermediate_size: 4096
17
+ is_llama_config: true
18
+ max_position_embeddings: 4096
19
+ num_attention_heads: 32
20
+ num_hidden_layers: 24
21
+ num_key_value_heads: 32
22
+ pad_token_id: null
23
+ pretraining_tp: 1
24
+ rms_norm_eps: 1.0e-05
25
+ rope_scaling: null
26
+ rope_theta: 10000.0
27
+ tie_word_embeddings: true
28
+ use_cache: true
29
+ vocab_size: 50257
30
+ optimizer:
31
+ accumulate_grad_in_fp32: true
32
+ clip_grad: 1.0
33
+ learning_rate_scheduler:
34
+ learning_rate: 0.0001
35
+ lr_decay_style: linear
36
+ lr_warmup_style: linear
37
+ lr_warmup_steps: 1
38
+ min_decay_lr: 1.0e-05
39
+ optimizer_factory:
40
+ adam_beta1: 0.9
41
+ adam_beta2: 0.95
42
+ adam_eps: 1.0e-08
43
+ name: adamW
44
+ torch_adam_is_fused: true
45
+ weight_decay: 0.01
46
+ zero_stage: 1
47
+ parallelism:
48
+ dp: 1
49
+ expert_parallel_size: 1
50
+ pp: 8
51
+ pp_engine: 1f1b
52
+ tp: 1
53
+ tp_linear_async_communication: false
54
+ tp_mode: REDUCE_SCATTER
55
+ profiler:
56
+ profiler_export_path: /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4
57
+ tokenizer:
58
+ tokenizer_max_length: null
59
+ tokenizer_name_or_path: openai-community/gpt2
60
+ tokenizer_revision: null
61
+ data_stages:
62
+ - name: Training Stage
63
+ start_training_step: 1
64
+ data:
65
+ dataset:
66
+ dataset_overwrite_cache: false
67
+ dataset_processing_num_proc_per_process: 64
68
+ hf_dataset_config_name: null
69
+ hf_dataset_or_datasets: roneneldan/TinyStories
70
+ hf_dataset_splits: train
71
+ text_column_name: text
72
+ num_loading_workers: 0
73
+ seed: 42
74
+ lighteval: null
75
+ tokens:
76
+ train_steps: 20
77
+ val_check_interval: -1
78
+ batch_accumulation_per_replica: 256
79
+ limit_test_batches: 0
80
+ limit_val_batches: 0
81
+ micro_batch_size: 4
82
+ sequence_length: 4096
83
+ logging:
84
+ iteration_step_info_interval: 1
85
+ log_level: info
86
+ log_level_replica: info
87
+ checkpoints:
88
+ checkpoint_interval: 100000
89
+ checkpoints_path: /dev/null
90
+ resume_checkpoint_path: null
llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/log.out ADDED
@@ -0,0 +1,253 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ========================
2
+ START TIME: Wed Jul 3 21:25:49 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 21:25:56.666000 140669559994176 torch/distributed/run.py:757]
18
+ W0703 21:25:56.666000 140669559994176 torch/distributed/run.py:757] *****************************************
19
+ W0703 21:25:56.666000 140669559994176 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 21:25:56.666000 140669559994176 torch/distributed/run.py:757] *****************************************
21
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Config:
22
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Config(general=GeneralArgs(project='bench_cluster',
23
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: run='%date_%jobid',
24
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: seed=42,
25
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: step=None,
26
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: consumed_train_samples=None,
27
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: benchmark_csv_path=None,
28
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: ignore_sanity_checks=True),
29
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: parallelism=ParallelismArgs(dp=1,
30
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pp=8,
31
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tp=1,
32
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7ff7b36f08e0>,
33
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
34
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tp_linear_async_communication=False,
35
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: expert_parallel_size=1),
36
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
37
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: eos_token_id=2,
38
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hidden_act='silu',
39
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hidden_size=2048,
40
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: initializer_range=0.02,
41
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: intermediate_size=4096,
42
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: is_llama_config=True,
43
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: max_position_embeddings=4096,
44
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_attention_heads=32,
45
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_hidden_layers=24,
46
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_key_value_heads=32,
47
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pad_token_id=None,
48
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pretraining_tp=1,
49
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rms_norm_eps=1e-05,
50
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rope_scaling=None,
51
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rope_theta=10000.0,
52
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tie_word_embeddings=True,
53
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: use_cache=True,
54
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: vocab_size=50257),
55
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: init_method=RandomInit(std=0.025),
56
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: dtype=torch.bfloat16,
57
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: make_vocab_size_divisible_by=1,
58
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: ddp_bucket_cap_mb=25),
59
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
60
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tokenizer_revision=None,
61
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tokenizer_max_length=None),
62
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
63
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: checkpoint_interval=100000,
64
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: save_initial_state=False,
65
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: resume_checkpoint_path=None,
66
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: checkpoints_path_is_shared_file_system=False),
67
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: logging=LoggingArgs(log_level='info',
68
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: log_level_replica='info',
69
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: iteration_step_info_interval=1),
70
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tokens=TokensArgs(sequence_length=4096,
71
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: train_steps=20,
72
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: micro_batch_size=4,
73
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: batch_accumulation_per_replica=256,
74
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: val_check_interval=-1,
75
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: limit_val_batches=0,
76
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: limit_test_batches=0),
77
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
78
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: adam_beta1=0.9,
79
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: adam_beta2=0.95,
80
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: torch_adam_is_fused=True,
81
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: name='adamW'),
82
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: zero_stage=1,
83
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: weight_decay=0.01,
84
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: clip_grad=1.0,
85
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: accumulate_grad_in_fp32=True,
86
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
87
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lr_warmup_steps=1,
88
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lr_warmup_style='linear',
89
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lr_decay_style='linear',
90
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lr_decay_steps=19,
91
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lr_decay_starting_step=None,
92
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: min_decay_lr=1e-05)),
93
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: data_stages=[DatasetStageArgs(name='Training Stage',
94
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: start_training_step=1,
95
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
96
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hf_dataset_splits='train',
97
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hf_dataset_config_name=None,
98
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: dataset_processing_num_proc_per_process=64,
99
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: dataset_overwrite_cache=False,
100
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: text_column_name='text'),
101
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: seed=42,
102
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_loading_workers=0))],
103
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4')),
104
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lighteval=None)
105
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Model Config:
106
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: LlamaConfig(bos_token_id=1,
107
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: eos_token_id=2,
108
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hidden_act='silu',
109
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hidden_size=2048,
110
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: initializer_range=0.02,
111
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: intermediate_size=4096,
112
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: is_llama_config=True,
113
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: max_position_embeddings=4096,
114
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_attention_heads=32,
115
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_hidden_layers=24,
116
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_key_value_heads=32,
117
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pad_token_id=None,
118
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pretraining_tp=1,
119
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rms_norm_eps=1e-05,
120
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rope_scaling=None,
121
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rope_theta=10000.0,
122
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tie_word_embeddings=True,
123
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: use_cache=True,
124
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: vocab_size=50257)
125
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Building model..
126
+ [default0]:07/03/2024 21:26:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Setting PP block ranks...
127
+ [default4]:07/03/2024 21:26:34 [INFO|DP=0|PP=4|TP=0|ip-26-0-160-225]: Local number of parameters: 126M (240.02MiB)
128
+ [default4]:07/03/2024 21:26:34 [INFO|DP=0|PP=4|TP=0|ip-26-0-160-225]: [After model building] Memory usage: 243.03MiB. Peak allocated: 245.06MiB Peak reserved: 262.00MiB
129
+ [default4]:07/03/2024 21:26:34 [INFO|DP=0|PP=4|TP=0|ip-26-0-160-225]: No checkpoint path provided.
130
+ [default3]:07/03/2024 21:26:34 [INFO|DP=0|PP=3|TP=0|ip-26-0-160-225]: Local number of parameters: 168M (320.03MiB)
131
+ [default3]:07/03/2024 21:26:34 [INFO|DP=0|PP=3|TP=0|ip-26-0-160-225]: [After model building] Memory usage: 324.04MiB. Peak allocated: 326.07MiB Peak reserved: 336.00MiB
132
+ [default3]:07/03/2024 21:26:34 [INFO|DP=0|PP=3|TP=0|ip-26-0-160-225]: No checkpoint path provided.
133
+ [default0]:07/03/2024 21:26:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Total number of parameters: 1.21G (2312.82MiB)
134
+ [default0]:07/03/2024 21:26:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Local number of parameters: 271M (516.35MiB)
135
+ [default2]:07/03/2024 21:26:34 [INFO|DP=0|PP=2|TP=0|ip-26-0-160-225]: Local number of parameters: 126M (240.02MiB)
136
+ [default2]:07/03/2024 21:26:34 [INFO|DP=0|PP=2|TP=0|ip-26-0-160-225]: [After model building] Memory usage: 243.03MiB. Peak allocated: 245.06MiB Peak reserved: 262.00MiB
137
+ [default2]:07/03/2024 21:26:34 [INFO|DP=0|PP=2|TP=0|ip-26-0-160-225]: No checkpoint path provided.
138
+ [default0]:07/03/2024 21:26:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [After model building] Memory usage: 520.36MiB. Peak allocated: 522.39MiB Peak reserved: 534.00MiB
139
+ [default0]:07/03/2024 21:26:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: No checkpoint path provided.
140
+ [default0]:07/03/2024 21:26:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Parametrizing model parameters using StandardParametrizator
141
+ [default5]:07/03/2024 21:26:34 [INFO|DP=0|PP=5|TP=0|ip-26-0-160-225]: Local number of parameters: 126M (240.02MiB)
142
+ [default5]:07/03/2024 21:26:34 [INFO|DP=0|PP=5|TP=0|ip-26-0-160-225]: [After model building] Memory usage: 243.03MiB. Peak allocated: 245.06MiB Peak reserved: 262.00MiB
143
+ [default5]:07/03/2024 21:26:34 [INFO|DP=0|PP=5|TP=0|ip-26-0-160-225]: No checkpoint path provided.
144
+ [default1]:07/03/2024 21:26:34 [INFO|DP=0|PP=1|TP=0|ip-26-0-160-225]: Local number of parameters: 126M (240.02MiB)
145
+ [default1]:07/03/2024 21:26:34 [INFO|DP=0|PP=1|TP=0|ip-26-0-160-225]: [After model building] Memory usage: 243.03MiB. Peak allocated: 245.06MiB Peak reserved: 262.00MiB
146
+ [default1]:07/03/2024 21:26:34 [INFO|DP=0|PP=1|TP=0|ip-26-0-160-225]: No checkpoint path provided.
147
+ [default6]:07/03/2024 21:26:34 [INFO|DP=0|PP=6|TP=0|ip-26-0-160-225]: Local number of parameters: 168M (320.03MiB)
148
+ [default6]:07/03/2024 21:26:34 [INFO|DP=0|PP=6|TP=0|ip-26-0-160-225]: [After model building] Memory usage: 324.04MiB. Peak allocated: 326.07MiB Peak reserved: 336.00MiB
149
+ [default6]:07/03/2024 21:26:34 [INFO|DP=0|PP=6|TP=0|ip-26-0-160-225]: No checkpoint path provided.
150
+ [default7]:07/03/2024 21:26:34 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: Local number of parameters: 103M (196.32MiB)
151
+ [default7]:07/03/2024 21:26:34 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: [After model building] Memory usage: 196.33MiB. Peak allocated: 196.33MiB Peak reserved: 200.00MiB
152
+ [default7]:07/03/2024 21:26:34 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: No checkpoint path provided.
153
+ [default0]:07/03/2024 21:26:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Optimizer Building] Using LearningRateForSP as learning rate
154
+ [default0]:07/03/2024 21:26:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [ZeRO sharding] Size of optimizer params per rank:
155
+ [default0]:07/03/2024 21:26:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [ZeRO sharding] DP Rank 0 has 271M out of 271M (100.00%) params' optimizer states
156
+ [default0]:07/03/2024 21:26:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
157
+ [default0]:07/03/2024 21:26:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Using `datasets` library
158
+ [default0]:07/03/2024 21:26:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
159
+ [default0]:07/03/2024 21:26:36 [WARNING|DP=0|PP=0|TP=0|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
160
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
161
+ [default0]:07/03/2024 21:26:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Training Plan] There are 1 training stages
162
+ [default0]:07/03/2024 21:26:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Stage Training Stage] start from step 1
163
+ [default0]:07/03/2024 21:26:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]:
164
+ [default0]:07/03/2024 21:26:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Start training] datetime: 2024-07-03 21:26:38.472195 | mbs: 4 | grad_accum: 256 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
165
+ [default0]:07/03/2024 21:26:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
166
+ [default0]:07/03/2024 21:26:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 2585.75MiB. Peak allocated 2585.75MiB. Peak reserved: 2602.00MiB
167
+ [default4]:07/03/2024 21:26:38 [WARNING|DP=0|PP=4|TP=0|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
168
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
169
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
170
+ [default2]:07/03/2024 21:26:38 [WARNING|DP=0|PP=2|TP=0|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
171
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
172
+ [default3]:07/03/2024 21:26:38 [WARNING|DP=0|PP=3|TP=0|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
173
+ [default6]:07/03/2024 21:26:38 [WARNING|DP=0|PP=6|TP=0|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
174
+ [default5]:07/03/2024 21:26:38 [WARNING|DP=0|PP=5|TP=0|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
175
+ [default1]:07/03/2024 21:26:38 [WARNING|DP=0|PP=1|TP=0|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
176
+ [default7]:07/03/2024 21:26:38 [WARNING|DP=0|PP=7|TP=0|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
177
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
178
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
179
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
180
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
181
+ [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.)
182
+ [default7]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
183
+ [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.)
184
+ [default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
185
+ [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.)
186
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
187
+ [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.)
188
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
189
+ [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.)
190
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
191
+ [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.)
192
+ [default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
193
+ [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.)
194
+ [default1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
195
+ [default0]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: Attempting to run cuBLAS, but there was no current CUDA context! Attempting to set the primary context... (Triggered internally at ../aten/src/ATen/cuda/CublasHandlePool.cpp:135.)
196
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
197
+ [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.)
198
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
199
+ [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
200
+ [default7]: warnings.warn(
201
+ [default0]:07/03/2024 21:27:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 2651.89MiB. Peak allocated 41968.83MiB. Peak reserved: 42212.00MiB
202
+ [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
203
+ [default0]: warnings.warn(
204
+ [default7]:07/03/2024 21:27:33 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 1 / 20 | consumed_tokens: 4.19M | elapsed_time_per_iteration_ms: 54.2K | tokens_per_sec: 77.4K | tokens_per_sec_per_gpu: 9.67K | global_batch_size: 1.02K | lm_loss: 11.1 | lr: 0.0001 | model_tflops_per_gpu: 87.8 | hardware_tflops_per_gpu: 87.8 | grad_norm: 24.9 | cuda_memory_allocated: 1.92G | cuda_max_memory_reserved: 13G | hd_total_memory_tb: 312G | hd_used_memory_tb: 66.5G | hd_free_memory_tb: 246G
205
+ [default0]:07/03/2024 21:27:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 4717.29MiB. Peak reserved: 43000.00MiB
206
+ [default0]:07/03/2024 21:28:02 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
207
+ [default0]:07/03/2024 21:28:02 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 4717.31MiB. Peak reserved: 44472.00MiB
208
+ [default7]:07/03/2024 21:28:02 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 2 / 20 | consumed_tokens: 8.39M | elapsed_time_per_iteration_ms: 29.2K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 17.9K | global_batch_size: 1.02K | lm_loss: 11.1 | lr: 9.53e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 25.1 | cuda_memory_allocated: 1.92G | cuda_max_memory_reserved: 13G | hd_total_memory_tb: 312G | hd_used_memory_tb: 66.5G | hd_free_memory_tb: 246G
209
+ [default0]:07/03/2024 21:28:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
210
+ [default0]:07/03/2024 21:28:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 4717.31MiB. Peak reserved: 44472.00MiB
211
+ [default7]:07/03/2024 21:28:32 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 3 / 20 | consumed_tokens: 12.6M | elapsed_time_per_iteration_ms: 29.3K | tokens_per_sec: 143K | tokens_per_sec_per_gpu: 17.9K | global_batch_size: 1.02K | lm_loss: 9.49 | lr: 9.05e-05 | model_tflops_per_gpu: 162 | hardware_tflops_per_gpu: 162 | grad_norm: 21.5 | cuda_memory_allocated: 1.92G | cuda_max_memory_reserved: 13G | hd_total_memory_tb: 312G | hd_used_memory_tb: 66.5G | hd_free_memory_tb: 246G
212
+ [default0]:STAGE:2024-07-03 21:28:32 333932:333932 ActivityProfilerController.cpp:314] Completed Stage: Warm Up
213
+ [default0]:07/03/2024 21:29:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
214
+ [default0]:07/03/2024 21:29:01 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 4717.31MiB. Peak reserved: 44472.00MiB
215
+ [default7]:07/03/2024 21:29:01 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 4 / 20 | consumed_tokens: 16.8M | elapsed_time_per_iteration_ms: 29.2K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 9.36 | lr: 8.58e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 21.4 | cuda_memory_allocated: 1.92G | cuda_max_memory_reserved: 13G | hd_total_memory_tb: 312G | hd_used_memory_tb: 66.5G | hd_free_memory_tb: 246G
216
+ [default0]:07/03/2024 21:29:30 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
217
+ [default7]:07/03/2024 21:29:30 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 5 / 20 | consumed_tokens: 21M | elapsed_time_per_iteration_ms: 29.2K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 9.02 | lr: 8.11e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 12.7
218
+ [default7]:07/03/2024 21:29:59 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 6 / 20 | consumed_tokens: 25.2M | elapsed_time_per_iteration_ms: 29.1K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 10.3 | lr: 7.63e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 47.1
219
+ [default0]:STAGE:2024-07-03 21:30:24 333932:333932 ActivityProfilerController.cpp:320] Completed Stage: Collection
220
+ [default0]:STAGE:2024-07-03 21:30:26 333932:333932 ActivityProfilerController.cpp:324] Completed Stage: Post Processing
221
+ [default0]:07/03/2024 21:33:21 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
222
+ [default0]:07/03/2024 21:33:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
223
+ [default7]:07/03/2024 21:33:50 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 7 / 20 | consumed_tokens: 29.4M | elapsed_time_per_iteration_ms: 231K | tokens_per_sec: 18.1K | tokens_per_sec_per_gpu: 2.27K | global_batch_size: 1.02K | lm_loss: 8.68 | lr: 7.16e-05 | model_tflops_per_gpu: 20.6 | hardware_tflops_per_gpu: 20.6 | grad_norm: 5.58
224
+ [default7]:07/03/2024 21:34:20 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 8 / 20 | consumed_tokens: 33.6M | elapsed_time_per_iteration_ms: 29.2K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 8.32 | lr: 6.68e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 4.78
225
+ [default0]:07/03/2024 21:34:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
226
+ [default0]:07/03/2024 21:34:49 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
227
+ [default7]:07/03/2024 21:34:49 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 9 / 20 | consumed_tokens: 37.7M | elapsed_time_per_iteration_ms: 29.2K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 7.95 | lr: 6.21e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 3.31
228
+ [default0]:07/03/2024 21:35:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
229
+ [default7]:07/03/2024 21:35:18 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 10 / 20 | consumed_tokens: 41.9M | elapsed_time_per_iteration_ms: 29.2K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 7.69 | lr: 5.74e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 4.31
230
+ [default0]:07/03/2024 21:35:47 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
231
+ [default7]:07/03/2024 21:35:47 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 11 / 20 | consumed_tokens: 46.1M | elapsed_time_per_iteration_ms: 29.1K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 7.45 | lr: 5.26e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 2.5
232
+ [default0]:07/03/2024 21:36:16 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
233
+ [default7]:07/03/2024 21:36:16 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 12 / 20 | consumed_tokens: 50.3M | elapsed_time_per_iteration_ms: 29.1K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 7.37 | lr: 4.79e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 5.02
234
+ [default0]:07/03/2024 21:36:45 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
235
+ [default7]:07/03/2024 21:36:45 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 13 / 20 | consumed_tokens: 54.5M | elapsed_time_per_iteration_ms: 29.1K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 7.31 | lr: 4.32e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 6.04
236
+ [default0]:07/03/2024 21:37:14 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
237
+ [default7]:07/03/2024 21:37:14 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 14 / 20 | consumed_tokens: 58.7M | elapsed_time_per_iteration_ms: 29.1K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 7.19 | lr: 3.84e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 5.29
238
+ [default0]:07/03/2024 21:37:44 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
239
+ [default7]:07/03/2024 21:37:44 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 15 / 20 | consumed_tokens: 62.9M | elapsed_time_per_iteration_ms: 29.1K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 7.06 | lr: 3.37e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 2.7
240
+ [default0]:07/03/2024 21:38:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
241
+ [default7]:07/03/2024 21:38:13 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 16 / 20 | consumed_tokens: 67.1M | elapsed_time_per_iteration_ms: 29.1K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 6.97 | lr: 2.89e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 2.02
242
+ [default0]:07/03/2024 21:38:42 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
243
+ [default7]:07/03/2024 21:38:42 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 17 / 20 | consumed_tokens: 71.3M | elapsed_time_per_iteration_ms: 29.1K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 6.91 | lr: 2.42e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 2.02
244
+ [default0]:07/03/2024 21:39:11 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
245
+ [default7]:07/03/2024 21:39:11 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 18 / 20 | consumed_tokens: 75.5M | elapsed_time_per_iteration_ms: 29.1K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 6.86 | lr: 1.95e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 2
246
+ [default7]:07/03/2024 21:39:40 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 19 / 20 | consumed_tokens: 79.7M | elapsed_time_per_iteration_ms: 29.1K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 6.81 | lr: 1.47e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 2.03
247
+ [default0]:07/03/2024 21:39:40 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 4717.29MiB. Peak allocated 44034.23MiB. Peak reserved: 44472.00MiB
248
+ [default7]:07/03/2024 21:40:09 [INFO|DP=0|PP=7|TP=0|ip-26-0-160-225]: iteration: 20 / 20 | consumed_tokens: 83.9M | elapsed_time_per_iteration_ms: 29.1K | tokens_per_sec: 144K | tokens_per_sec_per_gpu: 18K | global_batch_size: 1.02K | lm_loss: 6.77 | lr: 1e-05 | model_tflops_per_gpu: 163 | hardware_tflops_per_gpu: 163 | grad_norm: 1.94
249
+ Saved 1 csv files over 1 completed logs
250
+ Processing file: /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/profiler/ip-26-0-160-225_333932.1720042364135456939.pt.trace.json
251
+ Results written to /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/profiler.csv
252
+ 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.
253
+
llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/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
+ 1,4190000.0000000005,54200.0,77400.0,9670.0,1020.0,11.1,0.0001,87.8,87.8,24.9,4717.29,4717.31,44472.0
3
+ 2,8390000.0,29200.0,144000.0,17900.0,1020.0,11.1,9.53e-05,163.0,163.0,25.1,4717.29,4717.31,44472.0
4
+ 3,12600000.0,29300.0,143000.0,17900.0,1020.0,9.49,9.05e-05,162.0,162.0,21.5,4717.29,4717.31,44472.0
5
+ 4,16800000.0,29200.0,144000.0,18000.0,1020.0,9.36,8.58e-05,163.0,163.0,21.4,4717.29,44034.23,44472.0
6
+ 5,21000000.0,29200.0,144000.0,18000.0,1020.0,9.02,8.11e-05,163.0,163.0,12.7,,,
7
+ 6,25200000.0,29100.0,144000.0,18000.0,1020.0,10.3,7.63e-05,163.0,163.0,47.1,4717.29,44034.23,44472.0
8
+ 7,29400000.0,231000.0,18100.0,2270.0,1020.0,8.68,7.16e-05,20.6,20.6,5.58,,,
9
+ 8,33600000.0,29200.0,144000.0,18000.0,1020.0,8.32,6.68e-05,163.0,163.0,4.78,4717.29,44034.23,44472.0
10
+ 9,37700000.0,29200.0,144000.0,18000.0,1020.0,7.95,6.21e-05,163.0,163.0,3.31,4717.29,44034.23,44472.0
11
+ 10,41900000.0,29200.0,144000.0,18000.0,1020.0,7.69,5.74e-05,163.0,163.0,4.31,4717.29,44034.23,44472.0
12
+ 11,46100000.0,29100.0,144000.0,18000.0,1020.0,7.45,5.26e-05,163.0,163.0,2.5,4717.29,44034.23,44472.0
13
+ 12,50300000.0,29100.0,144000.0,18000.0,1020.0,7.37,4.79e-05,163.0,163.0,5.02,4717.29,44034.23,44472.0
14
+ 13,54500000.0,29100.0,144000.0,18000.0,1020.0,7.31,4.32e-05,163.0,163.0,6.04,4717.29,44034.23,44472.0
15
+ 14,58700000.0,29100.0,144000.0,18000.0,1020.0,7.19,3.84e-05,163.0,163.0,5.29,4717.29,44034.23,44472.0
16
+ 15,62900000.0,29100.0,144000.0,18000.0,1020.0,7.06,3.37e-05,163.0,163.0,2.7,4717.29,44034.23,44472.0
17
+ 16,67099999.99999999,29100.0,144000.0,18000.0,1020.0,6.97,2.89e-05,163.0,163.0,2.02,4717.29,44034.23,44472.0
18
+ 17,71300000.0,29100.0,144000.0,18000.0,1020.0,6.91,2.42e-05,163.0,163.0,2.02,4717.29,44034.23,44472.0
19
+ 18,75500000.0,29100.0,144000.0,18000.0,1020.0,6.86,1.95e-05,163.0,163.0,2.0,,,
20
+ 19,79700000.0,29100.0,144000.0,18000.0,1020.0,6.81,1.47e-05,163.0,163.0,2.03,4717.29,44034.23,44472.0
21
+ 20,83900000.0,29100.0,144000.0,18000.0,1020.0,6.77,1e-05,163.0,163.0,1.94,,,
llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/profiler.csv ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ forward,backward
2
+ 0ms 718μs,1ms 520μs
llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/profiler/ip-26-0-160-225_333932.1720042364135456939.pt.trace.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9bae39aea385561e40de37d1077602d4afda59d8c248c053fa799030d8a2aef1
3
+ size 6092122913
llama-1B/8_GPUS/dp-1_tp-1_pp-8_mbz-4/status.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ completed