Hugging Face's logo

nanotron
/
old_bench

Model card Files Community
3outeille HF staff commited on
Commit
ed0e41b
•
1 Parent(s): ae6e601

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

Browse files
Files changed (8) hide show
  1. .gitattributes +1 -0
  2. llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/bench.slurm +111 -0
  3. llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/config.yaml +90 -0
  4. llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/log.out +255 -0
  5. llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/log_metrics.csv +21 -0
  6. llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/profiler.csv +2 -0
  7. llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/profiler/ip-26-0-163-220_391612.1720042520675458251.pt.trace.json +3 -0
  8. llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/status.txt +1 -0
.gitattributes CHANGED
@@ -103,3 +103,4 @@ llama-1B/64_GPUS/dp-4_tp-16_pp-1_mbz-4/profiler/ip-26-0-161-153_1502820.17200029
103
  llama-1B/8_GPUS/dp-1_tp-2_pp-4_mbz-8/profiler/ip-26-0-164-207_582119.1720040069056805143.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
 
 
103
  llama-1B/8_GPUS/dp-1_tp-2_pp-4_mbz-8/profiler/ip-26-0-164-207_582119.1720040069056805143.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
llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/bench.slurm ADDED
@@ -0,0 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/bin/bash
2
+
3
+ #SBATCH --job-name=bench_cluster
4
+ #SBATCH --time=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-4_tp-2_pp-1_mbz-8/log.out
13
+ #SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/log.out
14
+
15
+ # Function to update status based on squeue output
16
+ update_status() {
17
+ job_id=$1
18
+ status_file=$2
19
+ # For unknown reasons, it doenst update status for pending. It only works for running
20
+ while true; do
21
+ job_status=$(squeue --job $job_id --noheader --format=%T)
22
+ echo "Job status: $job_status"
23
+ if [ -z "$job_status" ]; then
24
+ # Job has finished or is not found
25
+ break
26
+ elif [ "$job_status" = "RUNNING" ]; then
27
+ printf "running" > $status_file
28
+ break
29
+ fi
30
+ sleep 10
31
+ done
32
+ }
33
+
34
+ # Misc initializations.
35
+ echo "========================"
36
+ echo "START TIME: $(date)"
37
+ source /fsx/ferdinandmom/miniforge3/etc/profile.d/conda.sh
38
+ conda activate /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster
39
+ echo python3 version = $(python3 --version)
40
+ echo "========================"
41
+
42
+ # Slurm stuff
43
+ export HOSTNAMES=$(scontrol show hostnames "$SLURM_JOB_NODELIST")
44
+ export MASTER_ADDR=$(scontrol show hostnames "$SLURM_JOB_NODELIST" | head -n 1)
45
+ export MASTER_PORT=$((1024 + RANDOM % 64511))
46
+
47
+ export TMPDIR=/scratch
48
+ export HF_DATASETS_CACHE="/admin/home/ferdinand_mom/.cache"
49
+ export CUBLAS_WORKSPACE_CONFIG=":4096:8"
50
+ export CUDA_DEVICE_MAX_CONNECTIONS="1"
51
+
52
+ huggingface-cli login --token $HUGGINGFACE_TOKEN
53
+
54
+
55
+ NANOTRON_REPO="/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron"
56
+ CMD="$NANOTRON_REPO/run_train.py --config-file /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/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-4_tp-2_pp-1_mbz-8/status.txt &
76
+
77
+ # Run the main command
78
+ srun -u $LAUNCHER $CMD
79
+ exit_status=$?
80
+
81
+ # Update status based on the exit status of `srun`
82
+ if [ $exit_status -eq 0 ]; then
83
+ printf "completed" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/status.txt
84
+ else
85
+ if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/log.out; then
86
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/status.txt
87
+ elif grep -q " CUDA error: an illegal memory access" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/log.out; then
88
+ printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/status.txt
89
+ elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/log.out; then
90
+ printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/status.txt
91
+ else
92
+ printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/status.txt
93
+ fi
94
+ fi
95
+
96
+ # Run the report script if the job completed successfully
97
+ if [ $exit_status -eq 0 ]; then
98
+ python /fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py report --inp_dir /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8 --is_logs
99
+ python /fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py report --inp_dir /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8 --is_profiler
100
+ fi
101
+
102
+
103
+ # Push to hub the folder using huggingface_cli
104
+ huggingface-cli upload nanotron/bench_cluster /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8 llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8 --commit-message "Upload llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8"
105
+
106
+ # Verify the upload
107
+ if [ $? -eq 0 ]; then
108
+ echo "Uploading to Huggingface Hub successful"
109
+ else
110
+ echo "Failed to upload to Huggingface Hub"
111
+ fi
llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/config.yaml ADDED
@@ -0,0 +1,90 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ general:
2
+ project: bench_cluster
3
+ seed: 42
4
+ model:
5
+ ddp_bucket_cap_mb: 25
6
+ dtype: bfloat16
7
+ init_method:
8
+ std: 0.025
9
+ make_vocab_size_divisible_by: 1
10
+ model_config:
11
+ bos_token_id: 1
12
+ eos_token_id: 2
13
+ hidden_act: silu
14
+ hidden_size: 2048
15
+ initializer_range: 0.02
16
+ intermediate_size: 4096
17
+ is_llama_config: true
18
+ max_position_embeddings: 4096
19
+ num_attention_heads: 32
20
+ num_hidden_layers: 24
21
+ num_key_value_heads: 32
22
+ pad_token_id: null
23
+ pretraining_tp: 1
24
+ rms_norm_eps: 1.0e-05
25
+ rope_scaling: null
26
+ rope_theta: 10000.0
27
+ tie_word_embeddings: true
28
+ use_cache: true
29
+ vocab_size: 50257
30
+ optimizer:
31
+ accumulate_grad_in_fp32: true
32
+ clip_grad: 1.0
33
+ learning_rate_scheduler:
34
+ learning_rate: 0.0001
35
+ lr_decay_style: linear
36
+ lr_warmup_style: linear
37
+ lr_warmup_steps: 1
38
+ min_decay_lr: 1.0e-05
39
+ optimizer_factory:
40
+ adam_beta1: 0.9
41
+ adam_beta2: 0.95
42
+ adam_eps: 1.0e-08
43
+ name: adamW
44
+ torch_adam_is_fused: true
45
+ weight_decay: 0.01
46
+ zero_stage: 1
47
+ parallelism:
48
+ dp: 4
49
+ expert_parallel_size: 1
50
+ pp: 1
51
+ pp_engine: 1f1b
52
+ tp: 2
53
+ tp_linear_async_communication: false
54
+ tp_mode: REDUCE_SCATTER
55
+ profiler:
56
+ profiler_export_path: /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8
57
+ tokenizer:
58
+ tokenizer_max_length: null
59
+ tokenizer_name_or_path: openai-community/gpt2
60
+ tokenizer_revision: null
61
+ data_stages:
62
+ - name: Training Stage
63
+ start_training_step: 1
64
+ data:
65
+ dataset:
66
+ dataset_overwrite_cache: false
67
+ dataset_processing_num_proc_per_process: 64
68
+ hf_dataset_config_name: null
69
+ hf_dataset_or_datasets: roneneldan/TinyStories
70
+ hf_dataset_splits: train
71
+ text_column_name: text
72
+ num_loading_workers: 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: 8
82
+ sequence_length: 4096
83
+ logging:
84
+ iteration_step_info_interval: 1
85
+ log_level: info
86
+ log_level_replica: info
87
+ checkpoints:
88
+ checkpoint_interval: 100000
89
+ checkpoints_path: /dev/null
90
+ resume_checkpoint_path: null
llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/log.out ADDED
@@ -0,0 +1,255 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ========================
2
+ START TIME: Wed Jul 3 21:30:53 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 21:30:56.193000 139902981371712 torch/distributed/run.py:757]
18
+ W0703 21:30:56.193000 139902981371712 torch/distributed/run.py:757] *****************************************
19
+ W0703 21:30:56.193000 139902981371712 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 21:30:56.193000 139902981371712 torch/distributed/run.py:757] *****************************************
21
+ [default0]:07/03/2024 21:31:12 [WARNING|DP=0|PP=0|TP=0|ip-26-0-163-220]: [Vocab Size Padding] Padded vocab (size: 50257) with 1 dummy tokens (new size: 50258)
22
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Config:
23
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Config(general=GeneralArgs(project='bench_cluster',
24
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: run='%date_%jobid',
25
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: seed=42,
26
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: step=None,
27
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: consumed_train_samples=None,
28
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: benchmark_csv_path=None,
29
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: ignore_sanity_checks=True),
30
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: parallelism=ParallelismArgs(dp=4,
31
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: pp=1,
32
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: tp=2,
33
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f9d57790940>,
34
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
35
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: tp_linear_async_communication=False,
36
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: expert_parallel_size=1),
37
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
38
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: eos_token_id=2,
39
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: hidden_act='silu',
40
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: hidden_size=2048,
41
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: initializer_range=0.02,
42
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: intermediate_size=4096,
43
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: is_llama_config=True,
44
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: max_position_embeddings=4096,
45
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: num_attention_heads=32,
46
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: num_hidden_layers=24,
47
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: num_key_value_heads=32,
48
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: pad_token_id=None,
49
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: pretraining_tp=1,
50
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: rms_norm_eps=1e-05,
51
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: rope_scaling=None,
52
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: rope_theta=10000.0,
53
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: tie_word_embeddings=True,
54
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: use_cache=True,
55
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: vocab_size=50258),
56
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: init_method=RandomInit(std=0.025),
57
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: dtype=torch.bfloat16,
58
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: make_vocab_size_divisible_by=1,
59
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: ddp_bucket_cap_mb=25),
60
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
61
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: tokenizer_revision=None,
62
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: tokenizer_max_length=None),
63
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
64
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: checkpoint_interval=100000,
65
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: save_initial_state=False,
66
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: resume_checkpoint_path=None,
67
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: checkpoints_path_is_shared_file_system=False),
68
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: logging=LoggingArgs(log_level='info',
69
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: log_level_replica='info',
70
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration_step_info_interval=1),
71
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: tokens=TokensArgs(sequence_length=4096,
72
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: train_steps=20,
73
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: micro_batch_size=8,
74
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: batch_accumulation_per_replica=32,
75
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: val_check_interval=-1,
76
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: limit_val_batches=0,
77
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: limit_test_batches=0),
78
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
79
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: adam_beta1=0.9,
80
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: adam_beta2=0.95,
81
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: torch_adam_is_fused=True,
82
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: name='adamW'),
83
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: zero_stage=1,
84
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: weight_decay=0.01,
85
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: clip_grad=1.0,
86
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: accumulate_grad_in_fp32=True,
87
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
88
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: lr_warmup_steps=1,
89
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: lr_warmup_style='linear',
90
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: lr_decay_style='linear',
91
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: lr_decay_steps=19,
92
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: lr_decay_starting_step=None,
93
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: min_decay_lr=1e-05)),
94
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: data_stages=[DatasetStageArgs(name='Training Stage',
95
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: start_training_step=1,
96
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
97
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: hf_dataset_splits='train',
98
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: hf_dataset_config_name=None,
99
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: dataset_processing_num_proc_per_process=64,
100
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: dataset_overwrite_cache=False,
101
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: text_column_name='text'),
102
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: seed=42,
103
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: num_loading_workers=0))],
104
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8')),
105
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: lighteval=None)
106
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Model Config:
107
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: LlamaConfig(bos_token_id=1,
108
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: eos_token_id=2,
109
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: hidden_act='silu',
110
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: hidden_size=2048,
111
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: initializer_range=0.02,
112
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: intermediate_size=4096,
113
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: is_llama_config=True,
114
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: max_position_embeddings=4096,
115
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: num_attention_heads=32,
116
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: num_hidden_layers=24,
117
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: num_key_value_heads=32,
118
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: pad_token_id=None,
119
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: pretraining_tp=1,
120
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: rms_norm_eps=1e-05,
121
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: rope_scaling=None,
122
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: rope_theta=10000.0,
123
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: tie_word_embeddings=True,
124
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: use_cache=True,
125
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: vocab_size=50258)
126
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Building model..
127
+ [default0]:07/03/2024 21:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Setting PP block ranks...
128
+ [default3]:07/03/2024 21:31:22 [INFO|DP=1|PP=0|TP=1|ip-26-0-163-220]: No checkpoint path provided.
129
+ [default2]:07/03/2024 21:31:22 [INFO|DP=1|PP=0|TP=0|ip-26-0-163-220]: No checkpoint path provided.
130
+ [default4]:07/03/2024 21:31:22 [INFO|DP=2|PP=0|TP=0|ip-26-0-163-220]: No checkpoint path provided.
131
+ [default7]:07/03/2024 21:31:22 [INFO|DP=3|PP=0|TP=1|ip-26-0-163-220]: No checkpoint path provided.
132
+ [default1]:07/03/2024 21:31:22 [INFO|DP=0|PP=0|TP=1|ip-26-0-163-220]: Local number of parameters: 555M (1058.35MiB)
133
+ [default1]:07/03/2024 21:31:22 [INFO|DP=0|PP=0|TP=1|ip-26-0-163-220]: [After model building] Memory usage: 1082.37MiB. Peak allocated: 1182.56MiB Peak reserved: 1200.00MiB
134
+ [default0]:07/03/2024 21:31:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Total number of parameters: 1.11G (2116.70MiB)
135
+ [default0]:07/03/2024 21:31:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Local number of parameters: 555M (1058.35MiB)
136
+ [default0]:07/03/2024 21:31:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: [After model building] Memory usage: 1082.37MiB. Peak allocated: 1182.56MiB Peak reserved: 1200.00MiB
137
+ [default0]:07/03/2024 21:31:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: No checkpoint path provided.
138
+ [default0]:07/03/2024 21:31:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Parametrizing model parameters using StandardParametrizator
139
+ [default1]:07/03/2024 21:31:22 [INFO|DP=0|PP=0|TP=1|ip-26-0-163-220]: No checkpoint path provided.
140
+ [default5]:07/03/2024 21:31:22 [INFO|DP=2|PP=0|TP=1|ip-26-0-163-220]: No checkpoint path provided.
141
+ [default6]:07/03/2024 21:31:22 [INFO|DP=3|PP=0|TP=0|ip-26-0-163-220]: No checkpoint path provided.
142
+ [default0]:07/03/2024 21:31:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: [Optimizer Building] Using LearningRateForSP as learning rate
143
+ [default0]:07/03/2024 21:31:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: [ZeRO sharding] Size of optimizer params per rank:
144
+ [default0]:07/03/2024 21:31:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: [ZeRO sharding] DP Rank 0 has 139M out of 555M (25.00%) params' optimizer states
145
+ [default0]:07/03/2024 21:31:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: [ZeRO sharding] DP Rank 1 has 139M out of 555M (25.00%) params' optimizer states
146
+ [default0]:07/03/2024 21:31:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: [ZeRO sharding] DP Rank 2 has 139M out of 555M (25.00%) params' optimizer states
147
+ [default0]:07/03/2024 21:31:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: [ZeRO sharding] DP Rank 3 has 139M out of 555M (25.00%) params' optimizer states
148
+ [default0]:07/03/2024 21:31:28 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
149
+ [default0]:07/03/2024 21:31:28 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Using `datasets` library
150
+ [default0]:07/03/2024 21:31:28 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
151
+ [default0]:07/03/2024 21:31:28 [WARNING|DP=0|PP=0|TP=0|ip-26-0-163-220]: Repo card metadata block was not found. Setting CardData to empty.
152
+ [default0]:Repo card metadata block was not found. Setting CardData to empty.
153
+ [default0]:07/03/2024 21:31:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: [Training Plan] There are 1 training stages
154
+ [default0]:07/03/2024 21:31:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: [Stage Training Stage] start from step 1
155
+ [default0]:07/03/2024 21:31:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]:
156
+ [default0]:07/03/2024 21:31:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: [Start training] datetime: 2024-07-03 21:31:29.096784 | mbs: 8 | grad_accum: 32 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
157
+ [default0]:07/03/2024 21:31:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
158
+ [default0]:07/03/2024 21:31:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 3729.08MiB. Peak allocated 3729.08MiB. Peak reserved: 3848.00MiB
159
+ [default5]:07/03/2024 21:31:29 [WARNING|DP=2|PP=0|TP=1|ip-26-0-163-220]: Repo card metadata block was not found. Setting CardData to empty.
160
+ [default5]:Repo card metadata block was not found. Setting CardData to empty.
161
+ [default1]:Repo card metadata block was not found. Setting CardData to empty.
162
+ [default6]:Repo card metadata block was not found. Setting CardData to empty.
163
+ [default2]:Repo card metadata block was not found. Setting CardData to empty.
164
+ [default3]:Repo card metadata block was not found. Setting CardData to empty.
165
+ [default3]:07/03/2024 21:31:29 [WARNING|DP=1|PP=0|TP=1|ip-26-0-163-220]: Repo card metadata block was not found. Setting CardData to empty.
166
+ [default4]:07/03/2024 21:31:29 [WARNING|DP=2|PP=0|TP=0|ip-26-0-163-220]: Repo card metadata block was not found. Setting CardData to empty.
167
+ [default7]:07/03/2024 21:31:29 [WARNING|DP=3|PP=0|TP=1|ip-26-0-163-220]: Repo card metadata block was not found. Setting CardData to empty.
168
+ [default2]:07/03/2024 21:31:29 [WARNING|DP=1|PP=0|TP=0|ip-26-0-163-220]: Repo card metadata block was not found. Setting CardData to empty.
169
+ [default1]:07/03/2024 21:31:29 [WARNING|DP=0|PP=0|TP=1|ip-26-0-163-220]: Repo card metadata block was not found. Setting CardData to empty.
170
+ [default7]:Repo card metadata block was not found. Setting CardData to empty.
171
+ [default6]:07/03/2024 21:31:29 [WARNING|DP=3|PP=0|TP=0|ip-26-0-163-220]: Repo card metadata block was not found. Setting CardData to empty.
172
+ [default4]:Repo card metadata block was not found. Setting CardData to empty.
173
+ [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.)
174
+ [default1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
175
+ [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.)
176
+ [default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
177
+ [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.)
178
+ [default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
179
+ [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.)
180
+ [default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
181
+ [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.)
182
+ [default6]: 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
+ [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.)
186
+ [default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
187
+ [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.)
188
+ [default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
189
+ [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
190
+ [default1]: warnings.warn(
191
+ [default0]:07/03/2024 21:31:48 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 3805.65MiB. Peak allocated 42590.44MiB. Peak reserved: 44376.00MiB
192
+ [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
193
+ [default0]: warnings.warn(
194
+ [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
195
+ [default7]: warnings.warn(
196
+ [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
197
+ [default6]: warnings.warn(
198
+ [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
199
+ [default5]: 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
+ [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
203
+ [default2]: warnings.warn(
204
+ [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
205
+ [default3]: warnings.warn(
206
+ [default0]:07/03/2024 21:31:49 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 1 / 20 | consumed_tokens: 4.19M | elapsed_time_per_iteration_ms: 20.5K | tokens_per_sec: 205K | tokens_per_sec_per_gpu: 25.6K | global_batch_size: 1.02K | lm_loss: 11.5 | lr: 0.0001 | model_tflops_per_gpu: 232 | hardware_tflops_per_gpu: 232 | grad_norm: 26.4 | cuda_memory_allocated: 5.1G | cuda_max_memory_reserved: 46.5G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.6G | hd_free_memory_tb: 247G
207
+ [default0]:07/03/2024 21:31:49 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 7247.21MiB. Peak reserved: 44376.00MiB
208
+ [default0]:07/03/2024 21:32:05 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.93MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
209
+ [default0]:07/03/2024 21:32:05 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 2 / 20 | consumed_tokens: 8.39M | elapsed_time_per_iteration_ms: 15.9K | tokens_per_sec: 263K | tokens_per_sec_per_gpu: 32.9K | global_batch_size: 1.02K | lm_loss: 11.5 | lr: 9.53e-05 | model_tflops_per_gpu: 298 | hardware_tflops_per_gpu: 298 | grad_norm: 26.6 | cuda_memory_allocated: 5.1G | cuda_max_memory_reserved: 46.5G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.6G | hd_free_memory_tb: 247G
210
+ [default0]:07/03/2024 21:32:05 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 7247.22MiB. Peak reserved: 44378.00MiB
211
+ [default0]:07/03/2024 21:32:21 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.93MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
212
+ [default0]:07/03/2024 21:32:21 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 3 / 20 | consumed_tokens: 12.6M | elapsed_time_per_iteration_ms: 15.8K | tokens_per_sec: 265K | tokens_per_sec_per_gpu: 33.1K | global_batch_size: 1.02K | lm_loss: 11.5 | lr: 9.05e-05 | model_tflops_per_gpu: 300 | hardware_tflops_per_gpu: 300 | grad_norm: 262 | cuda_memory_allocated: 5.1G | cuda_max_memory_reserved: 46.5G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.6G | hd_free_memory_tb: 247G
213
+ [default0]:07/03/2024 21:32:21 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 7247.22MiB. Peak reserved: 44378.00MiB
214
+ [default0]:STAGE:2024-07-03 21:32:21 391612:391612 ActivityProfilerController.cpp:314] Completed Stage: Warm Up
215
+ [default0]:07/03/2024 21:32:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.93MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
216
+ [default0]:07/03/2024 21:32:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 4 / 20 | consumed_tokens: 16.8M | elapsed_time_per_iteration_ms: 15.9K | tokens_per_sec: 264K | tokens_per_sec_per_gpu: 33K | global_batch_size: 1.02K | lm_loss: 14.6 | lr: 8.58e-05 | model_tflops_per_gpu: 300 | hardware_tflops_per_gpu: 300 | grad_norm: 29.1 | cuda_memory_allocated: 5.1G | cuda_max_memory_reserved: 46.5G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.6G | hd_free_memory_tb: 247G
217
+ [default0]:07/03/2024 21:32:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 7247.22MiB. Peak reserved: 44378.00MiB
218
+ [default0]:07/03/2024 21:32:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 5 / 20 | consumed_tokens: 21M | elapsed_time_per_iteration_ms: 15.8K | tokens_per_sec: 265K | tokens_per_sec_per_gpu: 33.1K | global_batch_size: 1.02K | lm_loss: 10.8 | lr: 8.11e-05 | model_tflops_per_gpu: 301 | hardware_tflops_per_gpu: 301 | grad_norm: 31.2
219
+ [default0]:07/03/2024 21:32:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
220
+ [default0]:07/03/2024 21:33:08 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 6 / 20 | consumed_tokens: 25.2M | elapsed_time_per_iteration_ms: 15.8K | tokens_per_sec: 266K | tokens_per_sec_per_gpu: 33.2K | global_batch_size: 1.02K | lm_loss: 10.6 | lr: 7.63e-05 | model_tflops_per_gpu: 301 | hardware_tflops_per_gpu: 301 | grad_norm: 27
221
+ [default0]:STAGE:2024-07-03 21:33:26 391612:391612 ActivityProfilerController.cpp:320] Completed Stage: Collection
222
+ [default0]:STAGE:2024-07-03 21:33:28 391612:391612 ActivityProfilerController.cpp:324] Completed Stage: Post Processing
223
+ [default0]:07/03/2024 21:35:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
224
+ [default0]:07/03/2024 21:36:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 7 / 20 | consumed_tokens: 29.4M | elapsed_time_per_iteration_ms: 15.7K | tokens_per_sec: 267K | tokens_per_sec_per_gpu: 33.4K | global_batch_size: 1.02K | lm_loss: 10.2 | lr: 7.16e-05 | model_tflops_per_gpu: 303 | hardware_tflops_per_gpu: 303 | grad_norm: 9.43
225
+ [default0]:07/03/2024 21:36:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
226
+ [default0]:07/03/2024 21:36:25 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 8 / 20 | consumed_tokens: 33.6M | elapsed_time_per_iteration_ms: 15.8K | tokens_per_sec: 266K | tokens_per_sec_per_gpu: 33.3K | global_batch_size: 1.02K | lm_loss: 12.8 | lr: 6.68e-05 | model_tflops_per_gpu: 302 | hardware_tflops_per_gpu: 302 | grad_norm: 77.8
227
+ [default0]:07/03/2024 21:36:25 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
228
+ [default0]:07/03/2024 21:36:40 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 9 / 20 | consumed_tokens: 37.7M | elapsed_time_per_iteration_ms: 15.8K | tokens_per_sec: 266K | tokens_per_sec_per_gpu: 33.3K | global_batch_size: 1.02K | lm_loss: 9.42 | lr: 6.21e-05 | model_tflops_per_gpu: 302 | hardware_tflops_per_gpu: 302 | grad_norm: 11.7
229
+ [default0]:07/03/2024 21:36:40 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
230
+ [default0]:07/03/2024 21:36:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 10 / 20 | consumed_tokens: 41.9M | elapsed_time_per_iteration_ms: 15.8K | tokens_per_sec: 265K | tokens_per_sec_per_gpu: 33.2K | global_batch_size: 1.02K | lm_loss: 9.19 | lr: 5.74e-05 | model_tflops_per_gpu: 301 | hardware_tflops_per_gpu: 301 | grad_norm: 6.73
231
+ [default0]:07/03/2024 21:36:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
232
+ [default0]:07/03/2024 21:37:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 11 / 20 | consumed_tokens: 46.1M | elapsed_time_per_iteration_ms: 15.8K | tokens_per_sec: 265K | tokens_per_sec_per_gpu: 33.1K | global_batch_size: 1.02K | lm_loss: 8.91 | lr: 5.26e-05 | model_tflops_per_gpu: 300 | hardware_tflops_per_gpu: 300 | grad_norm: 5.86
233
+ [default0]:07/03/2024 21:37:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
234
+ [default0]:07/03/2024 21:37:28 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 12 / 20 | consumed_tokens: 50.3M | elapsed_time_per_iteration_ms: 15.8K | tokens_per_sec: 266K | tokens_per_sec_per_gpu: 33.2K | global_batch_size: 1.02K | lm_loss: 8.53 | lr: 4.79e-05 | model_tflops_per_gpu: 301 | hardware_tflops_per_gpu: 301 | grad_norm: 6.22
235
+ [default0]:07/03/2024 21:37:28 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
236
+ [default0]:07/03/2024 21:37:44 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 13 / 20 | consumed_tokens: 54.5M | elapsed_time_per_iteration_ms: 15.8K | tokens_per_sec: 266K | tokens_per_sec_per_gpu: 33.2K | global_batch_size: 1.02K | lm_loss: 8.07 | lr: 4.32e-05 | model_tflops_per_gpu: 301 | hardware_tflops_per_gpu: 301 | grad_norm: 5.34
237
+ [default0]:07/03/2024 21:37:44 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
238
+ [default0]:07/03/2024 21:37:59 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 14 / 20 | consumed_tokens: 58.7M | elapsed_time_per_iteration_ms: 15.8K | tokens_per_sec: 266K | tokens_per_sec_per_gpu: 33.2K | global_batch_size: 1.02K | lm_loss: 7.67 | lr: 3.84e-05 | model_tflops_per_gpu: 301 | hardware_tflops_per_gpu: 301 | grad_norm: 4.53
239
+ [default0]:07/03/2024 21:37:59 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
240
+ [default0]:07/03/2024 21:38:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 15 / 20 | consumed_tokens: 62.9M | elapsed_time_per_iteration_ms: 15.9K | tokens_per_sec: 264K | tokens_per_sec_per_gpu: 33K | global_batch_size: 1.02K | lm_loss: 7.51 | lr: 3.37e-05 | model_tflops_per_gpu: 299 | hardware_tflops_per_gpu: 299 | grad_norm: 7.61
241
+ [default0]:07/03/2024 21:38:15 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
242
+ [default0]:07/03/2024 21:38:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 16 / 20 | consumed_tokens: 67.1M | elapsed_time_per_iteration_ms: 15.8K | tokens_per_sec: 265K | tokens_per_sec_per_gpu: 33.2K | global_batch_size: 1.02K | lm_loss: 7.4 | lr: 2.89e-05 | model_tflops_per_gpu: 301 | hardware_tflops_per_gpu: 301 | grad_norm: 7.23
243
+ [default0]:07/03/2024 21:38:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
244
+ [default0]:07/03/2024 21:38:47 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 17 / 20 | consumed_tokens: 71.3M | elapsed_time_per_iteration_ms: 15.8K | tokens_per_sec: 265K | tokens_per_sec_per_gpu: 33.2K | global_batch_size: 1.02K | lm_loss: 7.37 | lr: 2.42e-05 | model_tflops_per_gpu: 301 | hardware_tflops_per_gpu: 301 | grad_norm: 6.64
245
+ [default0]:07/03/2024 21:38:47 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
246
+ [default0]:07/03/2024 21:39:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 18 / 20 | consumed_tokens: 75.5M | elapsed_time_per_iteration_ms: 15.8K | tokens_per_sec: 266K | tokens_per_sec_per_gpu: 33.2K | global_batch_size: 1.02K | lm_loss: 7.33 | lr: 1.95e-05 | model_tflops_per_gpu: 301 | hardware_tflops_per_gpu: 301 | grad_norm: 7.06
247
+ [default0]:07/03/2024 21:39:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
248
+ [default0]:07/03/2024 21:39:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 19 / 20 | consumed_tokens: 79.7M | elapsed_time_per_iteration_ms: 15.8K | tokens_per_sec: 266K | tokens_per_sec_per_gpu: 33.2K | global_batch_size: 1.02K | lm_loss: 7.22 | lr: 1.47e-05 | model_tflops_per_gpu: 301 | hardware_tflops_per_gpu: 301 | grad_norm: 5.71
249
+ [default0]:07/03/2024 21:39:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: Memory usage: 4865.92MiB. Peak allocated 43650.72MiB. Peak reserved: 44378.00MiB
250
+ [default0]:07/03/2024 21:39:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-163-220]: iteration: 20 / 20 | consumed_tokens: 83.9M | elapsed_time_per_iteration_ms: 15.8K | tokens_per_sec: 266K | tokens_per_sec_per_gpu: 33.2K | global_batch_size: 1.02K | lm_loss: 7.11 | lr: 1e-05 | model_tflops_per_gpu: 301 | hardware_tflops_per_gpu: 301 | grad_norm: 4.39
251
+ Saved 1 csv files over 1 completed logs
252
+ Processing file: /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/profiler/ip-26-0-163-220_391612.1720042520675458251.pt.trace.json
253
+ Results written to /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/profiler.csv
254
+ 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.
255
+
llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/log_metrics.csv ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ iteration,consumed_tokens,elapsed_time_per_iteration_ms,tokens_per_sec,tokens_per_sec_per_gpu,global_batch_size,lm_loss,lr,model_tflops_per_gpu,hardware_tflops_per_gpu,grad_norm,memory_usage_MiB,peak_allocated_MiB,peak_reserved_MiB
2
+ 1,4190000.0000000005,20500.0,205000.0,25600.0,1020.0,11.5,0.0001,232.0,232.0,26.4,4865.93,43650.72,44378.0
3
+ 2,8390000.0,15900.0,263000.0,32900.0,1020.0,11.5,9.53e-05,298.0,298.0,26.6,4865.93,43650.72,44378.0
4
+ 3,12600000.0,15800.0,265000.0,33100.0,1020.0,11.5,9.05e-05,300.0,300.0,262.0,4865.93,43650.72,44378.0
5
+ 4,16800000.0,15900.0,264000.0,33000.0,1020.0,14.6,8.58e-05,300.0,300.0,29.1,4865.92,7247.22,44378.0
6
+ 5,21000000.0,15800.0,265000.0,33100.0,1020.0,10.8,8.11e-05,301.0,301.0,31.2,4865.92,43650.72,44378.0
7
+ 6,25200000.0,15800.0,266000.0,33200.0,1020.0,10.6,7.63e-05,301.0,301.0,27.0,4865.92,43650.72,44378.0
8
+ 7,29400000.0,15700.0,267000.0,33400.0,1020.0,10.2,7.16e-05,303.0,303.0,9.43,4865.92,43650.72,44378.0
9
+ 8,33600000.0,15800.0,266000.0,33300.0,1020.0,12.8,6.68e-05,302.0,302.0,77.8,4865.92,43650.72,44378.0
10
+ 9,37700000.0,15800.0,266000.0,33300.0,1020.0,9.42,6.21e-05,302.0,302.0,11.7,4865.92,43650.72,44378.0
11
+ 10,41900000.0,15800.0,265000.0,33200.0,1020.0,9.19,5.74e-05,301.0,301.0,6.73,4865.92,43650.72,44378.0
12
+ 11,46100000.0,15800.0,265000.0,33100.0,1020.0,8.91,5.26e-05,300.0,300.0,5.86,4865.92,43650.72,44378.0
13
+ 12,50300000.0,15800.0,266000.0,33200.0,1020.0,8.53,4.79e-05,301.0,301.0,6.22,4865.92,43650.72,44378.0
14
+ 13,54500000.0,15800.0,266000.0,33200.0,1020.0,8.07,4.32e-05,301.0,301.0,5.34,4865.92,43650.72,44378.0
15
+ 14,58700000.0,15800.0,266000.0,33200.0,1020.0,7.67,3.84e-05,301.0,301.0,4.53,4865.92,43650.72,44378.0
16
+ 15,62900000.0,15900.0,264000.0,33000.0,1020.0,7.51,3.37e-05,299.0,299.0,7.61,4865.92,43650.72,44378.0
17
+ 16,67099999.99999999,15800.0,265000.0,33200.0,1020.0,7.4,2.89e-05,301.0,301.0,7.23,4865.92,43650.72,44378.0
18
+ 17,71300000.0,15800.0,265000.0,33200.0,1020.0,7.37,2.42e-05,301.0,301.0,6.64,4865.92,43650.72,44378.0
19
+ 18,75500000.0,15800.0,266000.0,33200.0,1020.0,7.33,1.95e-05,301.0,301.0,7.06,4865.92,43650.72,44378.0
20
+ 19,79700000.0,15800.0,266000.0,33200.0,1020.0,7.22,1.47e-05,301.0,301.0,5.71,4865.92,43650.72,44378.0
21
+ 20,83900000.0,15800.0,266000.0,33200.0,1020.0,7.11,1e-05,301.0,301.0,4.39,,,
llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/profiler.csv ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ forward,backward
2
+ 1ms 69μs,0ms 403μs
llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/profiler/ip-26-0-163-220_391612.1720042520675458251.pt.trace.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e11743f8edd83a2ac26c971468d3369c7998996a5f83c01785e9d132f5dc1cfa
3
+ size 4520159402
llama-1B/8_GPUS/dp-4_tp-2_pp-1_mbz-8/status.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ completed