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

.gitattributes

@@ -38,3 +38,4 @@ llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-8/profiler/ip-26-0-170-31_2724547.1719930375
 llama-1B/16_GPUS/dp-1_tp-1_pp-16_mbz-1/profiler/ip-26-0-161-178_157959.1719931391546523089.pt.trace.json filter=lfs diff=lfs merge=lfs -text
 llama-1B/16_GPUS/dp-1_tp-2_pp-8_mbz-2/profiler/ip-26-0-163-134_1435141.1719931469116122110.pt.trace.json filter=lfs diff=lfs merge=lfs -text
 llama-1B/16_GPUS/dp-8_tp-1_pp-2_mbz-4/profiler/ip-26-0-171-56_3064380.1719933802382173616.pt.trace.json filter=lfs diff=lfs merge=lfs -text
+llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/profiler/ip-26-0-171-56_3084719.1719934470978269987.pt.trace.json filter=lfs diff=lfs merge=lfs -text

llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/bench.slurm

@@ -0,0 +1,111 @@
+#!/bin/bash
+
+#SBATCH --job-name=bench_cluster
+#SBATCH --time=00:59:00
+#SBATCH --partition=hopper-prod
+#SBATCH --nodes=2
+#SBATCH --gres=gpu:8
+#SBATCH --qos=high
+#SBATCH --ntasks-per-node=1
+#SBATCH --cpus-per-task=96
+#SBATCH --exclusive
+#SBATCH --output=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/log.out
+#SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/log.out
+
+# Function to update status based on squeue output
+update_status() {
+    job_id=$1
+    status_file=$2
+    # For unknown reasons, it doenst update status for pending. It only works for running 
+    while true; do
+        job_status=$(squeue --job $job_id --noheader --format=%T)
+        echo "Job status: $job_status"
+        if [ -z "$job_status" ]; then
+            # Job has finished or is not found
+            break
+        elif [ "$job_status" = "RUNNING" ]; then
+            printf "running" > $status_file
+            break
+        fi
+        sleep 10
+    done
+}
+
+# Misc initializations.
+echo "========================"
+echo "START TIME: $(date)"
+source /fsx/ferdinandmom/miniforge3/etc/profile.d/conda.sh
+conda activate /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster
+echo python3 version = $(python3 --version)
+echo "========================"
+
+# Slurm stuff
+export HOSTNAMES=$(scontrol show hostnames "$SLURM_JOB_NODELIST")
+export MASTER_ADDR=$(scontrol show hostnames "$SLURM_JOB_NODELIST" | head -n 1)
+export MASTER_PORT=$((1024 + RANDOM % 64511))
+
+export TMPDIR=/scratch
+export HF_DATASETS_CACHE="/admin/home/ferdinand_mom/.cache"
+export CUBLAS_WORKSPACE_CONFIG=":4096:8"
+export CUDA_DEVICE_MAX_CONNECTIONS="1"
+
+huggingface-cli login --token $HUGGINGFACE_TOKEN
+
+
+NANOTRON_REPO="/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron"
+CMD="$NANOTRON_REPO/run_train.py --config-file /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/config.yaml"
+
+LAUNCHER="torchrun \
+   --nproc_per_node 8 \
+   --nnodes 2 \
+   --rdzv_endpoint ${MASTER_ADDR}:${MASTER_PORT} \
+   --rdzv_backend c10d \
+   --max_restarts 0 \
+   --tee 3 \
+   --node_rank ${SLURM_PROCID}"
+
+# Checkout the bench_cluster branch
+cd $NANOTRON_REPO
+git checkout bench_cluster
+cd ..
+# Get the current job ID
+job_id=${SLURM_JOB_ID}
+
+# Update status to "pending" or "running" in the background
+update_status $job_id /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/status.txt &
+
+# Run the main command
+srun -u $LAUNCHER $CMD
+exit_status=$?
+
+# Update status based on the exit status of `srun`
+if [ $exit_status -eq 0 ]; then
+    printf "completed" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/status.txt
+else
+    if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/log.out; then
+        printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/status.txt
+    elif grep -q " CUDA error: an illegal memory access" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/log.out; then
+        printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/status.txt
+    elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/log.out; then
+        printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/status.txt
+    else
+        printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/status.txt
+    fi 
+fi
+
+# Run the report script if the job completed successfully
+if [ $exit_status -eq 0 ]; then
+    python /fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py report --inp_dir /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2 --is_logs
+    python /fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py report --inp_dir /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2 --is_profiler
+fi
+
+
+# Push to hub the folder using huggingface_cli
+huggingface-cli upload nanotron/bench_cluster /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2 llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2 --commit-message "Upload llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2"
+
+# Verify the upload
+if [ $? -eq 0 ]; then
+    echo "Uploading to Huggingface Hub successful"
+else
+    echo "Failed to upload to Huggingface Hub"
+fi

llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/config.yaml

@@ -0,0 +1,90 @@
+general:
+  project: bench_cluster
+  seed: 42
+model:
+  ddp_bucket_cap_mb: 25
+  dtype: bfloat16
+  init_method:
+    std: 0.025
+  make_vocab_size_divisible_by: 1
+  model_config:
+    bos_token_id: 1
+    eos_token_id: 2
+    hidden_act: silu
+    hidden_size: 2048
+    initializer_range: 0.02
+    intermediate_size: 4096
+    is_llama_config: true
+    max_position_embeddings: 4096
+    num_attention_heads: 32
+    num_hidden_layers: 24
+    num_key_value_heads: 32
+    pad_token_id: null
+    pretraining_tp: 1
+    rms_norm_eps: 1.0e-05
+    rope_scaling: null
+    rope_theta: 10000.0
+    tie_word_embeddings: true
+    use_cache: true
+    vocab_size: 50257
+optimizer:
+  accumulate_grad_in_fp32: true
+  clip_grad: 1.0
+  learning_rate_scheduler:
+    learning_rate: 0.0001
+    lr_decay_style: linear
+    lr_warmup_style: linear
+    lr_warmup_steps: 1
+    min_decay_lr: 1.0e-05
+  optimizer_factory:
+    adam_beta1: 0.9
+    adam_beta2: 0.95
+    adam_eps: 1.0e-08
+    name: adamW
+    torch_adam_is_fused: true
+  weight_decay: 0.01
+  zero_stage: 1
+parallelism:
+  dp: 4
+  expert_parallel_size: 1
+  pp: 4
+  pp_engine: 1f1b
+  tp: 1
+  tp_linear_async_communication: false
+  tp_mode: REDUCE_SCATTER
+profiler:
+  profiler_export_path: /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2
+tokenizer:
+  tokenizer_max_length: null
+  tokenizer_name_or_path: openai-community/gpt2
+  tokenizer_revision: null
+data_stages:
+- name: Training Stage
+  start_training_step: 1
+  data:
+    dataset:
+      dataset_overwrite_cache: false
+      dataset_processing_num_proc_per_process: 64
+      hf_dataset_config_name: null
+      hf_dataset_or_datasets: roneneldan/TinyStories
+      hf_dataset_splits: train
+      text_column_name: text
+    num_loading_workers: 32
+    seed: 42
+lighteval: null
+tokens:
+  train_steps: 20
+  val_check_interval: -1
+  batch_accumulation_per_replica: 128
+  limit_test_batches: 0
+  limit_val_batches: 0
+  micro_batch_size: 2
+  sequence_length: 4096
+logging:
+  iteration_step_info_interval: 1
+  log_level: info
+  log_level_replica: info
+checkpoints:
+  checkpoint_interval: 100000
+  checkpoints_path: /dev/null
+  resume_checkpoint_path: null

llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/log.out

@@ -0,0 +1,308 @@
+========================
+START TIME: Tue Jul  2 15:28:58 UTC 2024
+python3 version = Python 3.10.14
+========================
+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.
+Token is valid (permission: write).
+Your token has been saved to /admin/home/ferdinand_mom/.cache/huggingface/token
+Login successful
+Already on 'bench_cluster'
+M	examples/config_tiny_llama.py
+M	examples/config_tiny_llama.yaml
+M	examples/train_tiny_llama.sh
+M	src/nanotron/models/llama.py
+M	src/nanotron/trainer.py
+Your branch is up to date with 'origin/bench_cluster'.
+Job status: RUNNING
+W0702 15:29:01.305000 140441428072256 torch/distributed/run.py:757] 
+W0702 15:29:01.305000 140441428072256 torch/distributed/run.py:757] *****************************************
+W0702 15:29:01.305000 140441428072256 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. 
+W0702 15:29:01.305000 140441428072256 torch/distributed/run.py:757] *****************************************
+W0702 15:29:01.391000 140542556596032 torch/distributed/run.py:757] 
+W0702 15:29:01.391000 140542556596032 torch/distributed/run.py:757] *****************************************
+W0702 15:29:01.391000 140542556596032 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. 
+W0702 15:29:01.391000 140542556596032 torch/distributed/run.py:757] *****************************************
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: Config:
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: Config(general=GeneralArgs(project='bench_cluster',
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                            run='%date_%jobid',
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                            seed=42,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                            step=None,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                            consumed_train_samples=None,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                            benchmark_csv_path=None,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                            ignore_sanity_checks=True),
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:        parallelism=ParallelismArgs(dp=4,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                    pp=4,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                    tp=1,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                    pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f07876f8520>,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                    tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                    tp_linear_async_communication=False,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                    expert_parallel_size=1),
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:        model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 eos_token_id=2,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 hidden_act='silu',
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 hidden_size=2048,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 initializer_range=0.02,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 intermediate_size=4096,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 is_llama_config=True,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 max_position_embeddings=4096,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 num_attention_heads=32,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 num_hidden_layers=24,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 num_key_value_heads=32,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 pad_token_id=None,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 pretraining_tp=1,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 rms_norm_eps=1e-05,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 rope_scaling=None,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 rope_theta=10000.0,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 tie_word_embeddings=True,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 use_cache=True,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                 vocab_size=50257),
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                        init_method=RandomInit(std=0.025),
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                        dtype=torch.bfloat16,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                        make_vocab_size_divisible_by=1,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                        ddp_bucket_cap_mb=25),
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:        tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                tokenizer_revision=None,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                tokenizer_max_length=None),
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:        checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                    checkpoint_interval=100000,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                    save_initial_state=False,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                    resume_checkpoint_path=None,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                    checkpoints_path_is_shared_file_system=False),
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:        logging=LoggingArgs(log_level='info',
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                            log_level_replica='info',
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                            iteration_step_info_interval=1),
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:        tokens=TokensArgs(sequence_length=4096,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                          train_steps=20,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                          micro_batch_size=2,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                          batch_accumulation_per_replica=128,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                          val_check_interval=-1,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                          limit_val_batches=0,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                          limit_test_batches=0),
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:        optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                                     adam_beta1=0.9,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                                     adam_beta2=0.95,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                                     torch_adam_is_fused=True,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                                     name='adamW'),
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                zero_stage=1,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                weight_decay=0.01,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                clip_grad=1.0,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                accumulate_grad_in_fp32=True,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                                        lr_warmup_steps=1,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                                        lr_warmup_style='linear',
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                                        lr_decay_style='linear',
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                                        lr_decay_steps=19,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                                        lr_decay_starting_step=None,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                                        min_decay_lr=1e-05)),
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:        data_stages=[DatasetStageArgs(name='Training Stage',
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                      start_training_step=1,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                      data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                                                 hf_dataset_splits='train',
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                                                 hf_dataset_config_name=None,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                                                 dataset_processing_num_proc_per_process=64,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                                                 dataset_overwrite_cache=False,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                                                 text_column_name='text'),
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                    seed=42,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:                                                    num_loading_workers=32))],
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:        profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2')),
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:        lighteval=None)
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: Model Config:
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: LlamaConfig(bos_token_id=1,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             eos_token_id=2,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             hidden_act='silu',
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             hidden_size=2048,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             initializer_range=0.02,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             intermediate_size=4096,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             is_llama_config=True,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             max_position_embeddings=4096,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             num_attention_heads=32,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             num_hidden_layers=24,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             num_key_value_heads=32,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             pad_token_id=None,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             pretraining_tp=1,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             rms_norm_eps=1e-05,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             rope_scaling=None,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             rope_theta=10000.0,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             tie_word_embeddings=True,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             use_cache=True,
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:             vocab_size=50257)
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: Building model..
+[default0]:07/02/2024 15:29:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: Setting PP block ranks...
+[default0]:07/02/2024 15:29:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: Total number of parameters: 1.21G (2312.82MiB)
+[default0]:07/02/2024 15:29:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: Local number of parameters: 397M (756.37MiB)
+[default0]:07/02/2024 15:29:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: [After model building] Memory usage: 763.38MiB. Peak allocated: 765.41MiB Peak reserved: 792.00MiB
+[default0]:07/02/2024 15:29:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: No checkpoint path provided.
+[default0]:07/02/2024 15:29:32 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: Parametrizing model parameters using StandardParametrizator
+[default4]:07/02/2024 15:29:32 [INFO|DP=0|PP=1|TP=0|ip-26-0-171-56]: Local number of parameters: 294M (560.05MiB)
+[default3]:07/02/2024 15:29:32 [INFO|DP=3|PP=0|TP=0|ip-26-0-171-56]: No checkpoint path provided.
+[default4]:07/02/2024 15:29:32 [INFO|DP=0|PP=1|TP=0|ip-26-0-171-56]: [After model building] Memory usage: 567.07MiB. Peak allocated: 569.10MiB Peak reserved: 594.00MiB
+[default4]:07/02/2024 15:29:32 [INFO|DP=0|PP=1|TP=0|ip-26-0-171-56]: No checkpoint path provided.
+[default7]:07/02/2024 15:29:32 [INFO|DP=3|PP=1|TP=0|ip-26-0-171-56]: No checkpoint path provided.
+[default0]:07/02/2024 15:29:32 [INFO|DP=0|PP=2|TP=0|ip-26-0-175-132]: Local number of parameters: 252M (480.05MiB)
+[default4]:07/02/2024 15:29:32 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: Local number of parameters: 271M (516.35MiB)
+[default4]:07/02/2024 15:29:32 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: [After model building] Memory usage: 520.36MiB. Peak allocated: 522.39MiB Peak reserved: 534.00MiB
+[default3]:07/02/2024 15:29:32 [INFO|DP=3|PP=2|TP=0|ip-26-0-175-132]: No checkpoint path provided.
+[default0]:07/02/2024 15:29:32 [INFO|DP=0|PP=2|TP=0|ip-26-0-175-132]: [After model building] Memory usage: 486.06MiB. Peak allocated: 488.09MiB Peak reserved: 502.00MiB
+[default4]:07/02/2024 15:29:32 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: No checkpoint path provided.
+[default0]:07/02/2024 15:29:32 [INFO|DP=0|PP=2|TP=0|ip-26-0-175-132]: No checkpoint path provided.
+[default7]:07/02/2024 15:29:32 [INFO|DP=3|PP=3|TP=0|ip-26-0-175-132]: No checkpoint path provided.
+[default6]:07/02/2024 15:29:32 [INFO|DP=2|PP=1|TP=0|ip-26-0-171-56]: No checkpoint path provided.
+[default2]:07/02/2024 15:29:32 [INFO|DP=2|PP=2|TP=0|ip-26-0-175-132]: No checkpoint path provided.
+[default1]:07/02/2024 15:29:32 [INFO|DP=1|PP=2|TP=0|ip-26-0-175-132]: No checkpoint path provided.
+[default5]:07/02/2024 15:29:32 [INFO|DP=1|PP=1|TP=0|ip-26-0-171-56]: No checkpoint path provided.
+[default5]:07/02/2024 15:29:32 [INFO|DP=1|PP=3|TP=0|ip-26-0-175-132]: No checkpoint path provided.
+[default1]:07/02/2024 15:29:32 [INFO|DP=1|PP=0|TP=0|ip-26-0-171-56]: No checkpoint path provided.
+[default2]:07/02/2024 15:29:32 [INFO|DP=2|PP=0|TP=0|ip-26-0-171-56]: No checkpoint path provided.
+[default6]:07/02/2024 15:29:32 [INFO|DP=2|PP=3|TP=0|ip-26-0-175-132]: No checkpoint path provided.
+[default0]:07/02/2024 15:29:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: [Optimizer Building] Using LearningRateForSP as learning rate
+[default0]:07/02/2024 15:29:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: [ZeRO sharding] Size of optimizer params per rank:
+[default0]:07/02/2024 15:29:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: [ZeRO sharding] DP Rank 0 has 99.1M out of 397M (25.00%) params' optimizer states
+[default0]:07/02/2024 15:29:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: [ZeRO sharding] DP Rank 1 has 99.1M out of 397M (25.00%) params' optimizer states
+[default0]:07/02/2024 15:29:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: [ZeRO sharding] DP Rank 2 has 99.1M out of 397M (25.00%) params' optimizer states
+[default0]:07/02/2024 15:29:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: [ZeRO sharding] DP Rank 3 has 99.1M out of 397M (25.00%) params' optimizer states
+[default0]:07/02/2024 15:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
+[default0]:07/02/2024 15:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: Using `datasets` library
+[default0]:07/02/2024 15:29:38 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
+[default0]:07/02/2024 15:29:38 [WARNING|DP=0|PP=0|TP=0|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
+[default0]:Repo card metadata block was not found. Setting CardData to empty.
+[default0]:07/02/2024 15:29:39 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: [Training Plan] There are 1 training stages 
+[default0]:07/02/2024 15:29:39 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: [Stage Training Stage] start from step 1 
+[default0]:07/02/2024 15:29:39 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: 
+[default0]:07/02/2024 15:29:39 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: [Start training] datetime: 2024-07-02 15:29:39.365195 | mbs: 2 | grad_accum: 128 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
+[default0]:07/02/2024 15:29:39 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
+[default0]:07/02/2024 15:29:39 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 2654.31MiB. Peak allocated 2654.31MiB. Peak reserved: 2686.00MiB
+[default1]:Repo card metadata block was not found. Setting CardData to empty.
+[default3]:07/02/2024 15:29:39 [WARNING|DP=3|PP=0|TP=0|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
+[default4]:07/02/2024 15:29:39 [WARNING|DP=0|PP=1|TP=0|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
+[default4]:07/02/2024 15:29:39 [WARNING|DP=0|PP=3|TP=0|ip-26-0-175-132]: Repo card metadata block was not found. Setting CardData to empty.
+[default0]:07/02/2024 15:29:39 [WARNING|DP=0|PP=2|TP=0|ip-26-0-175-132]: Repo card metadata block was not found. Setting CardData to empty.
+[default2]:Repo card metadata block was not found. Setting CardData to empty.
+[default2]:07/02/2024 15:29:39 [WARNING|DP=2|PP=2|TP=0|ip-26-0-175-132]: Repo card metadata block was not found. Setting CardData to empty.
+[default3]:07/02/2024 15:29:39 [WARNING|DP=3|PP=2|TP=0|ip-26-0-175-132]: Repo card metadata block was not found. Setting CardData to empty.
+[default1]:07/02/2024 15:29:39 [WARNING|DP=1|PP=2|TP=0|ip-26-0-175-132]: Repo card metadata block was not found. Setting CardData to empty.
+[default7]:07/02/2024 15:29:39 [WARNING|DP=3|PP=1|TP=0|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
+[default5]:07/02/2024 15:29:39 [WARNING|DP=1|PP=1|TP=0|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
+[default1]:07/02/2024 15:29:39 [WARNING|DP=1|PP=0|TP=0|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
+[default3]:Repo card metadata block was not found. Setting CardData to empty.
+[default5]:07/02/2024 15:29:39 [WARNING|DP=1|PP=3|TP=0|ip-26-0-175-132]: Repo card metadata block was not found. Setting CardData to empty.
+[default5]:Repo card metadata block was not found. Setting CardData to empty.
+[default5]:Repo card metadata block was not found. Setting CardData to empty.
+[default6]:07/02/2024 15:29:39 [WARNING|DP=2|PP=3|TP=0|ip-26-0-175-132]: Repo card metadata block was not found. Setting CardData to empty.
+[default7]:07/02/2024 15:29:39 [WARNING|DP=3|PP=3|TP=0|ip-26-0-175-132]: Repo card metadata block was not found. Setting CardData to empty.
+[default2]:07/02/2024 15:29:39 [WARNING|DP=2|PP=0|TP=0|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
+[default3]:Repo card metadata block was not found. Setting CardData to empty.
+[default1]:Repo card metadata block was not found. Setting CardData to empty.
+[default4]:Repo card metadata block was not found. Setting CardData to empty.
+[default7]:Repo card metadata block was not found. Setting CardData to empty.
+[default0]:Repo card metadata block was not found. Setting CardData to empty.
+[default2]:Repo card metadata block was not found. Setting CardData to empty.
+[default7]:Repo card metadata block was not found. Setting CardData to empty.
+[default6]:Repo card metadata block was not found. Setting CardData to empty.
+[default4]:Repo card metadata block was not found. Setting CardData to empty.
+[default6]:07/02/2024 15:29:39 [WARNING|DP=2|PP=1|TP=0|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
+[default6]:Repo card metadata block was not found. Setting CardData to empty.
+[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.)
+[default4]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default5]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default7]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default0]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default0]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default3]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default1]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default6]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default4]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default5]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default7]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default2]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default1]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default0]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default0]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default3]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default6]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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.)
+[default2]:  return Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
+[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
+[default5]:  warnings.warn(
+[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
+[default1]:  warnings.warn(
+[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
+[default7]:  warnings.warn(
+[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
+[default3]:  warnings.warn(
+[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
+[default4]:  warnings.warn(
+[default0]:07/02/2024 15:30:17 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 2721.88MiB. Peak allocated 19951.87MiB. Peak reserved: 20266.00MiB
+[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
+[default0]:  warnings.warn(
+[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
+[default6]:  warnings.warn(
+[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
+[default2]:  warnings.warn(
+[default4]:07/02/2024 15:30:19 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 1 / 20 | consumed_tokens: 4.19M | elapsed_time_per_iteration_ms: 39.1K | tokens_per_sec: 107K | tokens_per_sec_per_gpu: 6.71K | global_batch_size: 1.02K | lm_loss: 11.1 | lr: 0.0001 | model_tflops_per_gpu: 60.9 | hardware_tflops_per_gpu: 60.9 | grad_norm: 25.1 | cuda_memory_allocated: 2.51G | cuda_max_memory_reserved: 8.75G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.5G | hd_free_memory_tb: 247G
+[default0]:07/02/2024 15:30:19 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 5183.74MiB. Peak reserved: 21784.00MiB
+[default0]:07/02/2024 15:30:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:30:36 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 2 / 20 | consumed_tokens: 8.39M | elapsed_time_per_iteration_ms: 17K | tokens_per_sec: 246K | tokens_per_sec_per_gpu: 15.4K | global_batch_size: 1.02K | lm_loss: 11.1 | lr: 9.53e-05 | model_tflops_per_gpu: 140 | hardware_tflops_per_gpu: 140 | grad_norm: 25.2 | cuda_memory_allocated: 2.51G | cuda_max_memory_reserved: 8.75G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.5G | hd_free_memory_tb: 247G
+[default0]:07/02/2024 15:30:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 5183.74MiB. Peak reserved: 21786.00MiB
+[default0]:07/02/2024 15:30:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:30:54 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 3 / 20 | consumed_tokens: 12.6M | elapsed_time_per_iteration_ms: 18.5K | tokens_per_sec: 226K | tokens_per_sec_per_gpu: 14.1K | global_batch_size: 1.02K | lm_loss: 11.4 | lr: 9.05e-05 | model_tflops_per_gpu: 128 | hardware_tflops_per_gpu: 128 | grad_norm: 217 | cuda_memory_allocated: 2.51G | cuda_max_memory_reserved: 8.75G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.5G | hd_free_memory_tb: 247G
+[default0]:07/02/2024 15:30:54 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 5183.74MiB. Peak reserved: 21786.00MiB
+[default0]:STAGE:2024-07-02 15:30:54 3084719:3084719 ActivityProfilerController.cpp:314] Completed Stage: Warm Up
+[default0]:07/02/2024 15:31:12 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:31:13 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 4 / 20 | consumed_tokens: 16.8M | elapsed_time_per_iteration_ms: 19K | tokens_per_sec: 221K | tokens_per_sec_per_gpu: 13.8K | global_batch_size: 1.02K | lm_loss: 13.8 | lr: 8.58e-05 | model_tflops_per_gpu: 125 | hardware_tflops_per_gpu: 125 | grad_norm: 22.5 | cuda_memory_allocated: 2.51G | cuda_max_memory_reserved: 8.75G | hd_total_memory_tb: 312G | hd_used_memory_tb: 65.5G | hd_free_memory_tb: 247G
+[default0]:07/02/2024 15:31:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 5183.74MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:31:31 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 5 / 20 | consumed_tokens: 21M | elapsed_time_per_iteration_ms: 17.4K | tokens_per_sec: 242K | tokens_per_sec_per_gpu: 15.1K | global_batch_size: 1.02K | lm_loss: 9.98 | lr: 8.11e-05 | model_tflops_per_gpu: 137 | hardware_tflops_per_gpu: 137 | grad_norm: 16.4
+[default0]:07/02/2024 15:31:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:31:51 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 6 / 20 | consumed_tokens: 25.2M | elapsed_time_per_iteration_ms: 19.8K | tokens_per_sec: 212K | tokens_per_sec_per_gpu: 13.3K | global_batch_size: 1.02K | lm_loss: 10.9 | lr: 7.63e-05 | model_tflops_per_gpu: 120 | hardware_tflops_per_gpu: 120 | grad_norm: 93.9
+[default0]:STAGE:2024-07-02 15:32:15 3084719:3084719 ActivityProfilerController.cpp:320] Completed Stage: Collection
+[default0]:STAGE:2024-07-02 15:32:18 3084719:3084719 ActivityProfilerController.cpp:324] Completed Stage: Post Processing
+[default0]:07/02/2024 15:35:09 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default0]:07/02/2024 15:35:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:35:22 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 7 / 20 | consumed_tokens: 29.4M | elapsed_time_per_iteration_ms: 212K | tokens_per_sec: 19.8K | tokens_per_sec_per_gpu: 1.24K | global_batch_size: 1.02K | lm_loss: 9.16 | lr: 7.16e-05 | model_tflops_per_gpu: 11.2 | hardware_tflops_per_gpu: 11.2 | grad_norm: 19.8
+[default4]:07/02/2024 15:35:41 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 8 / 20 | consumed_tokens: 33.6M | elapsed_time_per_iteration_ms: 18.5K | tokens_per_sec: 227K | tokens_per_sec_per_gpu: 14.2K | global_batch_size: 1.02K | lm_loss: 8.83 | lr: 6.68e-05 | model_tflops_per_gpu: 129 | hardware_tflops_per_gpu: 129 | grad_norm: 6.08
+[default0]:07/02/2024 15:35:41 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:35:59 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 9 / 20 | consumed_tokens: 37.7M | elapsed_time_per_iteration_ms: 17.8K | tokens_per_sec: 236K | tokens_per_sec_per_gpu: 14.7K | global_batch_size: 1.02K | lm_loss: 8.47 | lr: 6.21e-05 | model_tflops_per_gpu: 134 | hardware_tflops_per_gpu: 134 | grad_norm: 5.23
+[default0]:07/02/2024 15:35:59 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:36:17 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 10 / 20 | consumed_tokens: 41.9M | elapsed_time_per_iteration_ms: 18.8K | tokens_per_sec: 224K | tokens_per_sec_per_gpu: 14K | global_batch_size: 1.02K | lm_loss: 8.17 | lr: 5.74e-05 | model_tflops_per_gpu: 127 | hardware_tflops_per_gpu: 127 | grad_norm: 7.71
+[default0]:07/02/2024 15:36:17 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:36:35 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 11 / 20 | consumed_tokens: 46.1M | elapsed_time_per_iteration_ms: 18K | tokens_per_sec: 233K | tokens_per_sec_per_gpu: 14.6K | global_batch_size: 1.02K | lm_loss: 7.93 | lr: 5.26e-05 | model_tflops_per_gpu: 132 | hardware_tflops_per_gpu: 132 | grad_norm: 5.55
+[default0]:07/02/2024 15:36:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:36:53 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 12 / 20 | consumed_tokens: 50.3M | elapsed_time_per_iteration_ms: 17.7K | tokens_per_sec: 237K | tokens_per_sec_per_gpu: 14.8K | global_batch_size: 1.02K | lm_loss: 7.75 | lr: 4.79e-05 | model_tflops_per_gpu: 135 | hardware_tflops_per_gpu: 135 | grad_norm: 4.65
+[default0]:07/02/2024 15:36:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default0]:07/02/2024 15:37:13 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:37:13 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 13 / 20 | consumed_tokens: 54.5M | elapsed_time_per_iteration_ms: 19.7K | tokens_per_sec: 213K | tokens_per_sec_per_gpu: 13.3K | global_batch_size: 1.02K | lm_loss: 7.58 | lr: 4.32e-05 | model_tflops_per_gpu: 121 | hardware_tflops_per_gpu: 121 | grad_norm: 2.9
+[default0]:07/02/2024 15:37:33 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:37:33 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 14 / 20 | consumed_tokens: 58.7M | elapsed_time_per_iteration_ms: 20.2K | tokens_per_sec: 208K | tokens_per_sec_per_gpu: 13K | global_batch_size: 1.02K | lm_loss: 7.5 | lr: 3.84e-05 | model_tflops_per_gpu: 118 | hardware_tflops_per_gpu: 118 | grad_norm: 4.18
+[default4]:07/02/2024 15:37:50 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 15 / 20 | consumed_tokens: 62.9M | elapsed_time_per_iteration_ms: 17K | tokens_per_sec: 247K | tokens_per_sec_per_gpu: 15.4K | global_batch_size: 1.02K | lm_loss: 7.4 | lr: 3.37e-05 | model_tflops_per_gpu: 140 | hardware_tflops_per_gpu: 140 | grad_norm: 3.86
+[default0]:07/02/2024 15:37:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default0]:07/02/2024 15:38:08 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:38:08 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 16 / 20 | consumed_tokens: 67.1M | elapsed_time_per_iteration_ms: 18.2K | tokens_per_sec: 231K | tokens_per_sec_per_gpu: 14.4K | global_batch_size: 1.02K | lm_loss: 7.29 | lr: 2.89e-05 | model_tflops_per_gpu: 131 | hardware_tflops_per_gpu: 131 | grad_norm: 3.07
+[default0]:07/02/2024 15:38:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:38:27 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 17 / 20 | consumed_tokens: 71.3M | elapsed_time_per_iteration_ms: 18.8K | tokens_per_sec: 223K | tokens_per_sec_per_gpu: 13.9K | global_batch_size: 1.02K | lm_loss: 7.19 | lr: 2.42e-05 | model_tflops_per_gpu: 126 | hardware_tflops_per_gpu: 126 | grad_norm: 2.39
+[default0]:07/02/2024 15:38:47 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:38:47 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 18 / 20 | consumed_tokens: 75.5M | elapsed_time_per_iteration_ms: 20.3K | tokens_per_sec: 207K | tokens_per_sec_per_gpu: 12.9K | global_batch_size: 1.02K | lm_loss: 7.13 | lr: 1.95e-05 | model_tflops_per_gpu: 117 | hardware_tflops_per_gpu: 117 | grad_norm: 2.2
+[default0]:07/02/2024 15:39:06 [INFO|DP=0|PP=0|TP=0|ip-26-0-171-56]:  Memory usage: 3478.27MiB. Peak allocated 20708.26MiB. Peak reserved: 21786.00MiB
+[default4]:07/02/2024 15:39:07 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 19 / 20 | consumed_tokens: 79.7M | elapsed_time_per_iteration_ms: 19.3K | tokens_per_sec: 217K | tokens_per_sec_per_gpu: 13.6K | global_batch_size: 1.02K | lm_loss: 7.08 | lr: 1.47e-05 | model_tflops_per_gpu: 123 | hardware_tflops_per_gpu: 123 | grad_norm: 2.64
+[default4]:07/02/2024 15:39:25 [INFO|DP=0|PP=3|TP=0|ip-26-0-175-132]: iteration: 20 / 20 | consumed_tokens: 83.9M | elapsed_time_per_iteration_ms: 18.3K | tokens_per_sec: 229K | tokens_per_sec_per_gpu: 14.3K | global_batch_size: 1.02K | lm_loss: 7.03 | lr: 1e-05 | model_tflops_per_gpu: 130 | hardware_tflops_per_gpu: 130 | grad_norm: 2.3
+W0702 15:39:58.574000 140542556596032 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1203] The node 'ip-26-0-175-132.ec2.internal_108785_0' has failed to shutdown the rendezvous 'none' due to an error of type RendezvousConnectionError.
+W0702 15:39:58.581000 140542556596032 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1203] The node 'ip-26-0-175-132.ec2.internal_108785_0' has failed to shutdown the rendezvous 'none' due to an error of type RendezvousConnectionError.
+Saved 1 csv files over 1 completed logs
+Processing file: /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/profiler/ip-26-0-171-56_3084719.1719934470978269987.pt.trace.json
+Results written to /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/profiler.csv
+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.
+

llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/log_metrics.csv

@@ -0,0 +1,21 @@
+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
+1,4190000.0000000005,39100.0,107000.0,6710.0,1020.0,11.1,0.0001,60.9,60.9,25.1,3478.27,20708.26,21786.0
+2,8390000.0,17000.0,246000.0,15400.0,1020.0,11.1,9.53e-05,140.0,140.0,25.2,3478.27,20708.26,21786.0
+3,12600000.0,18500.0,226000.0,14100.0,1020.0,11.4,9.05e-05,128.0,128.0,217.0,3478.27,20708.26,21786.0
+4,16800000.0,19000.0,221000.0,13800.0,1020.0,13.8,8.58e-05,125.0,125.0,22.5,3478.27,5183.74,21786.0
+5,21000000.0,17400.0,242000.0,15100.0,1020.0,9.98,8.11e-05,137.0,137.0,16.4,3478.27,20708.26,21786.0
+6,25200000.0,19800.0,212000.0,13300.0,1020.0,10.9,7.63e-05,120.0,120.0,93.9,3478.27,20708.26,21786.0
+7,29400000.0,212000.0,19800.0,1240.0,1020.0,9.16,7.16e-05,11.2,11.2,19.8,,,
+8,33600000.0,18500.0,227000.0,14200.0,1020.0,8.83,6.68e-05,129.0,129.0,6.08,3478.27,20708.26,21786.0
+9,37700000.0,17800.0,236000.0,14700.0,1020.0,8.47,6.21e-05,134.0,134.0,5.23,3478.27,20708.26,21786.0
+10,41900000.0,18800.0,224000.0,14000.0,1020.0,8.17,5.74e-05,127.0,127.0,7.71,3478.27,20708.26,21786.0
+11,46100000.0,18000.0,233000.0,14600.0,1020.0,7.93,5.26e-05,132.0,132.0,5.55,3478.27,20708.26,21786.0
+12,50300000.0,17700.0,237000.0,14800.0,1020.0,7.75,4.79e-05,135.0,135.0,4.65,3478.27,20708.26,21786.0
+13,54500000.0,19700.0,213000.0,13300.0,1020.0,7.58,4.32e-05,121.0,121.0,2.9,3478.27,20708.26,21786.0
+14,58700000.0,20200.0,208000.0,13000.0,1020.0,7.5,3.84e-05,118.0,118.0,4.18,,,
+15,62900000.0,17000.0,247000.0,15400.0,1020.0,7.4,3.37e-05,140.0,140.0,3.86,3478.27,20708.26,21786.0
+16,67099999.99999999,18200.0,231000.0,14400.0,1020.0,7.29,2.89e-05,131.0,131.0,3.07,3478.27,20708.26,21786.0
+17,71300000.0,18800.0,223000.0,13900.0,1020.0,7.19,2.42e-05,126.0,126.0,2.39,3478.27,20708.26,21786.0
+18,75500000.0,20300.0,207000.0,12900.0,1020.0,7.13,1.95e-05,117.0,117.0,2.2,3478.27,20708.26,21786.0
+19,79700000.0,19300.0,217000.0,13600.0,1020.0,7.08,1.47e-05,123.0,123.0,2.64,,,
+20,83900000.0,18300.0,229000.0,14300.0,1020.0,7.03,1e-05,130.0,130.0,2.3,,,

llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/profiler.csv

@@ -0,0 +1,2 @@
+forward,backward
+0ms 890μs,1ms 193μs

llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/profiler/ip-26-0-171-56_3084719.1719934470978269987.pt.trace.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:22559333a23a2e6a0a9a18a34c8ba6306c5c513ae3fe602eee1fccb72ae6dac8
+size 5193950887

llama-1B/16_GPUS/dp-4_tp-1_pp-4_mbz-2/status.txt

@@ -0,0 +1 @@
+completed