Upload llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32
Browse files
llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32/bench.slurm
ADDED
|
@@ -0,0 +1,111 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/bin/bash
|
| 2 |
+
|
| 3 |
+
#SBATCH --job-name=bench_cluster
|
| 4 |
+
#SBATCH --time=02:00:00
|
| 5 |
+
#SBATCH --partition=hopper-prod
|
| 6 |
+
#SBATCH --nodes=1
|
| 7 |
+
#SBATCH --gres=gpu:8
|
| 8 |
+
#SBATCH --qos=normal
|
| 9 |
+
#SBATCH --ntasks-per-node=1
|
| 10 |
+
#SBATCH --cpus-per-task=96
|
| 11 |
+
#SBATCH --exclusive
|
| 12 |
+
#SBATCH --output=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32/log.out
|
| 13 |
+
#SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32/log.out
|
| 14 |
+
|
| 15 |
+
# Function to update status based on squeue output
|
| 16 |
+
update_status() {
|
| 17 |
+
job_id=$1
|
| 18 |
+
status_file=$2
|
| 19 |
+
# For unknown reasons, it doenst update status for pending. It only works for running
|
| 20 |
+
while true; do
|
| 21 |
+
job_status=$(squeue --job $job_id --noheader --format=%T)
|
| 22 |
+
echo "Job status: $job_status"
|
| 23 |
+
if [ -z "$job_status" ]; then
|
| 24 |
+
# Job has finished or is not found
|
| 25 |
+
break
|
| 26 |
+
elif [ "$job_status" = "RUNNING" ]; then
|
| 27 |
+
printf "running" > $status_file
|
| 28 |
+
break
|
| 29 |
+
fi
|
| 30 |
+
sleep 10
|
| 31 |
+
done
|
| 32 |
+
}
|
| 33 |
+
|
| 34 |
+
# Misc initializations.
|
| 35 |
+
echo "========================"
|
| 36 |
+
echo "START TIME: $(date)"
|
| 37 |
+
source /fsx/ferdinandmom/miniforge3/etc/profile.d/conda.sh
|
| 38 |
+
conda activate /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster
|
| 39 |
+
echo python3 version = $(python3 --version)
|
| 40 |
+
echo "========================"
|
| 41 |
+
|
| 42 |
+
# Slurm stuff
|
| 43 |
+
export HOSTNAMES=$(scontrol show hostnames "$SLURM_JOB_NODELIST")
|
| 44 |
+
export MASTER_ADDR=$(scontrol show hostnames "$SLURM_JOB_NODELIST" | head -n 1)
|
| 45 |
+
export MASTER_PORT=$((1024 + RANDOM % 64511))
|
| 46 |
+
|
| 47 |
+
export TMPDIR=/scratch
|
| 48 |
+
export HF_DATASETS_CACHE="/admin/home/ferdinand_mom/.cache"
|
| 49 |
+
export CUBLAS_WORKSPACE_CONFIG=":4096:8"
|
| 50 |
+
export CUDA_DEVICE_MAX_CONNECTIONS="1"
|
| 51 |
+
|
| 52 |
+
huggingface-cli login --token $HUGGINGFACE_TOKEN
|
| 53 |
+
|
| 54 |
+
|
| 55 |
+
NANOTRON_REPO="/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron"
|
| 56 |
+
CMD="$NANOTRON_REPO/run_train.py --config-file /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32/config.yaml"
|
| 57 |
+
|
| 58 |
+
LAUNCHER="torchrun \
|
| 59 |
+
--nproc_per_node 8 \
|
| 60 |
+
--nnodes 1 \
|
| 61 |
+
--rdzv_endpoint ${MASTER_ADDR}:${MASTER_PORT} \
|
| 62 |
+
--rdzv_backend c10d \
|
| 63 |
+
--max_restarts 0 \
|
| 64 |
+
--tee 3 \
|
| 65 |
+
--node_rank ${SLURM_PROCID}"
|
| 66 |
+
|
| 67 |
+
# Checkout the bench_cluster branch
|
| 68 |
+
cd $NANOTRON_REPO
|
| 69 |
+
git checkout bench_cluster
|
| 70 |
+
cd ..
|
| 71 |
+
# Get the current job ID
|
| 72 |
+
job_id=${SLURM_JOB_ID}
|
| 73 |
+
|
| 74 |
+
# Update status to "pending" or "running" in the background
|
| 75 |
+
update_status $job_id /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32/status.txt &
|
| 76 |
+
|
| 77 |
+
# Run the main command
|
| 78 |
+
srun -u $LAUNCHER $CMD
|
| 79 |
+
exit_status=$?
|
| 80 |
+
|
| 81 |
+
# Update status based on the exit status of `srun`
|
| 82 |
+
if [ $exit_status -eq 0 ]; then
|
| 83 |
+
printf "completed" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32/status.txt
|
| 84 |
+
else
|
| 85 |
+
if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32/log.out; then
|
| 86 |
+
printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32/status.txt
|
| 87 |
+
elif grep -q " CUDA error: an illegal memory access" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32/log.out; then
|
| 88 |
+
printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32/status.txt
|
| 89 |
+
elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32/log.out; then
|
| 90 |
+
printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32/status.txt
|
| 91 |
+
else
|
| 92 |
+
printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32/status.txt
|
| 93 |
+
fi
|
| 94 |
+
fi
|
| 95 |
+
|
| 96 |
+
# Run the report script if the job completed successfully
|
| 97 |
+
if [ $exit_status -eq 0 ]; then
|
| 98 |
+
python /fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py report --inp_dir /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32 --is_logs
|
| 99 |
+
python /fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py report --inp_dir /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32 --is_profiler
|
| 100 |
+
fi
|
| 101 |
+
|
| 102 |
+
|
| 103 |
+
# Push to hub the folder using huggingface_cli
|
| 104 |
+
huggingface-cli upload nanotron/bench_cluster /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32 llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32 --commit-message "Upload llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32"
|
| 105 |
+
|
| 106 |
+
# Verify the upload
|
| 107 |
+
if [ $? -eq 0 ]; then
|
| 108 |
+
echo "Uploading to Huggingface Hub successful"
|
| 109 |
+
else
|
| 110 |
+
echo "Failed to upload to Huggingface Hub"
|
| 111 |
+
fi
|
llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32/config.yaml
ADDED
|
@@ -0,0 +1,90 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
general:
|
| 2 |
+
project: bench_cluster
|
| 3 |
+
seed: 42
|
| 4 |
+
model:
|
| 5 |
+
ddp_bucket_cap_mb: 25
|
| 6 |
+
dtype: bfloat16
|
| 7 |
+
init_method:
|
| 8 |
+
std: 0.025
|
| 9 |
+
make_vocab_size_divisible_by: 1
|
| 10 |
+
model_config:
|
| 11 |
+
bos_token_id: 1
|
| 12 |
+
eos_token_id: 2
|
| 13 |
+
hidden_act: silu
|
| 14 |
+
hidden_size: 2048
|
| 15 |
+
initializer_range: 0.02
|
| 16 |
+
intermediate_size: 4096
|
| 17 |
+
is_llama_config: true
|
| 18 |
+
max_position_embeddings: 4096
|
| 19 |
+
num_attention_heads: 32
|
| 20 |
+
num_hidden_layers: 24
|
| 21 |
+
num_key_value_heads: 32
|
| 22 |
+
pad_token_id: null
|
| 23 |
+
pretraining_tp: 1
|
| 24 |
+
rms_norm_eps: 1.0e-05
|
| 25 |
+
rope_scaling: null
|
| 26 |
+
rope_theta: 10000.0
|
| 27 |
+
tie_word_embeddings: true
|
| 28 |
+
use_cache: true
|
| 29 |
+
vocab_size: 50257
|
| 30 |
+
optimizer:
|
| 31 |
+
accumulate_grad_in_fp32: true
|
| 32 |
+
clip_grad: 1.0
|
| 33 |
+
learning_rate_scheduler:
|
| 34 |
+
learning_rate: 0.0001
|
| 35 |
+
lr_decay_style: linear
|
| 36 |
+
lr_warmup_style: linear
|
| 37 |
+
lr_warmup_steps: 1
|
| 38 |
+
min_decay_lr: 1.0e-05
|
| 39 |
+
optimizer_factory:
|
| 40 |
+
adam_beta1: 0.9
|
| 41 |
+
adam_beta2: 0.95
|
| 42 |
+
adam_eps: 1.0e-08
|
| 43 |
+
name: adamW
|
| 44 |
+
torch_adam_is_fused: true
|
| 45 |
+
weight_decay: 0.01
|
| 46 |
+
zero_stage: 1
|
| 47 |
+
parallelism:
|
| 48 |
+
dp: 2
|
| 49 |
+
expert_parallel_size: 1
|
| 50 |
+
pp: 1
|
| 51 |
+
pp_engine: 1f1b
|
| 52 |
+
tp: 4
|
| 53 |
+
tp_linear_async_communication: false
|
| 54 |
+
tp_mode: REDUCE_SCATTER
|
| 55 |
+
profiler:
|
| 56 |
+
profiler_export_path: /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32
|
| 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: 16
|
| 79 |
+
limit_test_batches: 0
|
| 80 |
+
limit_val_batches: 0
|
| 81 |
+
micro_batch_size: 32
|
| 82 |
+
sequence_length: 4096
|
| 83 |
+
logging:
|
| 84 |
+
iteration_step_info_interval: 1
|
| 85 |
+
log_level: info
|
| 86 |
+
log_level_replica: info
|
| 87 |
+
checkpoints:
|
| 88 |
+
checkpoint_interval: 100000
|
| 89 |
+
checkpoints_path: /dev/null
|
| 90 |
+
resume_checkpoint_path: null
|
llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32/log.out
ADDED
|
@@ -0,0 +1,664 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
========================
|
| 2 |
+
START TIME: Wed Jul 3 23:45:41 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 |
+
fatal: Unable to create '/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/.git/index.lock': File exists.
|
| 10 |
+
|
| 11 |
+
Another git process seems to be running in this repository, e.g.
|
| 12 |
+
an editor opened by 'git commit'. Please make sure all processes
|
| 13 |
+
are terminated then try again. If it still fails, a git process
|
| 14 |
+
may have crashed in this repository earlier:
|
| 15 |
+
remove the file manually to continue.
|
| 16 |
+
Job status: RUNNING
|
| 17 |
+
W0703 23:45:49.311000 139891705947968 torch/distributed/run.py:757]
|
| 18 |
+
W0703 23:45:49.311000 139891705947968 torch/distributed/run.py:757] *****************************************
|
| 19 |
+
W0703 23:45:49.311000 139891705947968 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 23:45:49.311000 139891705947968 torch/distributed/run.py:757] *****************************************
|
| 21 |
+
[default0]:07/03/2024 23:46:10 [WARNING|DP=0|PP=0|TP=0|ip-26-0-164-187]: [Vocab Size Padding] Padded vocab (size: 50257) with 3 dummy tokens (new size: 50260)
|
| 22 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Config:
|
| 23 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Config(general=GeneralArgs(project='bench_cluster',
|
| 24 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: run='%date_%jobid',
|
| 25 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: seed=42,
|
| 26 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: step=None,
|
| 27 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: consumed_train_samples=None,
|
| 28 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: benchmark_csv_path=None,
|
| 29 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: ignore_sanity_checks=True),
|
| 30 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: parallelism=ParallelismArgs(dp=2,
|
| 31 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: pp=1,
|
| 32 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tp=4,
|
| 33 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f147f838880>,
|
| 34 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
|
| 35 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tp_linear_async_communication=False,
|
| 36 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: expert_parallel_size=1),
|
| 37 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
|
| 38 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: eos_token_id=2,
|
| 39 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: hidden_act='silu',
|
| 40 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: hidden_size=2048,
|
| 41 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: initializer_range=0.02,
|
| 42 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: intermediate_size=4096,
|
| 43 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: is_llama_config=True,
|
| 44 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: max_position_embeddings=4096,
|
| 45 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: num_attention_heads=32,
|
| 46 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: num_hidden_layers=24,
|
| 47 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: num_key_value_heads=32,
|
| 48 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: pad_token_id=None,
|
| 49 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: pretraining_tp=1,
|
| 50 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: rms_norm_eps=1e-05,
|
| 51 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: rope_scaling=None,
|
| 52 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: rope_theta=10000.0,
|
| 53 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tie_word_embeddings=True,
|
| 54 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: use_cache=True,
|
| 55 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: vocab_size=50260),
|
| 56 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: init_method=RandomInit(std=0.025),
|
| 57 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: dtype=torch.bfloat16,
|
| 58 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: make_vocab_size_divisible_by=1,
|
| 59 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: ddp_bucket_cap_mb=25),
|
| 60 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
|
| 61 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tokenizer_revision=None,
|
| 62 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tokenizer_max_length=None),
|
| 63 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
|
| 64 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: checkpoint_interval=100000,
|
| 65 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: save_initial_state=False,
|
| 66 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: resume_checkpoint_path=None,
|
| 67 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: checkpoints_path_is_shared_file_system=False),
|
| 68 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: logging=LoggingArgs(log_level='info',
|
| 69 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: log_level_replica='info',
|
| 70 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: iteration_step_info_interval=1),
|
| 71 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tokens=TokensArgs(sequence_length=4096,
|
| 72 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: train_steps=20,
|
| 73 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: micro_batch_size=32,
|
| 74 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: batch_accumulation_per_replica=16,
|
| 75 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: val_check_interval=-1,
|
| 76 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: limit_val_batches=0,
|
| 77 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: limit_test_batches=0),
|
| 78 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
|
| 79 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: adam_beta1=0.9,
|
| 80 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: adam_beta2=0.95,
|
| 81 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: torch_adam_is_fused=True,
|
| 82 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: name='adamW'),
|
| 83 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: zero_stage=1,
|
| 84 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: weight_decay=0.01,
|
| 85 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: clip_grad=1.0,
|
| 86 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: accumulate_grad_in_fp32=True,
|
| 87 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
|
| 88 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: lr_warmup_steps=1,
|
| 89 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: lr_warmup_style='linear',
|
| 90 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: lr_decay_style='linear',
|
| 91 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: lr_decay_steps=19,
|
| 92 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: lr_decay_starting_step=None,
|
| 93 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: min_decay_lr=1e-05)),
|
| 94 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: data_stages=[DatasetStageArgs(name='Training Stage',
|
| 95 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: start_training_step=1,
|
| 96 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
|
| 97 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: hf_dataset_splits='train',
|
| 98 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: hf_dataset_config_name=None,
|
| 99 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: dataset_processing_num_proc_per_process=64,
|
| 100 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: dataset_overwrite_cache=False,
|
| 101 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: text_column_name='text'),
|
| 102 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: seed=42,
|
| 103 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: num_loading_workers=0))],
|
| 104 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-2_tp-4_pp-1_mbz-32')),
|
| 105 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: lighteval=None)
|
| 106 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Model Config:
|
| 107 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: LlamaConfig(bos_token_id=1,
|
| 108 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: eos_token_id=2,
|
| 109 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: hidden_act='silu',
|
| 110 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: hidden_size=2048,
|
| 111 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: initializer_range=0.02,
|
| 112 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: intermediate_size=4096,
|
| 113 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: is_llama_config=True,
|
| 114 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: max_position_embeddings=4096,
|
| 115 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: num_attention_heads=32,
|
| 116 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: num_hidden_layers=24,
|
| 117 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: num_key_value_heads=32,
|
| 118 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: pad_token_id=None,
|
| 119 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: pretraining_tp=1,
|
| 120 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: rms_norm_eps=1e-05,
|
| 121 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: rope_scaling=None,
|
| 122 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: rope_theta=10000.0,
|
| 123 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tie_word_embeddings=True,
|
| 124 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: use_cache=True,
|
| 125 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: vocab_size=50260)
|
| 126 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Building model..
|
| 127 |
+
[default0]:07/03/2024 23:46:10 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Setting PP block ranks...
|
| 128 |
+
[default0]:07/03/2024 23:46:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Total number of parameters: 1.11G (2117.09MiB)
|
| 129 |
+
[default0]:07/03/2024 23:46:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Local number of parameters: 277M (529.27MiB)
|
| 130 |
+
[default0]:07/03/2024 23:46:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [After model building] Memory usage: 554.21MiB. Peak allocated: 606.24MiB Peak reserved: 608.00MiB
|
| 131 |
+
[default0]:07/03/2024 23:46:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: No checkpoint path provided.
|
| 132 |
+
[default0]:07/03/2024 23:46:23 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Parametrizing model parameters using StandardParametrizator
|
| 133 |
+
[default2]:07/03/2024 23:46:23 [INFO|DP=0|PP=0|TP=2|ip-26-0-164-187]: Local number of parameters: 277M (529.27MiB)
|
| 134 |
+
[default2]:07/03/2024 23:46:23 [INFO|DP=0|PP=0|TP=2|ip-26-0-164-187]: [After model building] Memory usage: 554.21MiB. Peak allocated: 606.24MiB Peak reserved: 608.00MiB
|
| 135 |
+
[default2]:07/03/2024 23:46:23 [INFO|DP=0|PP=0|TP=2|ip-26-0-164-187]: No checkpoint path provided.
|
| 136 |
+
[default1]:07/03/2024 23:46:23 [INFO|DP=0|PP=0|TP=1|ip-26-0-164-187]: Local number of parameters: 277M (529.27MiB)
|
| 137 |
+
[default1]:07/03/2024 23:46:23 [INFO|DP=0|PP=0|TP=1|ip-26-0-164-187]: [After model building] Memory usage: 554.21MiB. Peak allocated: 606.24MiB Peak reserved: 608.00MiB
|
| 138 |
+
[default1]:07/03/2024 23:46:23 [INFO|DP=0|PP=0|TP=1|ip-26-0-164-187]: No checkpoint path provided.
|
| 139 |
+
[default3]:07/03/2024 23:46:23 [INFO|DP=0|PP=0|TP=3|ip-26-0-164-187]: Local number of parameters: 277M (529.27MiB)
|
| 140 |
+
[default3]:07/03/2024 23:46:23 [INFO|DP=0|PP=0|TP=3|ip-26-0-164-187]: [After model building] Memory usage: 554.21MiB. Peak allocated: 606.24MiB Peak reserved: 608.00MiB
|
| 141 |
+
[default3]:07/03/2024 23:46:23 [INFO|DP=0|PP=0|TP=3|ip-26-0-164-187]: No checkpoint path provided.
|
| 142 |
+
[default4]:07/03/2024 23:46:23 [INFO|DP=1|PP=0|TP=0|ip-26-0-164-187]: No checkpoint path provided.
|
| 143 |
+
[default6]:07/03/2024 23:46:23 [INFO|DP=1|PP=0|TP=2|ip-26-0-164-187]: No checkpoint path provided.
|
| 144 |
+
[default7]:07/03/2024 23:46:23 [INFO|DP=1|PP=0|TP=3|ip-26-0-164-187]: No checkpoint path provided.
|
| 145 |
+
[default5]:07/03/2024 23:46:23 [INFO|DP=1|PP=0|TP=1|ip-26-0-164-187]: No checkpoint path provided.
|
| 146 |
+
[default0]:07/03/2024 23:46:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [Optimizer Building] Using LearningRateForSP as learning rate
|
| 147 |
+
[default0]:07/03/2024 23:46:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [ZeRO sharding] Size of optimizer params per rank:
|
| 148 |
+
[default0]:07/03/2024 23:46:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [ZeRO sharding] DP Rank 0 has 139M out of 277M (50.00%) params' optimizer states
|
| 149 |
+
[default0]:07/03/2024 23:46:26 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [ZeRO sharding] DP Rank 1 has 139M out of 277M (50.00%) params' optimizer states
|
| 150 |
+
[default0]:07/03/2024 23:46:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
|
| 151 |
+
[default0]:07/03/2024 23:46:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Using `datasets` library
|
| 152 |
+
[default0]:07/03/2024 23:46:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
|
| 153 |
+
[default0]:07/03/2024 23:46:27 [WARNING|DP=0|PP=0|TP=0|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 154 |
+
[default0]:Repo card metadata block was not found. Setting CardData to empty.
|
| 155 |
+
[default0]:07/03/2024 23:46:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [Training Plan] There are 1 training stages
|
| 156 |
+
[default0]:07/03/2024 23:46:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [Stage Training Stage] start from step 1
|
| 157 |
+
[default0]:07/03/2024 23:46:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]:
|
| 158 |
+
[default0]:07/03/2024 23:46:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [Start training] datetime: 2024-07-03 23:46:29.945526 | mbs: 32 | grad_accum: 16 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
|
| 159 |
+
[default0]:07/03/2024 23:46:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
|
| 160 |
+
[default0]:07/03/2024 23:46:29 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Memory usage: 2142.76MiB. Peak allocated 2142.76MiB. Peak reserved: 2198.00MiB
|
| 161 |
+
[default2]:07/03/2024 23:46:30 [WARNING|DP=0|PP=0|TP=2|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 162 |
+
[default5]:Repo card metadata block was not found. Setting CardData to empty.
|
| 163 |
+
[default1]:Repo card metadata block was not found. Setting CardData to empty.
|
| 164 |
+
[default4]:07/03/2024 23:46:30 [WARNING|DP=1|PP=0|TP=0|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 165 |
+
[default1]:07/03/2024 23:46:30 [WARNING|DP=0|PP=0|TP=1|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 166 |
+
[default6]:07/03/2024 23:46:30 [WARNING|DP=1|PP=0|TP=2|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 167 |
+
[default2]:Repo card metadata block was not found. Setting CardData to empty.
|
| 168 |
+
[default7]:07/03/2024 23:46:30 [WARNING|DP=1|PP=0|TP=3|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 169 |
+
[default3]:07/03/2024 23:46:30 [WARNING|DP=0|PP=0|TP=3|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 170 |
+
[default5]:07/03/2024 23:46:30 [WARNING|DP=1|PP=0|TP=1|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 171 |
+
[default7]:Repo card metadata block was not found. Setting CardData to empty.
|
| 172 |
+
[default3]:Repo card metadata block was not found. Setting CardData to empty.
|
| 173 |
+
[default6]:Repo card metadata block was not found. Setting CardData to empty.
|
| 174 |
+
[default4]:Repo card metadata block was not found. Setting CardData to empty.
|
| 175 |
+
[default2]:[rank2]: Traceback (most recent call last):
|
| 176 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 177 |
+
[default2]:[rank2]: trainer.train(dataloader)
|
| 178 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 179 |
+
[default2]:[rank2]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 180 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 181 |
+
[default2]:[rank2]: outputs = self.pipeline_engine.train_batch_iter(
|
| 182 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 183 |
+
[default2]:[rank2]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 184 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 185 |
+
[default2]:[rank2]: output = model(**micro_batch)
|
| 186 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 187 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 188 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 189 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 190 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 191 |
+
[default2]:[rank2]: sharded_logits = self.model(
|
| 192 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 193 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 194 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 195 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 196 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 197 |
+
[default2]:[rank2]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 198 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 199 |
+
[default2]:[rank2]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 200 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 201 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 202 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 203 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 204 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 205 |
+
[default2]:[rank2]: output = self.pp_block(**new_kwargs)
|
| 206 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 207 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 208 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 209 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 210 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 211 |
+
[default2]:[rank2]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 212 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 213 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 214 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 215 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 216 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 217 |
+
[default2]:[rank2]: merged_states = self.gate_up_proj(hidden_states)
|
| 218 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 219 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 220 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 221 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 222 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 223 |
+
[default2]:[rank2]: return column_linear(
|
| 224 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 225 |
+
[default2]:[rank2]: return F.linear(input, weight, bias)
|
| 226 |
+
[default2]:[rank2]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 512.00 MiB. GPU has a total capacity of 79.33 GiB of which 505.94 MiB is free. Including non-PyTorch memory, this process has 78.82 GiB memory in use. Of the allocated memory 67.10 GiB is allocated by PyTorch, and 272.09 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
|
| 227 |
+
[default1]:[rank1]: Traceback (most recent call last):
|
| 228 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 229 |
+
[default1]:[rank1]: trainer.train(dataloader)
|
| 230 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 231 |
+
[default1]:[rank1]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 232 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 233 |
+
[default1]:[rank1]: outputs = self.pipeline_engine.train_batch_iter(
|
| 234 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 235 |
+
[default1]:[rank1]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 236 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 237 |
+
[default1]:[rank1]: output = model(**micro_batch)
|
| 238 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 239 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 240 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 241 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 242 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 243 |
+
[default1]:[rank1]: sharded_logits = self.model(
|
| 244 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 245 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 246 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 247 |
+
[default0]:[rank0]: Traceback (most recent call last):
|
| 248 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 249 |
+
[default0]:[rank0]: trainer.train(dataloader)
|
| 250 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 251 |
+
[default0]:[rank0]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 252 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 253 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 254 |
+
[default1]:[rank1]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 255 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 256 |
+
[default1]:[rank1]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 257 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 258 |
+
[default0]:[rank0]: outputs = self.pipeline_engine.train_batch_iter(
|
| 259 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 260 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 261 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 262 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 263 |
+
[default0]:[rank0]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 264 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 265 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 266 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 267 |
+
[default1]:[rank1]: output = self.pp_block(**new_kwargs)
|
| 268 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 269 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 270 |
+
[default0]:[rank0]: output = model(**micro_batch)
|
| 271 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 272 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 273 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 274 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 275 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 276 |
+
[default1]:[rank1]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 277 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 278 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 279 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 280 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 281 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 282 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 283 |
+
[default0]:[rank0]: sharded_logits = self.model(
|
| 284 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 285 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 286 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 287 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 288 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 289 |
+
[default1]:[rank1]: merged_states = self.gate_up_proj(hidden_states)
|
| 290 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 291 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 292 |
+
[default3]:[rank3]: Traceback (most recent call last):
|
| 293 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 294 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 295 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 296 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 297 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 298 |
+
[default0]:[rank0]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 299 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 300 |
+
[default1]:[rank1]: return column_linear(
|
| 301 |
+
[default3]:[rank3]: trainer.train(dataloader)
|
| 302 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 303 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 304 |
+
[default0]:[rank0]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 305 |
+
[default1]:[rank1]: return F.linear(input, weight, bias)
|
| 306 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 307 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 308 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 309 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 310 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 311 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 312 |
+
[default0]:[rank0]: output = self.pp_block(**new_kwargs)
|
| 313 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 314 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 315 |
+
[default1]:[rank1]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 512.00 MiB. GPU has a total capacity of 79.33 GiB of which 505.94 MiB is free. Including non-PyTorch memory, this process has 78.82 GiB memory in use. Of the allocated memory 67.10 GiB is allocated by PyTorch, and 272.09 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
|
| 316 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 317 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 318 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 319 |
+
[default0]:[rank0]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 320 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 321 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 322 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 323 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 324 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
|
| 325 |
+
[default0]:[rank0]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 326 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 327 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 328 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 329 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 330 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 128, in forward
|
| 331 |
+
[default0]:[rank0]: return self.act(gate_states) * up_states
|
| 332 |
+
[default0]:[rank0]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 256.00 MiB. GPU
|
| 333 |
+
[default3]:[rank3]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 334 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 335 |
+
[default3]:[rank3]: outputs = self.pipeline_engine.train_batch_iter(
|
| 336 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 337 |
+
[default3]:[rank3]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 338 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 339 |
+
[default3]:[rank3]: output = model(**micro_batch)
|
| 340 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 341 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 342 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 343 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 344 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 345 |
+
[default3]:[rank3]: sharded_logits = self.model(
|
| 346 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 347 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 348 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 349 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 350 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 351 |
+
[default3]:[rank3]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 352 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 353 |
+
[default3]:[rank3]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 354 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 355 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 356 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 357 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 358 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 359 |
+
[default3]:[rank3]: output = self.pp_block(**new_kwargs)
|
| 360 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 361 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 362 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 363 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 364 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 365 |
+
[default3]:[rank3]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 366 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 367 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 368 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 369 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 370 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
|
| 371 |
+
[default3]:[rank3]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 372 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 373 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 374 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 375 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 376 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 128, in forward
|
| 377 |
+
[default3]:[rank3]: return self.act(gate_states) * up_states
|
| 378 |
+
[default3]:[rank3]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 256.00 MiB. GPU has a total capacity of 79.33 GiB of which 233.94 MiB is free. Including non-PyTorch memory, this process has 79.09 GiB memory in use. Of the allocated memory 67.85 GiB is allocated by PyTorch, and 16.09 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
|
| 379 |
+
[default5]:[rank5]: Traceback (most recent call last):
|
| 380 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 381 |
+
[default5]:[rank5]: trainer.train(dataloader)
|
| 382 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 383 |
+
[default5]:[rank5]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 384 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 385 |
+
[default5]:[rank5]: outputs = self.pipeline_engine.train_batch_iter(
|
| 386 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 387 |
+
[default5]:[rank5]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 388 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 389 |
+
[default5]:[rank5]: output = model(**micro_batch)
|
| 390 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 391 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 392 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 393 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 394 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 395 |
+
[default5]:[rank5]: sharded_logits = self.model(
|
| 396 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 397 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 398 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 399 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 400 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 401 |
+
[default5]:[rank5]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 402 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 403 |
+
[default5]:[rank5]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 404 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 405 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 406 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 407 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 408 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 409 |
+
[default5]:[rank5]: output = self.pp_block(**new_kwargs)
|
| 410 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 411 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 412 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 413 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 414 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 415 |
+
[default5]:[rank5]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 416 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 417 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 418 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 419 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 420 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 421 |
+
[default5]:[rank5]: merged_states = self.gate_up_proj(hidden_states)
|
| 422 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 423 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 424 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 425 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 426 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 427 |
+
[default5]:[rank5]: return column_linear(
|
| 428 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 429 |
+
[default5]:[rank5]: return F.linear(input, weight, bias)
|
| 430 |
+
[default5]:[rank5]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 512.00 MiB. GPU has a total capacity of 79.33 GiB of which 505.94 MiB is free. Including non-PyTorch memory, this process has 78.82 GiB memory in use. Of the allocated memory 67.10 GiB is allocated by PyTorch, and 272.09 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
|
| 431 |
+
[default6]:[rank6]: Traceback (most recent call last):
|
| 432 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 433 |
+
[default6]:[rank6]: trainer.train(dataloader)
|
| 434 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 435 |
+
[default6]:[rank6]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 436 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 437 |
+
[default6]:[rank6]: outputs = self.pipeline_engine.train_batch_iter(
|
| 438 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 439 |
+
[default6]:[rank6]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 440 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 441 |
+
[default6]:[rank6]: output = model(**micro_batch)
|
| 442 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 443 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 444 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 445 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 446 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 447 |
+
[default6]:[rank6]: sharded_logits = self.model(
|
| 448 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 449 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 450 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 451 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 452 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 453 |
+
[default6]:[rank6]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 454 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 455 |
+
[default6]:[rank6]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 456 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 457 |
+
[default7]:[rank7]: Traceback (most recent call last):
|
| 458 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 459 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 460 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 461 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 462 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 463 |
+
[default6]:[rank6]: output = self.pp_block(**new_kwargs)
|
| 464 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 465 |
+
[default7]:[rank7]: trainer.train(dataloader)
|
| 466 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 467 |
+
[default7]:[rank7]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 468 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 469 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 470 |
+
[default7]:[rank7]: outputs = self.pipeline_engine.train_batch_iter(
|
| 471 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 472 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 473 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 474 |
+
[default6]:[rank6]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 475 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 476 |
+
[default7]:[rank7]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 477 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 478 |
+
[default7]:[rank7]: output = model(**micro_batch)
|
| 479 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 480 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 481 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 482 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 483 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 484 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 485 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 486 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 487 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 488 |
+
[default7]:[rank7]: sharded_logits = self.model(
|
| 489 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 490 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 491 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 492 |
+
[default6]:[rank6]: merged_states = self.gate_up_proj(hidden_states)
|
| 493 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 494 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 495 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 496 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 497 |
+
[default7]:[rank7]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 498 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 499 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 500 |
+
[default7]:[rank7]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 501 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 502 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 503 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 504 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 505 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 506 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 507 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 508 |
+
[default6]:[rank6]: return column_linear(
|
| 509 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 510 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 511 |
+
[default6]:[rank6]: return F.linear(input, weight, bias)
|
| 512 |
+
[default7]:[rank7]: output = self.pp_block(**new_kwargs)
|
| 513 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 514 |
+
[default6]:[rank6]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 512.00 MiB. GPU has a total capacity of 79.33 GiB of which 505.94 MiB is free. Including non-PyTorch memory, this process has 78.82 GiB memory in use. Of the allocated memory 67.10 GiB is allocated by PyTorch, and 272.09 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
|
| 515 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 516 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 517 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 518 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 519 |
+
[default7]:[rank7]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 520 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 521 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 522 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 523 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 524 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
|
| 525 |
+
[default7]:[rank7]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 526 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 527 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 528 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 529 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 530 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 159, in forward
|
| 531 |
+
[default7]:[rank7]: return row_linear(
|
| 532 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 474, in row_linear
|
| 533 |
+
[default7]:[rank7]: out = F.linear(input, weight, bias)
|
| 534 |
+
[default7]:[rank7]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 512.00 MiB. GPU has a total capacity of 79.33 GiB of which 457.94 MiB is free. Including non-PyTorch memory, this process has 78.87 GiB memory in use. Of the allocated memory 68.10 GiB is allocated by PyTorch, and 16.09 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
|
| 535 |
+
[default4]:[rank4]: Traceback (most recent call last):
|
| 536 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 537 |
+
[default4]:[rank4]: trainer.train(dataloader)
|
| 538 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 539 |
+
[default4]:[rank4]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 540 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 541 |
+
[default4]:[rank4]: outputs = self.pipeline_engine.train_batch_iter(
|
| 542 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 543 |
+
[default4]:[rank4]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 544 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 545 |
+
[default4]:[rank4]: output = model(**micro_batch)
|
| 546 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 547 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 548 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 549 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 550 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 551 |
+
[default4]:[rank4]: sharded_logits = self.model(
|
| 552 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 553 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 554 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 555 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 556 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 557 |
+
[default4]:[rank4]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 558 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 559 |
+
[default4]:[rank4]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 560 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 561 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 562 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 563 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 564 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 565 |
+
[default4]:[rank4]: output = self.pp_block(**new_kwargs)
|
| 566 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 567 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 568 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 569 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 570 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 571 |
+
[default4]:[rank4]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 572 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 573 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 574 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 575 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 576 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
|
| 577 |
+
[default4]:[rank4]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 578 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 579 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 580 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 581 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 582 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 128, in forward
|
| 583 |
+
[default4]:[rank4]: return self.act(gate_states) * up_states
|
| 584 |
+
[default4]:[rank4]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 256.00 MiB. GPU has a total capacity of 79.33 GiB of which 41.94 MiB is free. Including non-PyTorch memory, this process has 79.28 GiB memory in use. Of the allocated memory 67.85 GiB is allocated by PyTorch, and 16.09 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
|
| 585 |
+
E0703 23:46:44.821000 139891705947968 torch/distributed/elastic/multiprocessing/api.py:826] failed (exitcode: 1) local_rank: 0 (pid: 34122) of binary: /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/python3.10
|
| 586 |
+
Traceback (most recent call last):
|
| 587 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/torchrun", line 8, in <module>
|
| 588 |
+
sys.exit(main())
|
| 589 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 347, in wrapper
|
| 590 |
+
return f(*args, **kwargs)
|
| 591 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 879, in main
|
| 592 |
+
run(args)
|
| 593 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 870, in run
|
| 594 |
+
elastic_launch(
|
| 595 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 132, in __call__
|
| 596 |
+
return launch_agent(self._config, self._entrypoint, list(args))
|
| 597 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 263, in launch_agent
|
| 598 |
+
raise ChildFailedError(
|
| 599 |
+
torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
|
| 600 |
+
============================================================
|
| 601 |
+
/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py FAILED
|
| 602 |
+
------------------------------------------------------------
|
| 603 |
+
Failures:
|
| 604 |
+
[1]:
|
| 605 |
+
time : 2024-07-03_23:46:44
|
| 606 |
+
host : ip-26-0-164-187.ec2.internal
|
| 607 |
+
rank : 1 (local_rank: 1)
|
| 608 |
+
exitcode : 1 (pid: 34123)
|
| 609 |
+
error_file: <N/A>
|
| 610 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 611 |
+
[2]:
|
| 612 |
+
time : 2024-07-03_23:46:44
|
| 613 |
+
host : ip-26-0-164-187.ec2.internal
|
| 614 |
+
rank : 2 (local_rank: 2)
|
| 615 |
+
exitcode : 1 (pid: 34124)
|
| 616 |
+
error_file: <N/A>
|
| 617 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 618 |
+
[3]:
|
| 619 |
+
time : 2024-07-03_23:46:44
|
| 620 |
+
host : ip-26-0-164-187.ec2.internal
|
| 621 |
+
rank : 3 (local_rank: 3)
|
| 622 |
+
exitcode : 1 (pid: 34125)
|
| 623 |
+
error_file: <N/A>
|
| 624 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 625 |
+
[4]:
|
| 626 |
+
time : 2024-07-03_23:46:44
|
| 627 |
+
host : ip-26-0-164-187.ec2.internal
|
| 628 |
+
rank : 4 (local_rank: 4)
|
| 629 |
+
exitcode : 1 (pid: 34126)
|
| 630 |
+
error_file: <N/A>
|
| 631 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 632 |
+
[5]:
|
| 633 |
+
time : 2024-07-03_23:46:44
|
| 634 |
+
host : ip-26-0-164-187.ec2.internal
|
| 635 |
+
rank : 5 (local_rank: 5)
|
| 636 |
+
exitcode : 1 (pid: 34127)
|
| 637 |
+
error_file: <N/A>
|
| 638 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 639 |
+
[6]:
|
| 640 |
+
time : 2024-07-03_23:46:44
|
| 641 |
+
host : ip-26-0-164-187.ec2.internal
|
| 642 |
+
rank : 6 (local_rank: 6)
|
| 643 |
+
exitcode : 1 (pid: 34128)
|
| 644 |
+
error_file: <N/A>
|
| 645 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 646 |
+
[7]:
|
| 647 |
+
time : 2024-07-03_23:46:44
|
| 648 |
+
host : ip-26-0-164-187.ec2.internal
|
| 649 |
+
rank : 7 (local_rank: 7)
|
| 650 |
+
exitcode : 1 (pid: 34129)
|
| 651 |
+
error_file: <N/A>
|
| 652 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 653 |
+
------------------------------------------------------------
|
| 654 |
+
Root Cause (first observed failure):
|
| 655 |
+
[0]:
|
| 656 |
+
time : 2024-07-03_23:46:44
|
| 657 |
+
host : ip-26-0-164-187.ec2.internal
|
| 658 |
+
rank : 0 (local_rank: 0)
|
| 659 |
+
exitcode : 1 (pid: 34122)
|
| 660 |
+
error_file: <N/A>
|
| 661 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 662 |
+
============================================================
|
| 663 |
+
srun: error: ip-26-0-164-187: task 0: Exited with exit code 1
|
| 664 |
+
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/8_GPUS/dp-2_tp-4_pp-1_mbz-32/status.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
oom
|