Upload llama-1B/16_GPUS/dp-4_tp-4_pp-1_mbz-32
Browse files
llama-1B/16_GPUS/dp-4_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=00:59:00
|
| 5 |
+
#SBATCH --partition=hopper-prod
|
| 6 |
+
#SBATCH --nodes=2
|
| 7 |
+
#SBATCH --gres=gpu:8
|
| 8 |
+
#SBATCH --qos=high
|
| 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/16_GPUS/dp-4_tp-4_pp-1_mbz-32/log.out
|
| 13 |
+
#SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_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/16_GPUS/dp-4_tp-4_pp-1_mbz-32/config.yaml"
|
| 57 |
+
|
| 58 |
+
LAUNCHER="torchrun \
|
| 59 |
+
--nproc_per_node 8 \
|
| 60 |
+
--nnodes 2 \
|
| 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/16_GPUS/dp-4_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/16_GPUS/dp-4_tp-4_pp-1_mbz-32/status.txt
|
| 84 |
+
else
|
| 85 |
+
if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-4_pp-1_mbz-32/log.out; then
|
| 86 |
+
printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_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/16_GPUS/dp-4_tp-4_pp-1_mbz-32/log.out; then
|
| 88 |
+
printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-4_pp-1_mbz-32/status.txt
|
| 89 |
+
elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-4_pp-1_mbz-32/log.out; then
|
| 90 |
+
printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-4_pp-1_mbz-32/status.txt
|
| 91 |
+
else
|
| 92 |
+
printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_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/16_GPUS/dp-4_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/16_GPUS/dp-4_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/16_GPUS/dp-4_tp-4_pp-1_mbz-32 llama-1B/16_GPUS/dp-4_tp-4_pp-1_mbz-32 --commit-message "Upload llama-1B/16_GPUS/dp-4_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/16_GPUS/dp-4_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: 4
|
| 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/16_GPUS/dp-4_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: 32
|
| 73 |
+
seed: 42
|
| 74 |
+
lighteval: null
|
| 75 |
+
tokens:
|
| 76 |
+
train_steps: 20
|
| 77 |
+
val_check_interval: -1
|
| 78 |
+
batch_accumulation_per_replica: 8
|
| 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/16_GPUS/dp-4_tp-4_pp-1_mbz-32/log.out
ADDED
|
@@ -0,0 +1,779 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
========================
|
| 2 |
+
START TIME: Tue Jul 2 19:50:07 UTC 2024
|
| 3 |
+
python3 version = Python 3.10.14
|
| 4 |
+
========================
|
| 5 |
+
The token has not been saved to the git credentials helper. Pass `add_to_git_credential=True` in this function directly or `--add-to-git-credential` if using via `huggingface-cli` if you want to set the git credential as well.
|
| 6 |
+
Token is valid (permission: write).
|
| 7 |
+
Your token has been saved to /admin/home/ferdinand_mom/.cache/huggingface/token
|
| 8 |
+
Login successful
|
| 9 |
+
Already on 'bench_cluster'
|
| 10 |
+
M examples/config_tiny_llama.py
|
| 11 |
+
M examples/config_tiny_llama.yaml
|
| 12 |
+
M examples/train_tiny_llama.sh
|
| 13 |
+
M src/nanotron/models/llama.py
|
| 14 |
+
M src/nanotron/trainer.py
|
| 15 |
+
Your branch is up to date with 'origin/bench_cluster'.
|
| 16 |
+
Job status: RUNNING
|
| 17 |
+
W0702 19:50:13.980000 140190952634176 torch/distributed/run.py:757]
|
| 18 |
+
W0702 19:50:13.980000 140190952634176 torch/distributed/run.py:757] *****************************************
|
| 19 |
+
W0702 19:50:13.980000 140190952634176 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 |
+
W0702 19:50:13.980000 140190952634176 torch/distributed/run.py:757] *****************************************
|
| 21 |
+
W0702 19:50:13.997000 140213654959936 torch/distributed/run.py:757]
|
| 22 |
+
W0702 19:50:13.997000 140213654959936 torch/distributed/run.py:757] *****************************************
|
| 23 |
+
W0702 19:50:13.997000 140213654959936 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.
|
| 24 |
+
W0702 19:50:13.997000 140213654959936 torch/distributed/run.py:757] *****************************************
|
| 25 |
+
[default0]:07/02/2024 19:50:37 [WARNING|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Vocab Size Padding] Padded vocab (size: 50257) with 3 dummy tokens (new size: 50260)
|
| 26 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Config:
|
| 27 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Config(general=GeneralArgs(project='bench_cluster',
|
| 28 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: run='%date_%jobid',
|
| 29 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: seed=42,
|
| 30 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: step=None,
|
| 31 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: consumed_train_samples=None,
|
| 32 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: benchmark_csv_path=None,
|
| 33 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: ignore_sanity_checks=True),
|
| 34 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: parallelism=ParallelismArgs(dp=4,
|
| 35 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pp=1,
|
| 36 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tp=4,
|
| 37 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f096239c910>,
|
| 38 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
|
| 39 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tp_linear_async_communication=False,
|
| 40 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: expert_parallel_size=1),
|
| 41 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
|
| 42 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: eos_token_id=2,
|
| 43 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hidden_act='silu',
|
| 44 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hidden_size=2048,
|
| 45 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: initializer_range=0.02,
|
| 46 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: intermediate_size=4096,
|
| 47 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: is_llama_config=True,
|
| 48 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: max_position_embeddings=4096,
|
| 49 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_attention_heads=32,
|
| 50 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_hidden_layers=24,
|
| 51 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_key_value_heads=32,
|
| 52 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pad_token_id=None,
|
| 53 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pretraining_tp=1,
|
| 54 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rms_norm_eps=1e-05,
|
| 55 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rope_scaling=None,
|
| 56 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rope_theta=10000.0,
|
| 57 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tie_word_embeddings=True,
|
| 58 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: use_cache=True,
|
| 59 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: vocab_size=50260),
|
| 60 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: init_method=RandomInit(std=0.025),
|
| 61 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: dtype=torch.bfloat16,
|
| 62 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: make_vocab_size_divisible_by=1,
|
| 63 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: ddp_bucket_cap_mb=25),
|
| 64 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
|
| 65 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tokenizer_revision=None,
|
| 66 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tokenizer_max_length=None),
|
| 67 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
|
| 68 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: checkpoint_interval=100000,
|
| 69 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: save_initial_state=False,
|
| 70 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: resume_checkpoint_path=None,
|
| 71 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: checkpoints_path_is_shared_file_system=False),
|
| 72 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: logging=LoggingArgs(log_level='info',
|
| 73 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: log_level_replica='info',
|
| 74 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: iteration_step_info_interval=1),
|
| 75 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tokens=TokensArgs(sequence_length=4096,
|
| 76 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: train_steps=20,
|
| 77 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: micro_batch_size=32,
|
| 78 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: batch_accumulation_per_replica=8,
|
| 79 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: val_check_interval=-1,
|
| 80 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: limit_val_batches=0,
|
| 81 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: limit_test_batches=0),
|
| 82 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
|
| 83 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: adam_beta1=0.9,
|
| 84 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: adam_beta2=0.95,
|
| 85 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: torch_adam_is_fused=True,
|
| 86 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: name='adamW'),
|
| 87 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: zero_stage=1,
|
| 88 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: weight_decay=0.01,
|
| 89 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: clip_grad=1.0,
|
| 90 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: accumulate_grad_in_fp32=True,
|
| 91 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
|
| 92 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lr_warmup_steps=1,
|
| 93 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lr_warmup_style='linear',
|
| 94 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lr_decay_style='linear',
|
| 95 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lr_decay_steps=19,
|
| 96 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lr_decay_starting_step=None,
|
| 97 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: min_decay_lr=1e-05)),
|
| 98 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: data_stages=[DatasetStageArgs(name='Training Stage',
|
| 99 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: start_training_step=1,
|
| 100 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
|
| 101 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hf_dataset_splits='train',
|
| 102 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hf_dataset_config_name=None,
|
| 103 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: dataset_processing_num_proc_per_process=64,
|
| 104 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: dataset_overwrite_cache=False,
|
| 105 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: text_column_name='text'),
|
| 106 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: seed=42,
|
| 107 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_loading_workers=32))],
|
| 108 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-4_tp-4_pp-1_mbz-32')),
|
| 109 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: lighteval=None)
|
| 110 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Model Config:
|
| 111 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: LlamaConfig(bos_token_id=1,
|
| 112 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: eos_token_id=2,
|
| 113 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hidden_act='silu',
|
| 114 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: hidden_size=2048,
|
| 115 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: initializer_range=0.02,
|
| 116 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: intermediate_size=4096,
|
| 117 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: is_llama_config=True,
|
| 118 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: max_position_embeddings=4096,
|
| 119 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_attention_heads=32,
|
| 120 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_hidden_layers=24,
|
| 121 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: num_key_value_heads=32,
|
| 122 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pad_token_id=None,
|
| 123 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: pretraining_tp=1,
|
| 124 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rms_norm_eps=1e-05,
|
| 125 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rope_scaling=None,
|
| 126 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: rope_theta=10000.0,
|
| 127 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: tie_word_embeddings=True,
|
| 128 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: use_cache=True,
|
| 129 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: vocab_size=50260)
|
| 130 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Building model..
|
| 131 |
+
[default0]:07/02/2024 19:50:37 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Setting PP block ranks...
|
| 132 |
+
[default0]:07/02/2024 19:50:51 [INFO|DP=2|PP=0|TP=0|ip-26-0-171-102]: No checkpoint path provided.
|
| 133 |
+
[default1]:07/02/2024 19:50:51 [INFO|DP=2|PP=0|TP=1|ip-26-0-171-102]: No checkpoint path provided.
|
| 134 |
+
[default2]:07/02/2024 19:50:51 [INFO|DP=2|PP=0|TP=2|ip-26-0-171-102]: No checkpoint path provided.
|
| 135 |
+
[default3]:07/02/2024 19:50:51 [INFO|DP=2|PP=0|TP=3|ip-26-0-171-102]: No checkpoint path provided.
|
| 136 |
+
[default4]:07/02/2024 19:50:51 [INFO|DP=3|PP=0|TP=0|ip-26-0-171-102]: No checkpoint path provided.
|
| 137 |
+
[default6]:07/02/2024 19:50:51 [INFO|DP=3|PP=0|TP=2|ip-26-0-171-102]: No checkpoint path provided.
|
| 138 |
+
[default7]:07/02/2024 19:50:51 [INFO|DP=3|PP=0|TP=3|ip-26-0-171-102]: No checkpoint path provided.
|
| 139 |
+
[default3]:07/02/2024 19:50:51 [INFO|DP=0|PP=0|TP=3|ip-26-0-160-225]: Local number of parameters: 277M (529.27MiB)
|
| 140 |
+
[default1]:07/02/2024 19:50:51 [INFO|DP=0|PP=0|TP=1|ip-26-0-160-225]: Local number of parameters: 277M (529.27MiB)
|
| 141 |
+
[default1]:07/02/2024 19:50:51 [INFO|DP=0|PP=0|TP=1|ip-26-0-160-225]: [After model building] Memory usage: 554.21MiB. Peak allocated: 606.24MiB Peak reserved: 608.00MiB
|
| 142 |
+
[default1]:07/02/2024 19:50:51 [INFO|DP=0|PP=0|TP=1|ip-26-0-160-225]: No checkpoint path provided.
|
| 143 |
+
[default3]:07/02/2024 19:50:51 [INFO|DP=0|PP=0|TP=3|ip-26-0-160-225]: [After model building] Memory usage: 554.21MiB. Peak allocated: 606.24MiB Peak reserved: 608.00MiB
|
| 144 |
+
[default3]:07/02/2024 19:50:51 [INFO|DP=0|PP=0|TP=3|ip-26-0-160-225]: No checkpoint path provided.
|
| 145 |
+
[default2]:07/02/2024 19:50:51 [INFO|DP=0|PP=0|TP=2|ip-26-0-160-225]: Local number of parameters: 277M (529.27MiB)
|
| 146 |
+
[default2]:07/02/2024 19:50:51 [INFO|DP=0|PP=0|TP=2|ip-26-0-160-225]: [After model building] Memory usage: 554.21MiB. Peak allocated: 606.24MiB Peak reserved: 608.00MiB
|
| 147 |
+
[default2]:07/02/2024 19:50:51 [INFO|DP=0|PP=0|TP=2|ip-26-0-160-225]: No checkpoint path provided.
|
| 148 |
+
[default0]:07/02/2024 19:50:51 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Total number of parameters: 1.11G (2117.09MiB)
|
| 149 |
+
[default0]:07/02/2024 19:50:51 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Local number of parameters: 277M (529.27MiB)
|
| 150 |
+
[default0]:07/02/2024 19:50:51 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [After model building] Memory usage: 554.21MiB. Peak allocated: 606.24MiB Peak reserved: 608.00MiB
|
| 151 |
+
[default0]:07/02/2024 19:50:51 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: No checkpoint path provided.
|
| 152 |
+
[default0]:07/02/2024 19:50:51 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Parametrizing model parameters using StandardParametrizator
|
| 153 |
+
[default5]:07/02/2024 19:50:51 [INFO|DP=3|PP=0|TP=1|ip-26-0-171-102]: No checkpoint path provided.
|
| 154 |
+
[default4]:07/02/2024 19:50:51 [INFO|DP=1|PP=0|TP=0|ip-26-0-160-225]: No checkpoint path provided.
|
| 155 |
+
[default5]:07/02/2024 19:50:51 [INFO|DP=1|PP=0|TP=1|ip-26-0-160-225]: No checkpoint path provided.
|
| 156 |
+
[default7]:07/02/2024 19:50:51 [INFO|DP=1|PP=0|TP=3|ip-26-0-160-225]: No checkpoint path provided.
|
| 157 |
+
[default6]:07/02/2024 19:50:51 [INFO|DP=1|PP=0|TP=2|ip-26-0-160-225]: No checkpoint path provided.
|
| 158 |
+
[default0]:07/02/2024 19:50:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Optimizer Building] Using LearningRateForSP as learning rate
|
| 159 |
+
[default0]:07/02/2024 19:50:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [ZeRO sharding] Size of optimizer params per rank:
|
| 160 |
+
[default0]:07/02/2024 19:50:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [ZeRO sharding] DP Rank 0 has 69.4M out of 277M (25.00%) params' optimizer states
|
| 161 |
+
[default0]:07/02/2024 19:50:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [ZeRO sharding] DP Rank 1 has 69.4M out of 277M (25.00%) params' optimizer states
|
| 162 |
+
[default0]:07/02/2024 19:50:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [ZeRO sharding] DP Rank 2 has 69.4M out of 277M (25.00%) params' optimizer states
|
| 163 |
+
[default0]:07/02/2024 19:50:53 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [ZeRO sharding] DP Rank 3 has 69.4M out of 277M (25.00%) params' optimizer states
|
| 164 |
+
[default0]:07/02/2024 19:50:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
|
| 165 |
+
[default0]:07/02/2024 19:50:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Using `datasets` library
|
| 166 |
+
[default0]:07/02/2024 19:50:56 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
|
| 167 |
+
[default0]:07/02/2024 19:50:56 [WARNING|DP=0|PP=0|TP=0|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
|
| 168 |
+
[default0]:Repo card metadata block was not found. Setting CardData to empty.
|
| 169 |
+
[default0]:07/02/2024 19:50:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Training Plan] There are 1 training stages
|
| 170 |
+
[default0]:07/02/2024 19:50:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Stage Training Stage] start from step 1
|
| 171 |
+
[default0]:07/02/2024 19:50:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]:
|
| 172 |
+
[default0]:07/02/2024 19:50:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: [Start training] datetime: 2024-07-02 19:50:57.481225 | mbs: 32 | grad_accum: 8 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
|
| 173 |
+
[default0]:07/02/2024 19:50:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
|
| 174 |
+
[default0]:07/02/2024 19:50:57 [INFO|DP=0|PP=0|TP=0|ip-26-0-160-225]: Memory usage: 1877.40MiB. Peak allocated 1877.40MiB. Peak reserved: 1934.00MiB
|
| 175 |
+
[default0]:07/02/2024 19:50:57 [WARNING|DP=2|PP=0|TP=0|ip-26-0-171-102]: Repo card metadata block was not found. Setting CardData to empty.
|
| 176 |
+
[default1]:07/02/2024 19:50:57 [WARNING|DP=2|PP=0|TP=1|ip-26-0-171-102]: Repo card metadata block was not found. Setting CardData to empty.
|
| 177 |
+
[default2]:07/02/2024 19:50:57 [WARNING|DP=2|PP=0|TP=2|ip-26-0-171-102]: Repo card metadata block was not found. Setting CardData to empty.
|
| 178 |
+
[default4]:07/02/2024 19:50:57 [WARNING|DP=3|PP=0|TP=0|ip-26-0-171-102]: Repo card metadata block was not found. Setting CardData to empty.
|
| 179 |
+
[default5]:Repo card metadata block was not found. Setting CardData to empty.
|
| 180 |
+
[default6]:07/02/2024 19:50:57 [WARNING|DP=3|PP=0|TP=2|ip-26-0-171-102]: Repo card metadata block was not found. Setting CardData to empty.
|
| 181 |
+
[default3]:07/02/2024 19:50:57 [WARNING|DP=2|PP=0|TP=3|ip-26-0-171-102]: Repo card metadata block was not found. Setting CardData to empty.
|
| 182 |
+
[default3]:Repo card metadata block was not found. Setting CardData to empty.
|
| 183 |
+
[default6]:Repo card metadata block was not found. Setting CardData to empty.
|
| 184 |
+
[default4]:Repo card metadata block was not found. Setting CardData to empty.
|
| 185 |
+
[default7]:07/02/2024 19:50:57 [WARNING|DP=3|PP=0|TP=3|ip-26-0-171-102]: Repo card metadata block was not found. Setting CardData to empty.
|
| 186 |
+
[default2]:Repo card metadata block was not found. Setting CardData to empty.
|
| 187 |
+
[default2]:07/02/2024 19:50:57 [WARNING|DP=0|PP=0|TP=2|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
|
| 188 |
+
[default7]:07/02/2024 19:50:57 [WARNING|DP=1|PP=0|TP=3|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
|
| 189 |
+
[default1]:07/02/2024 19:50:57 [WARNING|DP=0|PP=0|TP=1|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
|
| 190 |
+
[default3]:07/02/2024 19:50:57 [WARNING|DP=0|PP=0|TP=3|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
|
| 191 |
+
[default5]:07/02/2024 19:50:57 [WARNING|DP=1|PP=0|TP=1|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
|
| 192 |
+
[default5]:07/02/2024 19:50:57 [WARNING|DP=3|PP=0|TP=1|ip-26-0-171-102]: Repo card metadata block was not found. Setting CardData to empty.
|
| 193 |
+
[default5]:Repo card metadata block was not found. Setting CardData to empty.
|
| 194 |
+
[default2]:Repo card metadata block was not found. Setting CardData to empty.
|
| 195 |
+
[default3]:Repo card metadata block was not found. Setting CardData to empty.
|
| 196 |
+
[default1]:Repo card metadata block was not found. Setting CardData to empty.
|
| 197 |
+
[default0]:Repo card metadata block was not found. Setting CardData to empty.
|
| 198 |
+
[default7]:Repo card metadata block was not found. Setting CardData to empty.
|
| 199 |
+
[default1]:Repo card metadata block was not found. Setting CardData to empty.
|
| 200 |
+
[default7]:Repo card metadata block was not found. Setting CardData to empty.
|
| 201 |
+
[default6]:07/02/2024 19:50:57 [WARNING|DP=1|PP=0|TP=2|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
|
| 202 |
+
[default4]:07/02/2024 19:50:57 [WARNING|DP=1|PP=0|TP=0|ip-26-0-160-225]: Repo card metadata block was not found. Setting CardData to empty.
|
| 203 |
+
[default4]:Repo card metadata block was not found. Setting CardData to empty.
|
| 204 |
+
[default6]:Repo card metadata block was not found. Setting CardData to empty.
|
| 205 |
+
[default4]:[rank4]: Traceback (most recent call last):
|
| 206 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 207 |
+
[default4]:[rank4]: trainer.train(dataloader)
|
| 208 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 209 |
+
[default4]:[rank4]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 210 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 211 |
+
[default4]:[rank4]: outputs = self.pipeline_engine.train_batch_iter(
|
| 212 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 213 |
+
[default4]:[rank4]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 214 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 215 |
+
[default4]:[rank4]: output = model(**micro_batch)
|
| 216 |
+
[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
|
| 217 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 218 |
+
[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
|
| 219 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 220 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 221 |
+
[default4]:[rank4]: sharded_logits = self.model(
|
| 222 |
+
[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
|
| 223 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 224 |
+
[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
|
| 225 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 226 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 227 |
+
[default4]:[rank4]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 228 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 229 |
+
[default4]:[rank4]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 230 |
+
[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
|
| 231 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 232 |
+
[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
|
| 233 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 234 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 235 |
+
[default4]:[rank4]: output = self.pp_block(**new_kwargs)
|
| 236 |
+
[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
|
| 237 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 238 |
+
[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
|
| 239 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 240 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 241 |
+
[default4]:[rank4]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 242 |
+
[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
|
| 243 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 244 |
+
[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
|
| 245 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 246 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
|
| 247 |
+
[default4]:[rank4]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 248 |
+
[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
|
| 249 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 250 |
+
[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
|
| 251 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 252 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 159, in forward
|
| 253 |
+
[default4]:[rank4]: return row_linear(
|
| 254 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 474, in row_linear
|
| 255 |
+
[default4]:[rank4]: out = F.linear(input, weight, bias)
|
| 256 |
+
[default4]:[rank4]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 512.00 MiB. GPU has a total capacity of 79.33 GiB of which 289.94 MiB is free. Including non-PyTorch memory, this process has 79.03 GiB memory in use. Of the allocated memory 70.97 GiB is allocated by PyTorch, and 12.70 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)
|
| 257 |
+
[default6]:[rank6]: Traceback (most recent call last):
|
| 258 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 259 |
+
[default6]:[rank6]: trainer.train(dataloader)
|
| 260 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 261 |
+
[default6]:[rank6]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 262 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 263 |
+
[default6]:[rank6]: outputs = self.pipeline_engine.train_batch_iter(
|
| 264 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 265 |
+
[default6]:[rank6]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 266 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 267 |
+
[default6]:[rank6]: output = model(**micro_batch)
|
| 268 |
+
[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
|
| 269 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 270 |
+
[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
|
| 271 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 272 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 273 |
+
[default6]:[rank6]: sharded_logits = self.model(
|
| 274 |
+
[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
|
| 275 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 276 |
+
[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
|
| 277 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 278 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 279 |
+
[default6]:[rank6]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 280 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 281 |
+
[default6]:[rank6]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 282 |
+
[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
|
| 283 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 284 |
+
[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
|
| 285 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 286 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 287 |
+
[default6]:[rank6]: output = self.pp_block(**new_kwargs)
|
| 288 |
+
[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
|
| 289 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 290 |
+
[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
|
| 291 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 292 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 293 |
+
[default6]:[rank6]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 294 |
+
[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
|
| 295 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 296 |
+
[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
|
| 297 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 298 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
|
| 299 |
+
[default6]:[rank6]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 300 |
+
[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
|
| 301 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 302 |
+
[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
|
| 303 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 304 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 159, in forward
|
| 305 |
+
[default6]:[rank6]: return row_linear(
|
| 306 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 474, in row_linear
|
| 307 |
+
[default6]:[rank6]: out = F.linear(input, weight, bias)
|
| 308 |
+
[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 65.94 MiB is free. Including non-PyTorch memory, this process has 79.25 GiB memory in use. Of the allocated memory 70.97 GiB is allocated by PyTorch, and 12.70 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)
|
| 309 |
+
[default7]:[rank7]: Traceback (most recent call last):
|
| 310 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 311 |
+
[default7]:[rank7]: trainer.train(dataloader)
|
| 312 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 313 |
+
[default7]:[rank7]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 314 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 315 |
+
[default7]:[rank7]: outputs = self.pipeline_engine.train_batch_iter(
|
| 316 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 317 |
+
[default7]:[rank7]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 318 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 319 |
+
[default7]:[rank7]: output = model(**micro_batch)
|
| 320 |
+
[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
|
| 321 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 322 |
+
[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
|
| 323 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 324 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 325 |
+
[default7]:[rank7]: sharded_logits = self.model(
|
| 326 |
+
[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
|
| 327 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 328 |
+
[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
|
| 329 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 330 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 331 |
+
[default7]:[rank7]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 332 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 333 |
+
[default7]:[rank7]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 334 |
+
[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
|
| 335 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 336 |
+
[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
|
| 337 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 338 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 339 |
+
[default7]:[rank7]: output = self.pp_block(**new_kwargs)
|
| 340 |
+
[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
|
| 341 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 342 |
+
[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
|
| 343 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 344 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 345 |
+
[default7]:[rank7]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 346 |
+
[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
|
| 347 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 348 |
+
[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
|
| 349 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 350 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
|
| 351 |
+
[default7]:[rank7]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 352 |
+
[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
|
| 353 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 354 |
+
[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
|
| 355 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 356 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 159, in forward
|
| 357 |
+
[default7]:[rank7]: return row_linear(
|
| 358 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 479, in row_linear
|
| 359 |
+
[default7]:[rank7]: out = differentiable_reduce_scatter_sum(out, group=group)
|
| 360 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/distributed_differentiable_primitives.py", line 145, in differentiable_reduce_scatter_sum
|
| 361 |
+
[default7]:[rank7]: return DifferentiableReduceScatterSum.apply(tensor, group)
|
| 362 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/function.py", line 598, in apply
|
| 363 |
+
[default7]:[rank7]: return super().apply(*args, **kwargs) # type: ignore[misc]
|
| 364 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/distributed_differentiable_primitives.py", line 111, in forward
|
| 365 |
+
[default7]:[rank7]: sharded_tensor = torch.empty(
|
| 366 |
+
[default7]:[rank7]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 128.00 MiB. GPU has a total capacity of 79.33 GiB of which 17.94 MiB is free. Including non-PyTorch memory, this process has 79.30 GiB memory in use. Of the allocated memory 71.47 GiB is allocated by PyTorch, and 12.70 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)
|
| 367 |
+
[default5]:[rank5]: Traceback (most recent call last):
|
| 368 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 369 |
+
[default5]:[rank5]: trainer.train(dataloader)
|
| 370 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 371 |
+
[default5]:[rank5]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 372 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 373 |
+
[default5]:[rank5]: outputs = self.pipeline_engine.train_batch_iter(
|
| 374 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 375 |
+
[default5]:[rank5]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 376 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 377 |
+
[default5]:[rank5]: output = model(**micro_batch)
|
| 378 |
+
[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
|
| 379 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 380 |
+
[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
|
| 381 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 382 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 383 |
+
[default5]:[rank5]: sharded_logits = self.model(
|
| 384 |
+
[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
|
| 385 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 386 |
+
[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
|
| 387 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 388 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 389 |
+
[default5]:[rank5]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 390 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 391 |
+
[default5]:[rank5]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 392 |
+
[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
|
| 393 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 394 |
+
[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
|
| 395 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 396 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 397 |
+
[default5]:[rank5]: output = self.pp_block(**new_kwargs)
|
| 398 |
+
[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
|
| 399 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 400 |
+
[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
|
| 401 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 402 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 403 |
+
[default5]:[rank5]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_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/models/llama.py", line 172, in forward
|
| 409 |
+
[default5]:[rank5]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 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/parallel/tensor_parallel/nn.py", line 159, in forward
|
| 415 |
+
[default5]:[rank5]: return row_linear(
|
| 416 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 474, in row_linear
|
| 417 |
+
[default5]:[rank5]: out = F.linear(input, weight, bias)
|
| 418 |
+
[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 41.94 MiB is free. Including non-PyTorch memory, this process has 79.28 GiB memory in use. Of the allocated memory 70.97 GiB is allocated by PyTorch, and 12.70 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)
|
| 419 |
+
[default0]:[rank0]: Traceback (most recent call last):
|
| 420 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 421 |
+
[default0]:[rank0]: trainer.train(dataloader)
|
| 422 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 423 |
+
[default0]:[rank0]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 424 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 425 |
+
[default0]:[rank0]: outputs = self.pipeline_engine.train_batch_iter(
|
| 426 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 427 |
+
[default0]:[rank0]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 428 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 429 |
+
[default0]:[rank0]: output = model(**micro_batch)
|
| 430 |
+
[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
|
| 431 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 432 |
+
[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
|
| 433 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 434 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 435 |
+
[default0]:[rank0]: sharded_logits = self.model(
|
| 436 |
+
[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
|
| 437 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 438 |
+
[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
|
| 439 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 440 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 441 |
+
[default0]:[rank0]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 442 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 443 |
+
[default0]:[rank0]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 444 |
+
[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
|
| 445 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 446 |
+
[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
|
| 447 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 448 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 449 |
+
[default0]:[rank0]: output = self.pp_block(**new_kwargs)
|
| 450 |
+
[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
|
| 451 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 452 |
+
[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
|
| 453 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 454 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 631, in forward
|
| 455 |
+
[default0]:[rank0]: output = self.attn(hidden_states=hidden_states, sequence_mask=sequence_mask)
|
| 456 |
+
[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
|
| 457 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 458 |
+
[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
|
| 459 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 460 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 565, in forward
|
| 461 |
+
[default0]:[rank0]: key_value_states = key_value_states.permute(1, 2, 0, 3, 4).contiguous()
|
| 462 |
+
[default0]:[rank0]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 256.00 MiB. GPU
|
| 463 |
+
W0702 19:51:06.212000 140213654959936 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 1653252 closing signal SIGTERM
|
| 464 |
+
W0702 19:51:06.213000 140213654959936 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 1653253 closing signal SIGTERM
|
| 465 |
+
W0702 19:51:06.220000 140213654959936 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 1653254 closing signal SIGTERM
|
| 466 |
+
W0702 19:51:06.223000 140213654959936 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 1653255 closing signal SIGTERM
|
| 467 |
+
[default3]:[rank11]: Traceback (most recent call last):
|
| 468 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 469 |
+
[default3]:[rank11]: trainer.train(dataloader)
|
| 470 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 471 |
+
[default3]:[rank11]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 472 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 473 |
+
[default3]:[rank11]: outputs = self.pipeline_engine.train_batch_iter(
|
| 474 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 475 |
+
[default3]:[rank11]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 476 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 477 |
+
[default3]:[rank11]: output = model(**micro_batch)
|
| 478 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 479 |
+
[default3]:[rank11]: return self._call_impl(*args, **kwargs)
|
| 480 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 481 |
+
[default3]:[rank11]: return forward_call(*args, **kwargs)
|
| 482 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 483 |
+
[default3]:[rank11]: sharded_logits = self.model(
|
| 484 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 485 |
+
[default3]:[rank11]: return self._call_impl(*args, **kwargs)
|
| 486 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 487 |
+
[default3]:[rank11]: return forward_call(*args, **kwargs)
|
| 488 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 489 |
+
[default3]:[rank11]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 490 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 491 |
+
[default3]:[rank11]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 492 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 493 |
+
[default3]:[rank11]: return self._call_impl(*args, **kwargs)
|
| 494 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 495 |
+
[default3]:[rank11]: return forward_call(*args, **kwargs)
|
| 496 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 497 |
+
[default3]:[rank11]: output = self.pp_block(**new_kwargs)
|
| 498 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 499 |
+
[default3]:[rank11]: return self._call_impl(*args, **kwargs)
|
| 500 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 501 |
+
[default3]:[rank11]: return forward_call(*args, **kwargs)
|
| 502 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 503 |
+
[default3]:[rank11]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 504 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 505 |
+
[default3]:[rank11]: return self._call_impl(*args, **kwargs)
|
| 506 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 507 |
+
[default3]:[rank11]: return forward_call(*args, **kwargs)
|
| 508 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
|
| 509 |
+
[default3]:[rank11]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 510 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 511 |
+
[default3]:[rank11]: return self._call_impl(*args, **kwargs)
|
| 512 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 513 |
+
[default3]:[rank11]: return forward_call(*args, **kwargs)
|
| 514 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 159, in forward
|
| 515 |
+
[default3]:[rank11]: return row_linear(
|
| 516 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 474, in row_linear
|
| 517 |
+
[default3]:[rank11]: out = F.linear(input, weight, bias)
|
| 518 |
+
[default3]:[rank11]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 512.00 MiB. GPU has a total capacity of 79.33 GiB of which 481.94 MiB is free. Including non-PyTorch memory, this process has 78.85 GiB memory in use. Of the allocated memory 70.97 GiB is allocated by PyTorch, and 12.70 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)
|
| 519 |
+
[default0]:[rank8]: Traceback (most recent call last):
|
| 520 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 521 |
+
[default0]:[rank8]: trainer.train(dataloader)
|
| 522 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 523 |
+
[default0]:[rank8]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 524 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 525 |
+
[default0]:[rank8]: outputs = self.pipeline_engine.train_batch_iter(
|
| 526 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 527 |
+
[default0]:[rank8]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 528 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 529 |
+
[default0]:[rank8]: output = model(**micro_batch)
|
| 530 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 531 |
+
[default0]:[rank8]: return self._call_impl(*args, **kwargs)
|
| 532 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 533 |
+
[default0]:[rank8]: return forward_call(*args, **kwargs)
|
| 534 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 535 |
+
[default0]:[rank8]: sharded_logits = self.model(
|
| 536 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 537 |
+
[default0]:[rank8]: return self._call_impl(*args, **kwargs)
|
| 538 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 539 |
+
[default0]:[rank8]: return forward_call(*args, **kwargs)
|
| 540 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 541 |
+
[default0]:[rank8]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 542 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 543 |
+
[default0]:[rank8]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 544 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 545 |
+
[default0]:[rank8]: return self._call_impl(*args, **kwargs)
|
| 546 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 547 |
+
[default0]:[rank8]: return forward_call(*args, **kwargs)
|
| 548 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 549 |
+
[default0]:[rank8]: output = self.pp_block(**new_kwargs)
|
| 550 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 551 |
+
[default0]:[rank8]: return self._call_impl(*args, **kwargs)
|
| 552 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 553 |
+
[default0]:[rank8]: return forward_call(*args, **kwargs)
|
| 554 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 555 |
+
[default0]:[rank8]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 556 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 557 |
+
[default0]:[rank8]: return self._call_impl(*args, **kwargs)
|
| 558 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 559 |
+
[default0]:[rank8]: return forward_call(*args, **kwargs)
|
| 560 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
|
| 561 |
+
[default0]:[rank8]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 562 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 563 |
+
[default0]:[rank8]: return self._call_impl(*args, **kwargs)
|
| 564 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 565 |
+
[default0]:[rank8]: return forward_call(*args, **kwargs)
|
| 566 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 159, in forward
|
| 567 |
+
[default0]:[rank8]: return row_linear(
|
| 568 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 479, in row_linear
|
| 569 |
+
[default0]:[rank8]: out = differentiable_reduce_scatter_sum(out, group=group)
|
| 570 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/distributed_differentiable_primitives.py", line 145, in differentiable_reduce_scatter_sum
|
| 571 |
+
[default1]:[rank9]: Traceback (most recent call last):
|
| 572 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 573 |
+
[default1]:[rank9]: trainer.train(dataloader)
|
| 574 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 575 |
+
[default1]:[rank9]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 576 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 577 |
+
[default1]:[rank9]: outputs = self.pipeline_engine.train_batch_iter(
|
| 578 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 579 |
+
[default1]:[rank9]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 580 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 581 |
+
[default1]:[rank9]: output = model(**micro_batch)
|
| 582 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 583 |
+
[default1]:[rank9]: return self._call_impl(*args, **kwargs)
|
| 584 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 585 |
+
[default1]:[rank9]: return forward_call(*args, **kwargs)
|
| 586 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 587 |
+
[default1]:[rank9]: sharded_logits = self.model(
|
| 588 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 589 |
+
[default1]:[rank9]: return self._call_impl(*args, **kwargs)
|
| 590 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 591 |
+
[default1]:[rank9]: return forward_call(*args, **kwargs)
|
| 592 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 593 |
+
[default1]:[rank9]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 594 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 595 |
+
[default1]:[rank9]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 596 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 597 |
+
[default1]:[rank9]: return self._call_impl(*args, **kwargs)
|
| 598 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 599 |
+
[default1]:[rank9]: return forward_call(*args, **kwargs)
|
| 600 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 601 |
+
[default1]:[rank9]: output = self.pp_block(**new_kwargs)
|
| 602 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 603 |
+
[default1]:[rank9]: return self._call_impl(*args, **kwargs)
|
| 604 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 605 |
+
[default1]:[rank9]: return forward_call(*args, **kwargs)
|
| 606 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 607 |
+
[default1]:[rank9]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 608 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 609 |
+
[default1]:[rank9]: return self._call_impl(*args, **kwargs)
|
| 610 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 611 |
+
[default1]:[rank9]: return forward_call(*args, **kwargs)
|
| 612 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
|
| 613 |
+
[default1]:[rank9]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 614 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 615 |
+
[default1]:[rank9]: return self._call_impl(*args, **kwargs)
|
| 616 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 617 |
+
[default1]:[rank9]: return forward_call(*args, **kwargs)
|
| 618 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 159, in forward
|
| 619 |
+
[default1]:[rank9]: return row_linear(
|
| 620 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 474, in row_linear
|
| 621 |
+
[default1]:[rank9]: out = F.linear(input, weight, bias)
|
| 622 |
+
[default1]:[rank9]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 512.00 MiB. GPU has a total capacity of 79.33 GiB of which 309.94 MiB is free. Including non-PyTorch memory, this process has 79.01 GiB memory in use. Of the allocated memory 70.97 GiB is allocated by PyTorch, and 12.70 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)
|
| 623 |
+
[default0]:[rank8]: return DifferentiableReduceScatterSum.apply(tensor, group)
|
| 624 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/function.py", line 598, in apply
|
| 625 |
+
[default0]:[rank8]: return super().apply(*args, **kwargs) # type: ignore[misc]
|
| 626 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/distributed_differentiable_primitives.py", line 111, in forward
|
| 627 |
+
[default0]:[rank8]: sharded_tensor = torch.empty(
|
| 628 |
+
[default0]:[rank8]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 128.00 MiB. GPU
|
| 629 |
+
[default2]:[rank10]: Traceback (most recent call last):
|
| 630 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 631 |
+
[default2]:[rank10]: trainer.train(dataloader)
|
| 632 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 633 |
+
[default2]:[rank10]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 634 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 635 |
+
[default2]:[rank10]: outputs = self.pipeline_engine.train_batch_iter(
|
| 636 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 637 |
+
[default2]:[rank10]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 638 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 639 |
+
[default2]:[rank10]: output = model(**micro_batch)
|
| 640 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 641 |
+
[default2]:[rank10]: return self._call_impl(*args, **kwargs)
|
| 642 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 643 |
+
[default2]:[rank10]: return forward_call(*args, **kwargs)
|
| 644 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 645 |
+
[default2]:[rank10]: sharded_logits = self.model(
|
| 646 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 647 |
+
[default2]:[rank10]: return self._call_impl(*args, **kwargs)
|
| 648 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 649 |
+
[default2]:[rank10]: return forward_call(*args, **kwargs)
|
| 650 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 651 |
+
[default2]:[rank10]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 652 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 653 |
+
[default2]:[rank10]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 654 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 655 |
+
[default2]:[rank10]: return self._call_impl(*args, **kwargs)
|
| 656 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 657 |
+
[default2]:[rank10]: return forward_call(*args, **kwargs)
|
| 658 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 659 |
+
[default2]:[rank10]: output = self.pp_block(**new_kwargs)
|
| 660 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 661 |
+
[default2]:[rank10]: return self._call_impl(*args, **kwargs)
|
| 662 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 663 |
+
[default2]:[rank10]: return forward_call(*args, **kwargs)
|
| 664 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 665 |
+
[default2]:[rank10]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 666 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 667 |
+
[default2]:[rank10]: return self._call_impl(*args, **kwargs)
|
| 668 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 669 |
+
[default2]:[rank10]: return forward_call(*args, **kwargs)
|
| 670 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
|
| 671 |
+
[default2]:[rank10]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 672 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 673 |
+
[default2]:[rank10]: return self._call_impl(*args, **kwargs)
|
| 674 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 675 |
+
[default2]:[rank10]: return forward_call(*args, **kwargs)
|
| 676 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 159, in forward
|
| 677 |
+
[default2]:[rank10]: return row_linear(
|
| 678 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 474, in row_linear
|
| 679 |
+
[default2]:[rank10]: out = F.linear(input, weight, bias)
|
| 680 |
+
[default2]:[rank10]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 512.00 MiB. GPU has a total capacity of 79.33 GiB of which 223.94 MiB is free. Including non-PyTorch memory, this process has 79.10 GiB memory in use. Of the allocated memory 70.97 GiB is allocated by PyTorch, and 12.70 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)
|
| 681 |
+
E0702 19:51:08.750000 140213654959936 torch/distributed/elastic/multiprocessing/api.py:826] failed (exitcode: 1) local_rank: 4 (pid: 1653256) of binary: /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/python3.10
|
| 682 |
+
Traceback (most recent call last):
|
| 683 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/torchrun", line 8, in <module>
|
| 684 |
+
sys.exit(main())
|
| 685 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 347, in wrapper
|
| 686 |
+
return f(*args, **kwargs)
|
| 687 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 879, in main
|
| 688 |
+
run(args)
|
| 689 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 870, in run
|
| 690 |
+
elastic_launch(
|
| 691 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 132, in __call__
|
| 692 |
+
return launch_agent(self._config, self._entrypoint, list(args))
|
| 693 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 263, in launch_agent
|
| 694 |
+
raise ChildFailedError(
|
| 695 |
+
torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
|
| 696 |
+
============================================================
|
| 697 |
+
/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py FAILED
|
| 698 |
+
------------------------------------------------------------
|
| 699 |
+
Failures:
|
| 700 |
+
[1]:
|
| 701 |
+
time : 2024-07-02_19:51:06
|
| 702 |
+
host : ip-26-0-160-225.ec2.internal
|
| 703 |
+
rank : 5 (local_rank: 5)
|
| 704 |
+
exitcode : 1 (pid: 1653257)
|
| 705 |
+
error_file: <N/A>
|
| 706 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 707 |
+
[2]:
|
| 708 |
+
time : 2024-07-02_19:51:06
|
| 709 |
+
host : ip-26-0-160-225.ec2.internal
|
| 710 |
+
rank : 6 (local_rank: 6)
|
| 711 |
+
exitcode : 1 (pid: 1653258)
|
| 712 |
+
error_file: <N/A>
|
| 713 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 714 |
+
[3]:
|
| 715 |
+
time : 2024-07-02_19:51:06
|
| 716 |
+
host : ip-26-0-160-225.ec2.internal
|
| 717 |
+
rank : 7 (local_rank: 7)
|
| 718 |
+
exitcode : 1 (pid: 1653259)
|
| 719 |
+
error_file: <N/A>
|
| 720 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 721 |
+
------------------------------------------------------------
|
| 722 |
+
Root Cause (first observed failure):
|
| 723 |
+
[0]:
|
| 724 |
+
time : 2024-07-02_19:51:06
|
| 725 |
+
host : ip-26-0-160-225.ec2.internal
|
| 726 |
+
rank : 4 (local_rank: 4)
|
| 727 |
+
exitcode : 1 (pid: 1653256)
|
| 728 |
+
error_file: <N/A>
|
| 729 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 730 |
+
============================================================
|
| 731 |
+
srun: error: ip-26-0-160-225: task 0: Exited with exit code 1
|
| 732 |
+
W0702 19:51:10.255000 140185285814016 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1252] The node 'ip-26-0-171-102.ec2.internal_3646767_0' has failed to send a keep-alive heartbeat to the rendezvous 'none' due to an error of type RendezvousConnectionError.
|
| 733 |
+
W0702 19:51:11.232000 140190952634176 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 3646841 closing signal SIGTERM
|
| 734 |
+
W0702 19:51:11.232000 140190952634176 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 3646842 closing signal SIGTERM
|
| 735 |
+
W0702 19:51:11.233000 140190952634176 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 3646843 closing signal SIGTERM
|
| 736 |
+
W0702 19:51:11.233000 140190952634176 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 3646844 closing signal SIGTERM
|
| 737 |
+
W0702 19:51:11.233000 140190952634176 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 3646845 closing signal SIGTERM
|
| 738 |
+
W0702 19:51:11.238000 140190952634176 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 3646846 closing signal SIGTERM
|
| 739 |
+
W0702 19:51:11.238000 140190952634176 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 3646847 closing signal SIGTERM
|
| 740 |
+
W0702 19:51:11.244000 140190952634176 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 3646848 closing signal SIGTERM
|
| 741 |
+
W0702 19:51:13.786000 140190952634176 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1203] The node 'ip-26-0-171-102.ec2.internal_3646767_0' has failed to shutdown the rendezvous 'none' due to an error of type RendezvousConnectionError.
|
| 742 |
+
W0702 19:51:13.797000 140190952634176 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1203] The node 'ip-26-0-171-102.ec2.internal_3646767_0' has failed to shutdown the rendezvous 'none' due to an error of type RendezvousConnectionError.
|
| 743 |
+
Traceback (most recent call last):
|
| 744 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/rendezvous/c10d_rendezvous_backend.py", line 113, in _call_store
|
| 745 |
+
return getattr(self._store, store_op)(*args, **kwargs)
|
| 746 |
+
torch.distributed.DistNetworkError: Broken pipe
|
| 747 |
+
|
| 748 |
+
The above exception was the direct cause of the following exception:
|
| 749 |
+
|
| 750 |
+
Traceback (most recent call last):
|
| 751 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/torchrun", line 8, in <module>
|
| 752 |
+
sys.exit(main())
|
| 753 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 347, in wrapper
|
| 754 |
+
return f(*args, **kwargs)
|
| 755 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 879, in main
|
| 756 |
+
run(args)
|
| 757 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 870, in run
|
| 758 |
+
elastic_launch(
|
| 759 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 132, in __call__
|
| 760 |
+
return launch_agent(self._config, self._entrypoint, list(args))
|
| 761 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 254, in launch_agent
|
| 762 |
+
result = agent.run()
|
| 763 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 123, in wrapper
|
| 764 |
+
result = f(*args, **kwargs)
|
| 765 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 733, in run
|
| 766 |
+
result = self._invoke_run(role)
|
| 767 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 908, in _invoke_run
|
| 768 |
+
num_nodes_waiting = rdzv_handler.num_nodes_waiting()
|
| 769 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/rendezvous/dynamic_rendezvous.py", line 1174, in num_nodes_waiting
|
| 770 |
+
self._state_holder.sync()
|
| 771 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/rendezvous/dynamic_rendezvous.py", line 419, in sync
|
| 772 |
+
get_response = self._backend.get_state()
|
| 773 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/rendezvous/c10d_rendezvous_backend.py", line 73, in get_state
|
| 774 |
+
base64_state: bytes = self._call_store("get", self._key)
|
| 775 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/rendezvous/c10d_rendezvous_backend.py", line 115, in _call_store
|
| 776 |
+
raise RendezvousConnectionError(
|
| 777 |
+
torch.distributed.elastic.rendezvous.api.RendezvousConnectionError: The connection to the C10d store has failed. See inner exception for details.
|
| 778 |
+
srun: error: ip-26-0-171-102: task 1: Exited with exit code 1
|
| 779 |
+
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-4_pp-1_mbz-32/status.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
oom
|