Upload llama-1B/16_GPUS/dp-1_tp-4_pp-4_mbz-32
Browse files
llama-1B/16_GPUS/dp-1_tp-4_pp-4_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-1_tp-4_pp-4_mbz-32/log.out
|
| 13 |
+
#SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-4_pp-4_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-1_tp-4_pp-4_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-1_tp-4_pp-4_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-1_tp-4_pp-4_mbz-32/status.txt
|
| 84 |
+
else
|
| 85 |
+
if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-4_pp-4_mbz-32/log.out; then
|
| 86 |
+
printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-4_pp-4_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-1_tp-4_pp-4_mbz-32/log.out; then
|
| 88 |
+
printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-4_pp-4_mbz-32/status.txt
|
| 89 |
+
elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-4_pp-4_mbz-32/log.out; then
|
| 90 |
+
printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-4_pp-4_mbz-32/status.txt
|
| 91 |
+
else
|
| 92 |
+
printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-4_pp-4_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-1_tp-4_pp-4_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-1_tp-4_pp-4_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-1_tp-4_pp-4_mbz-32 llama-1B/16_GPUS/dp-1_tp-4_pp-4_mbz-32 --commit-message "Upload llama-1B/16_GPUS/dp-1_tp-4_pp-4_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-1_tp-4_pp-4_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: 1
|
| 49 |
+
expert_parallel_size: 1
|
| 50 |
+
pp: 4
|
| 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-1_tp-4_pp-4_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: 32
|
| 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-1_tp-4_pp-4_mbz-32/log.out
ADDED
|
@@ -0,0 +1,800 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
========================
|
| 2 |
+
START TIME: Tue Jul 2 14:14:57 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 14:15:05.914000 140251158693696 torch/distributed/run.py:757]
|
| 18 |
+
W0702 14:15:05.914000 140251158693696 torch/distributed/run.py:757] *****************************************
|
| 19 |
+
W0702 14:15:05.914000 140251158693696 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 14:15:05.914000 140251158693696 torch/distributed/run.py:757] *****************************************
|
| 21 |
+
W0702 14:15:06.663000 140497670219584 torch/distributed/run.py:757]
|
| 22 |
+
W0702 14:15:06.663000 140497670219584 torch/distributed/run.py:757] *****************************************
|
| 23 |
+
W0702 14:15:06.663000 140497670219584 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 14:15:06.663000 140497670219584 torch/distributed/run.py:757] *****************************************
|
| 25 |
+
[default0]:07/02/2024 14:15:31 [WARNING|DP=0|PP=0|TP=0|ip-26-0-170-31]: [Vocab Size Padding] Padded vocab (size: 50257) with 3 dummy tokens (new size: 50260)
|
| 26 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Config:
|
| 27 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Config(general=GeneralArgs(project='bench_cluster',
|
| 28 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: run='%date_%jobid',
|
| 29 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: seed=42,
|
| 30 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: step=None,
|
| 31 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: consumed_train_samples=None,
|
| 32 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: benchmark_csv_path=None,
|
| 33 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: ignore_sanity_checks=True),
|
| 34 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: parallelism=ParallelismArgs(dp=1,
|
| 35 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: pp=4,
|
| 36 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tp=4,
|
| 37 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f459fdb06a0>,
|
| 38 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
|
| 39 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tp_linear_async_communication=False,
|
| 40 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: expert_parallel_size=1),
|
| 41 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
|
| 42 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: eos_token_id=2,
|
| 43 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: hidden_act='silu',
|
| 44 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: hidden_size=2048,
|
| 45 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: initializer_range=0.02,
|
| 46 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: intermediate_size=4096,
|
| 47 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: is_llama_config=True,
|
| 48 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: max_position_embeddings=4096,
|
| 49 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: num_attention_heads=32,
|
| 50 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: num_hidden_layers=24,
|
| 51 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: num_key_value_heads=32,
|
| 52 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: pad_token_id=None,
|
| 53 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: pretraining_tp=1,
|
| 54 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: rms_norm_eps=1e-05,
|
| 55 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: rope_scaling=None,
|
| 56 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: rope_theta=10000.0,
|
| 57 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tie_word_embeddings=True,
|
| 58 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: use_cache=True,
|
| 59 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: vocab_size=50260),
|
| 60 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: init_method=RandomInit(std=0.025),
|
| 61 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: dtype=torch.bfloat16,
|
| 62 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: make_vocab_size_divisible_by=1,
|
| 63 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: ddp_bucket_cap_mb=25),
|
| 64 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
|
| 65 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tokenizer_revision=None,
|
| 66 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tokenizer_max_length=None),
|
| 67 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
|
| 68 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: checkpoint_interval=100000,
|
| 69 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: save_initial_state=False,
|
| 70 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: resume_checkpoint_path=None,
|
| 71 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: checkpoints_path_is_shared_file_system=False),
|
| 72 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: logging=LoggingArgs(log_level='info',
|
| 73 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: log_level_replica='info',
|
| 74 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: iteration_step_info_interval=1),
|
| 75 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tokens=TokensArgs(sequence_length=4096,
|
| 76 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: train_steps=20,
|
| 77 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: micro_batch_size=32,
|
| 78 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: batch_accumulation_per_replica=32,
|
| 79 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: val_check_interval=-1,
|
| 80 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: limit_val_batches=0,
|
| 81 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: limit_test_batches=0),
|
| 82 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
|
| 83 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: adam_beta1=0.9,
|
| 84 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: adam_beta2=0.95,
|
| 85 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: torch_adam_is_fused=True,
|
| 86 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: name='adamW'),
|
| 87 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: zero_stage=1,
|
| 88 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: weight_decay=0.01,
|
| 89 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: clip_grad=1.0,
|
| 90 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: accumulate_grad_in_fp32=True,
|
| 91 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
|
| 92 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: lr_warmup_steps=1,
|
| 93 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: lr_warmup_style='linear',
|
| 94 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: lr_decay_style='linear',
|
| 95 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: lr_decay_steps=19,
|
| 96 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: lr_decay_starting_step=None,
|
| 97 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: min_decay_lr=1e-05)),
|
| 98 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: data_stages=[DatasetStageArgs(name='Training Stage',
|
| 99 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: start_training_step=1,
|
| 100 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
|
| 101 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: hf_dataset_splits='train',
|
| 102 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: hf_dataset_config_name=None,
|
| 103 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: dataset_processing_num_proc_per_process=64,
|
| 104 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: dataset_overwrite_cache=False,
|
| 105 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: text_column_name='text'),
|
| 106 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: seed=42,
|
| 107 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: num_loading_workers=32))],
|
| 108 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/16_GPUS/dp-1_tp-4_pp-4_mbz-32')),
|
| 109 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: lighteval=None)
|
| 110 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Model Config:
|
| 111 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: LlamaConfig(bos_token_id=1,
|
| 112 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: eos_token_id=2,
|
| 113 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: hidden_act='silu',
|
| 114 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: hidden_size=2048,
|
| 115 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: initializer_range=0.02,
|
| 116 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: intermediate_size=4096,
|
| 117 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: is_llama_config=True,
|
| 118 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: max_position_embeddings=4096,
|
| 119 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: num_attention_heads=32,
|
| 120 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: num_hidden_layers=24,
|
| 121 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: num_key_value_heads=32,
|
| 122 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: pad_token_id=None,
|
| 123 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: pretraining_tp=1,
|
| 124 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: rms_norm_eps=1e-05,
|
| 125 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: rope_scaling=None,
|
| 126 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: rope_theta=10000.0,
|
| 127 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: tie_word_embeddings=True,
|
| 128 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: use_cache=True,
|
| 129 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: vocab_size=50260)
|
| 130 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Building model..
|
| 131 |
+
[default0]:07/02/2024 14:15:31 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Setting PP block ranks...
|
| 132 |
+
[default6]:07/02/2024 14:15:47 [INFO|DP=0|PP=3|TP=2|ip-26-0-171-56]: Local number of parameters: 67.7M (129.12MiB)
|
| 133 |
+
[default3]:07/02/2024 14:15:47 [INFO|DP=0|PP=0|TP=3|ip-26-0-170-31]: Local number of parameters: 99.2M (189.14MiB)
|
| 134 |
+
[default3]:07/02/2024 14:15:47 [INFO|DP=0|PP=0|TP=3|ip-26-0-170-31]: [After model building] Memory usage: 197.07MiB. Peak allocated: 199.10MiB Peak reserved: 200.00MiB
|
| 135 |
+
[default3]:07/02/2024 14:15:47 [INFO|DP=0|PP=0|TP=3|ip-26-0-170-31]: No checkpoint path provided.
|
| 136 |
+
[default2]:07/02/2024 14:15:47 [INFO|DP=0|PP=0|TP=2|ip-26-0-170-31]: Local number of parameters: 99.2M (189.14MiB)
|
| 137 |
+
[default2]:07/02/2024 14:15:47 [INFO|DP=0|PP=0|TP=2|ip-26-0-170-31]: [After model building] Memory usage: 197.07MiB. Peak allocated: 199.10MiB Peak reserved: 200.00MiB
|
| 138 |
+
[default2]:07/02/2024 14:15:47 [INFO|DP=0|PP=0|TP=2|ip-26-0-170-31]: No checkpoint path provided.
|
| 139 |
+
[default7]:07/02/2024 14:15:47 [INFO|DP=0|PP=1|TP=3|ip-26-0-170-31]: Local number of parameters: 73.4M (140.05MiB)
|
| 140 |
+
[default7]:07/02/2024 14:15:47 [INFO|DP=0|PP=1|TP=3|ip-26-0-170-31]: [After model building] Memory usage: 147.07MiB. Peak allocated: 149.10MiB Peak reserved: 150.00MiB
|
| 141 |
+
[default7]:07/02/2024 14:15:47 [INFO|DP=0|PP=1|TP=3|ip-26-0-170-31]: No checkpoint path provided.
|
| 142 |
+
[default4]:07/02/2024 14:15:47 [INFO|DP=0|PP=1|TP=0|ip-26-0-170-31]: Local number of parameters: 73.4M (140.05MiB)
|
| 143 |
+
[default4]:07/02/2024 14:15:47 [INFO|DP=0|PP=1|TP=0|ip-26-0-170-31]: [After model building] Memory usage: 147.07MiB. Peak allocated: 149.10MiB Peak reserved: 150.00MiB
|
| 144 |
+
[default4]:07/02/2024 14:15:47 [INFO|DP=0|PP=1|TP=0|ip-26-0-170-31]: No checkpoint path provided.
|
| 145 |
+
[default1]:07/02/2024 14:15:47 [INFO|DP=0|PP=0|TP=1|ip-26-0-170-31]: Local number of parameters: 99.2M (189.14MiB)
|
| 146 |
+
[default1]:07/02/2024 14:15:47 [INFO|DP=0|PP=0|TP=1|ip-26-0-170-31]: [After model building] Memory usage: 197.07MiB. Peak allocated: 199.10MiB Peak reserved: 200.00MiB
|
| 147 |
+
[default1]:07/02/2024 14:15:47 [INFO|DP=0|PP=0|TP=1|ip-26-0-170-31]: No checkpoint path provided.
|
| 148 |
+
[default5]:07/02/2024 14:15:47 [INFO|DP=0|PP=1|TP=1|ip-26-0-170-31]: Local number of parameters: 73.4M (140.05MiB)
|
| 149 |
+
[default5]:07/02/2024 14:15:47 [INFO|DP=0|PP=1|TP=1|ip-26-0-170-31]: [After model building] Memory usage: 147.07MiB. Peak allocated: 149.10MiB Peak reserved: 150.00MiB
|
| 150 |
+
[default5]:07/02/2024 14:15:47 [INFO|DP=0|PP=1|TP=1|ip-26-0-170-31]: No checkpoint path provided.
|
| 151 |
+
[default6]:07/02/2024 14:15:47 [INFO|DP=0|PP=1|TP=2|ip-26-0-170-31]: Local number of parameters: 73.4M (140.05MiB)
|
| 152 |
+
[default6]:07/02/2024 14:15:47 [INFO|DP=0|PP=1|TP=2|ip-26-0-170-31]: [After model building] Memory usage: 147.07MiB. Peak allocated: 149.10MiB Peak reserved: 150.00MiB
|
| 153 |
+
[default6]:07/02/2024 14:15:47 [INFO|DP=0|PP=1|TP=2|ip-26-0-170-31]: No checkpoint path provided.
|
| 154 |
+
[default0]:07/02/2024 14:15:47 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Total number of parameters: 1.21G (2313.42MiB)
|
| 155 |
+
[default0]:07/02/2024 14:15:47 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Local number of parameters: 99.2M (189.14MiB)
|
| 156 |
+
[default0]:07/02/2024 14:15:47 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [After model building] Memory usage: 197.07MiB. Peak allocated: 199.10MiB Peak reserved: 200.00MiB
|
| 157 |
+
[default0]:07/02/2024 14:15:47 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: No checkpoint path provided.
|
| 158 |
+
[default0]:07/02/2024 14:15:47 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Parametrizing model parameters using StandardParametrizator
|
| 159 |
+
[default0]:07/02/2024 14:15:47 [INFO|DP=0|PP=2|TP=0|ip-26-0-171-56]: Local number of parameters: 62.9M (120.05MiB)
|
| 160 |
+
[default0]:07/02/2024 14:15:47 [INFO|DP=0|PP=2|TP=0|ip-26-0-171-56]: [After model building] Memory usage: 126.06MiB. Peak allocated: 128.09MiB Peak reserved: 130.00MiB
|
| 161 |
+
[default0]:07/02/2024 14:15:47 [INFO|DP=0|PP=2|TP=0|ip-26-0-171-56]: No checkpoint path provided.
|
| 162 |
+
[default1]:07/02/2024 14:15:47 [INFO|DP=0|PP=2|TP=1|ip-26-0-171-56]: Local number of parameters: 62.9M (120.05MiB)
|
| 163 |
+
[default1]:07/02/2024 14:15:47 [INFO|DP=0|PP=2|TP=1|ip-26-0-171-56]: [After model building] Memory usage: 126.06MiB. Peak allocated: 128.09MiB Peak reserved: 130.00MiB
|
| 164 |
+
[default1]:07/02/2024 14:15:47 [INFO|DP=0|PP=2|TP=1|ip-26-0-171-56]: No checkpoint path provided.
|
| 165 |
+
[default7]:07/02/2024 14:15:47 [INFO|DP=0|PP=3|TP=3|ip-26-0-171-56]: Local number of parameters: 67.7M (129.12MiB)
|
| 166 |
+
[default7]:07/02/2024 14:15:47 [INFO|DP=0|PP=3|TP=3|ip-26-0-171-56]: [After model building] Memory usage: 134.05MiB. Peak allocated: 136.08MiB Peak reserved: 138.00MiB
|
| 167 |
+
[default7]:07/02/2024 14:15:47 [INFO|DP=0|PP=3|TP=3|ip-26-0-171-56]: No checkpoint path provided.
|
| 168 |
+
[default2]:07/02/2024 14:15:47 [INFO|DP=0|PP=2|TP=2|ip-26-0-171-56]: Local number of parameters: 62.9M (120.05MiB)
|
| 169 |
+
[default2]:07/02/2024 14:15:47 [INFO|DP=0|PP=2|TP=2|ip-26-0-171-56]: [After model building] Memory usage: 126.06MiB. Peak allocated: 128.09MiB Peak reserved: 130.00MiB
|
| 170 |
+
[default2]:07/02/2024 14:15:47 [INFO|DP=0|PP=2|TP=2|ip-26-0-171-56]: No checkpoint path provided.
|
| 171 |
+
[default3]:07/02/2024 14:15:47 [INFO|DP=0|PP=2|TP=3|ip-26-0-171-56]: Local number of parameters: 62.9M (120.05MiB)
|
| 172 |
+
[default5]:07/02/2024 14:15:47 [INFO|DP=0|PP=3|TP=1|ip-26-0-171-56]: Local number of parameters: 67.7M (129.12MiB)
|
| 173 |
+
[default5]:07/02/2024 14:15:47 [INFO|DP=0|PP=3|TP=1|ip-26-0-171-56]: [After model building] Memory usage: 134.05MiB. Peak allocated: 136.08MiB Peak reserved: 138.00MiB
|
| 174 |
+
[default3]:07/02/2024 14:15:47 [INFO|DP=0|PP=2|TP=3|ip-26-0-171-56]: [After model building] Memory usage: 126.06MiB. Peak allocated: 128.09MiB Peak reserved: 130.00MiB
|
| 175 |
+
[default4]:07/02/2024 14:15:47 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-56]: Local number of parameters: 67.7M (129.12MiB)
|
| 176 |
+
[default4]:07/02/2024 14:15:47 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-56]: [After model building] Memory usage: 134.05MiB. Peak allocated: 136.08MiB Peak reserved: 138.00MiB
|
| 177 |
+
[default4]:07/02/2024 14:15:47 [INFO|DP=0|PP=3|TP=0|ip-26-0-171-56]: No checkpoint path provided.
|
| 178 |
+
[default3]:07/02/2024 14:15:47 [INFO|DP=0|PP=2|TP=3|ip-26-0-171-56]: No checkpoint path provided.
|
| 179 |
+
[default5]:07/02/2024 14:15:47 [INFO|DP=0|PP=3|TP=1|ip-26-0-171-56]: No checkpoint path provided.
|
| 180 |
+
[default6]:07/02/2024 14:15:47 [INFO|DP=0|PP=3|TP=2|ip-26-0-171-56]: [After model building] Memory usage: 134.05MiB. Peak allocated: 136.08MiB Peak reserved: 138.00MiB
|
| 181 |
+
[default6]:07/02/2024 14:15:47 [INFO|DP=0|PP=3|TP=2|ip-26-0-171-56]: No checkpoint path provided.
|
| 182 |
+
[default0]:07/02/2024 14:15:49 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [Optimizer Building] Using LearningRateForSP as learning rate
|
| 183 |
+
[default0]:07/02/2024 14:15:49 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [ZeRO sharding] Size of optimizer params per rank:
|
| 184 |
+
[default0]:07/02/2024 14:15:49 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [ZeRO sharding] DP Rank 0 has 99.2M out of 99.2M (100.00%) params' optimizer states
|
| 185 |
+
[default0]:07/02/2024 14:15:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
|
| 186 |
+
[default0]:07/02/2024 14:15:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Using `datasets` library
|
| 187 |
+
[default0]:07/02/2024 14:15:50 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
|
| 188 |
+
[default0]:07/02/2024 14:15:50 [WARNING|DP=0|PP=0|TP=0|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
|
| 189 |
+
[default0]:Repo card metadata block was not found. Setting CardData to empty.
|
| 190 |
+
[default0]:07/02/2024 14:15:52 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [Training Plan] There are 1 training stages
|
| 191 |
+
[default0]:07/02/2024 14:15:52 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [Stage Training Stage] start from step 1
|
| 192 |
+
[default0]:07/02/2024 14:15:52 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]:
|
| 193 |
+
[default0]:07/02/2024 14:15:52 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: [Start training] datetime: 2024-07-02 14:15:52.769469 | mbs: 32 | grad_accum: 32 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
|
| 194 |
+
[default0]:07/02/2024 14:15:52 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
|
| 195 |
+
[default0]:07/02/2024 14:15:52 [INFO|DP=0|PP=0|TP=0|ip-26-0-170-31]: Memory usage: 953.61MiB. Peak allocated 953.61MiB. Peak reserved: 960.00MiB
|
| 196 |
+
[default5]:07/02/2024 14:15:52 [WARNING|DP=0|PP=3|TP=1|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
|
| 197 |
+
[default7]:Repo card metadata block was not found. Setting CardData to empty.
|
| 198 |
+
[default3]:07/02/2024 14:15:52 [WARNING|DP=0|PP=2|TP=3|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
|
| 199 |
+
[default7]:07/02/2024 14:15:52 [WARNING|DP=0|PP=3|TP=3|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
|
| 200 |
+
[default2]:07/02/2024 14:15:52 [WARNING|DP=0|PP=2|TP=2|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
|
| 201 |
+
[default5]:Repo card metadata block was not found. Setting CardData to empty.
|
| 202 |
+
[default2]:Repo card metadata block was not found. Setting CardData to empty.
|
| 203 |
+
[default3]:Repo card metadata block was not found. Setting CardData to empty.
|
| 204 |
+
[default3]:07/02/2024 14:15:52 [WARNING|DP=0|PP=0|TP=3|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
|
| 205 |
+
[default2]:Repo card metadata block was not found. Setting CardData to empty.
|
| 206 |
+
[default2]:07/02/2024 14:15:52 [WARNING|DP=0|PP=0|TP=2|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
|
| 207 |
+
[default7]:07/02/2024 14:15:52 [WARNING|DP=0|PP=1|TP=3|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
|
| 208 |
+
[default1]:Repo card metadata block was not found. Setting CardData to empty.
|
| 209 |
+
[default7]:Repo card metadata block was not found. Setting CardData to empty.
|
| 210 |
+
[default5]:Repo card metadata block was not found. Setting CardData to empty.
|
| 211 |
+
[default4]:Repo card metadata block was not found. Setting CardData to empty.
|
| 212 |
+
[default4]:07/02/2024 14:15:52 [WARNING|DP=0|PP=1|TP=0|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
|
| 213 |
+
[default1]:07/02/2024 14:15:52 [WARNING|DP=0|PP=0|TP=1|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
|
| 214 |
+
[default3]:Repo card metadata block was not found. Setting CardData to empty.
|
| 215 |
+
[default5]:07/02/2024 14:15:52 [WARNING|DP=0|PP=1|TP=1|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
|
| 216 |
+
[default6]:Repo card metadata block was not found. Setting CardData to empty.
|
| 217 |
+
[default6]:07/02/2024 14:15:52 [WARNING|DP=0|PP=1|TP=2|ip-26-0-170-31]: Repo card metadata block was not found. Setting CardData to empty.
|
| 218 |
+
[default0]:Repo card metadata block was not found. Setting CardData to empty.
|
| 219 |
+
[default0]:07/02/2024 14:15:52 [WARNING|DP=0|PP=2|TP=0|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
|
| 220 |
+
[default1]:07/02/2024 14:15:53 [WARNING|DP=0|PP=2|TP=1|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
|
| 221 |
+
[default6]:Repo card metadata block was not found. Setting CardData to empty.
|
| 222 |
+
[default6]:07/02/2024 14:15:53 [WARNING|DP=0|PP=3|TP=2|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
|
| 223 |
+
[default1]:Repo card metadata block was not found. Setting CardData to empty.
|
| 224 |
+
[default4]:Repo card metadata block was not found. Setting CardData to empty.
|
| 225 |
+
[default4]:07/02/2024 14:15:53 [WARNING|DP=0|PP=3|TP=0|ip-26-0-171-56]: Repo card metadata block was not found. Setting CardData to empty.
|
| 226 |
+
[default2]:[rank2]: Traceback (most recent call last):
|
| 227 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 228 |
+
[default2]:[rank2]: trainer.train(dataloader)
|
| 229 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 230 |
+
[default2]:[rank2]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 231 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 232 |
+
[default2]:[rank2]: outputs = self.pipeline_engine.train_batch_iter(
|
| 233 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 234 |
+
[default2]:[rank2]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 235 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 236 |
+
[default2]:[rank2]: output = model(**micro_batch)
|
| 237 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 238 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 239 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 240 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 241 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 242 |
+
[default2]:[rank2]: sharded_logits = self.model(
|
| 243 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 244 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 245 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 246 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 247 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 248 |
+
[default2]:[rank2]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 249 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 250 |
+
[default2]:[rank2]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 251 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 252 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 253 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 254 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 255 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 256 |
+
[default2]:[rank2]: output = self.pp_block(**new_kwargs)
|
| 257 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 258 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 259 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 260 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 261 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 262 |
+
[default2]:[rank2]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 263 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 264 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 265 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 266 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 267 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
|
| 268 |
+
[default2]:[rank2]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 269 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 270 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 271 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 272 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 273 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 128, in forward
|
| 274 |
+
[default2]:[rank2]: return self.act(gate_states) * up_states
|
| 275 |
+
[default2]:[rank2]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 256.00 MiB. GPU has a total capacity of 79.33 GiB of which 205.94 MiB is free. Including non-PyTorch memory, this process has 79.11 GiB memory in use. Of the allocated memory 69.81 GiB is allocated by PyTorch, and 141.60 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)
|
| 276 |
+
[default0]:[rank0]: Traceback (most recent call last):
|
| 277 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 278 |
+
[default0]:[rank0]: trainer.train(dataloader)
|
| 279 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 280 |
+
[default0]:[rank0]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 281 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 282 |
+
[default0]:[rank0]: outputs = self.pipeline_engine.train_batch_iter(
|
| 283 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 284 |
+
[default0]:[rank0]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 285 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 286 |
+
[default0]:[rank0]: output = model(**micro_batch)
|
| 287 |
+
[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
|
| 288 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 289 |
+
[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
|
| 290 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 291 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 292 |
+
[default0]:[rank0]: sharded_logits = self.model(
|
| 293 |
+
[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
|
| 294 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 295 |
+
[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
|
| 296 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 297 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 298 |
+
[default0]:[rank0]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 299 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 300 |
+
[default0]:[rank0]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 301 |
+
[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
|
| 302 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 303 |
+
[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
|
| 304 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 305 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 306 |
+
[default0]:[rank0]: output = self.pp_block(**new_kwargs)
|
| 307 |
+
[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
|
| 308 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 309 |
+
[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
|
| 310 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 311 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 312 |
+
[default0]:[rank0]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 313 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 314 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 315 |
+
[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
|
| 316 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 317 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
|
| 318 |
+
[default0]:[rank0]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 319 |
+
[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
|
| 320 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 321 |
+
[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
|
| 322 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 323 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 159, in forward
|
| 324 |
+
[default0]:[rank0]: return row_linear(
|
| 325 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 474, in row_linear
|
| 326 |
+
[default0]:[rank0]: out = F.linear(input, weight, bias)
|
| 327 |
+
[default0]:[rank0]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 512.00 MiB. GPU
|
| 328 |
+
[default1]:[rank1]: Traceback (most recent call last):
|
| 329 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 330 |
+
[default1]:[rank1]: trainer.train(dataloader)
|
| 331 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 332 |
+
[default1]:[rank1]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 333 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 334 |
+
[default1]:[rank1]: outputs = self.pipeline_engine.train_batch_iter(
|
| 335 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 336 |
+
[default1]:[rank1]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 337 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 338 |
+
[default1]:[rank1]: output = model(**micro_batch)
|
| 339 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 340 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 341 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 342 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 343 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 344 |
+
[default1]:[rank1]: sharded_logits = self.model(
|
| 345 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 346 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 347 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 348 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 349 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 350 |
+
[default1]:[rank1]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 351 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 352 |
+
[default1]:[rank1]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 353 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 354 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 355 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 356 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 357 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 358 |
+
[default1]:[rank1]: output = self.pp_block(**new_kwargs)
|
| 359 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 360 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 361 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 362 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 363 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 364 |
+
[default1]:[rank1]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 365 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 366 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 367 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 368 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 369 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
|
| 370 |
+
[default1]:[rank1]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 371 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 372 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 373 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 374 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 375 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 159, in forward
|
| 376 |
+
[default1]:[rank1]: return row_linear(
|
| 377 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 474, in row_linear
|
| 378 |
+
[default1]:[rank1]: out = F.linear(input, weight, bias)
|
| 379 |
+
[default1]:[rank1]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 512.00 MiB. GPU has a total capacity of 79.33 GiB of which 35.94 MiB is free. Including non-PyTorch memory, this process has 79.28 GiB memory in use. Of the allocated memory 70.06 GiB is allocated by PyTorch, and 141.60 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)
|
| 380 |
+
[default1]:Exception in thread Thread-2 (_pin_memory_loop):
|
| 381 |
+
[default1]:Traceback (most recent call last):
|
| 382 |
+
[default1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/threading.py", line 1016, in _bootstrap_inner
|
| 383 |
+
[default1]: self.run()
|
| 384 |
+
[default1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/threading.py", line 953, in run
|
| 385 |
+
[default1]: self._target(*self._args, **self._kwargs)
|
| 386 |
+
[default1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/utils/data/_utils/pin_memory.py", line 54, in _pin_memory_loop
|
| 387 |
+
[default1]: do_one_step()
|
| 388 |
+
[default1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/utils/data/_utils/pin_memory.py", line 31, in do_one_step
|
| 389 |
+
[default1]: r = in_queue.get(timeout=MP_STATUS_CHECK_INTERVAL)
|
| 390 |
+
[default1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/queues.py", line 122, in get
|
| 391 |
+
[default1]: return _ForkingPickler.loads(res)
|
| 392 |
+
[default1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/multiprocessing/reductions.py", line 495, in rebuild_storage_fd
|
| 393 |
+
[default1]: fd = df.detach()
|
| 394 |
+
[default1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/resource_sharer.py", line 57, in detach
|
| 395 |
+
[default1]: with _resource_sharer.get_connection(self._id) as conn:
|
| 396 |
+
[default1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/resource_sharer.py", line 86, in get_connection
|
| 397 |
+
[default1]: c = Client(address, authkey=process.current_process().authkey)
|
| 398 |
+
[default1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/connection.py", line 508, in Client
|
| 399 |
+
[default1]: answer_challenge(c, authkey)
|
| 400 |
+
[default1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/connection.py", line 752, in answer_challenge
|
| 401 |
+
[default1]: message = connection.recv_bytes(256) # reject large message
|
| 402 |
+
[default1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/connection.py", line 216, in recv_bytes
|
| 403 |
+
[default1]: buf = self._recv_bytes(maxlength)
|
| 404 |
+
[default1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/connection.py", line 414, in _recv_bytes
|
| 405 |
+
[default1]: buf = self._recv(4)
|
| 406 |
+
[default1]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/connection.py", line 379, in _recv
|
| 407 |
+
[default1]: chunk = read(handle, remaining)
|
| 408 |
+
[default1]:ConnectionResetError: [Errno 104] Connection reset by peer
|
| 409 |
+
[default3]:[rank3]: Traceback (most recent call last):
|
| 410 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 411 |
+
[default3]:[rank3]: trainer.train(dataloader)
|
| 412 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 413 |
+
[default3]:[rank3]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 414 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 415 |
+
[default3]:[rank3]: outputs = self.pipeline_engine.train_batch_iter(
|
| 416 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 417 |
+
[default3]:[rank3]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 418 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 419 |
+
[default3]:[rank3]: output = model(**micro_batch)
|
| 420 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 421 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 422 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 423 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 424 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 425 |
+
[default3]:[rank3]: sharded_logits = self.model(
|
| 426 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 427 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 428 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 429 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 430 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 431 |
+
[default3]:[rank3]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 432 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 433 |
+
[default3]:[rank3]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 434 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 435 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 436 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 437 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 438 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 439 |
+
[default3]:[rank3]: output = self.pp_block(**new_kwargs)
|
| 440 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 441 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 442 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 443 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 444 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 445 |
+
[default3]:[rank3]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 446 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 447 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 448 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 449 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 450 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 172, in forward
|
| 451 |
+
[default3]:[rank3]: hidden_states = self.down_proj(self.split_silu_mul(merged_states))
|
| 452 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 453 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 454 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 455 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 456 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 159, in forward
|
| 457 |
+
[default3]:[rank3]: return row_linear(
|
| 458 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 474, in row_linear
|
| 459 |
+
[default3]:[rank3]: out = F.linear(input, weight, bias)
|
| 460 |
+
[default3]:[rank3]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 512.00 MiB. GPU has a total capacity of 79.33 GiB of which 335.94 MiB is free. Including non-PyTorch memory, this process has 78.99 GiB memory in use. Of the allocated memory 70.06 GiB is allocated by PyTorch, and 141.60 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)
|
| 461 |
+
[default0]:Exception in thread Thread-2 (_pin_memory_loop):
|
| 462 |
+
[default0]:Traceback (most recent call last):
|
| 463 |
+
[default0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/threading.py", line 1016, in _bootstrap_inner
|
| 464 |
+
[default0]: self.run()
|
| 465 |
+
[default0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/threading.py", line 953, in run
|
| 466 |
+
[default0]: self._target(*self._args, **self._kwargs)
|
| 467 |
+
[default0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/utils/data/_utils/pin_memory.py", line 54, in _pin_memory_loop
|
| 468 |
+
[default0]: do_one_step()
|
| 469 |
+
[default0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/utils/data/_utils/pin_memory.py", line 31, in do_one_step
|
| 470 |
+
[default0]: r = in_queue.get(timeout=MP_STATUS_CHECK_INTERVAL)
|
| 471 |
+
[default0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/queues.py", line 122, in get
|
| 472 |
+
[default0]: return _ForkingPickler.loads(res)
|
| 473 |
+
[default0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/multiprocessing/reductions.py", line 495, in rebuild_storage_fd
|
| 474 |
+
[default0]: fd = df.detach()
|
| 475 |
+
[default0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/resource_sharer.py", line 57, in detach
|
| 476 |
+
[default0]: with _resource_sharer.get_connection(self._id) as conn:
|
| 477 |
+
[default0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/resource_sharer.py", line 86, in get_connection
|
| 478 |
+
[default0]: c = Client(address, authkey=process.current_process().authkey)
|
| 479 |
+
[default0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/connection.py", line 508, in Client
|
| 480 |
+
[default0]: answer_challenge(c, authkey)
|
| 481 |
+
[default0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/connection.py", line 757, in answer_challenge
|
| 482 |
+
[default0]: response = connection.recv_bytes(256) # reject large message
|
| 483 |
+
[default0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/connection.py", line 216, in recv_bytes
|
| 484 |
+
[default0]: buf = self._recv_bytes(maxlength)
|
| 485 |
+
[default0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/connection.py", line 414, in _recv_bytes
|
| 486 |
+
[default0]: buf = self._recv(4)
|
| 487 |
+
[default0]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/multiprocessing/connection.py", line 379, in _recv
|
| 488 |
+
[default0]: chunk = read(handle, remaining)
|
| 489 |
+
[default0]:ConnectionResetError: [Errno 104] Connection reset by peer
|
| 490 |
+
W0702 14:16:13.047000 140497670219584 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2695551 closing signal SIGTERM
|
| 491 |
+
W0702 14:16:13.048000 140497670219584 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2695552 closing signal SIGTERM
|
| 492 |
+
W0702 14:16:13.049000 140497670219584 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2695553 closing signal SIGTERM
|
| 493 |
+
W0702 14:16:13.049000 140497670219584 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2695554 closing signal SIGTERM
|
| 494 |
+
[default4]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: c10d::allreduce_: an autograd kernel was not registered to the Autograd key(s) but we are trying to backprop through it. This may lead to silently incorrect behavior. This behavior is deprecated and will be removed in a future version of PyTorch. If your operator is differentiable, please ensure you have registered an autograd kernel to the correct Autograd key (e.g. DispatchKey::Autograd, DispatchKey::CompositeImplicitAutograd). If your operator is not differentiable, or to squash this warning and use the previous behavior, please register torch::CppFunction::makeFallthrough() to DispatchKey::Autograd. (Triggered internally at ../torch/csrc/autograd/autograd_not_implemented_fallback.cpp:63.)
|
| 495 |
+
[default4]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
|
| 496 |
+
[default5]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: c10d::allreduce_: an autograd kernel was not registered to the Autograd key(s) but we are trying to backprop through it. This may lead to silently incorrect behavior. This behavior is deprecated and will be removed in a future version of PyTorch. If your operator is differentiable, please ensure you have registered an autograd kernel to the correct Autograd key (e.g. DispatchKey::Autograd, DispatchKey::CompositeImplicitAutograd). If your operator is not differentiable, or to squash this warning and use the previous behavior, please register torch::CppFunction::makeFallthrough() to DispatchKey::Autograd. (Triggered internally at ../torch/csrc/autograd/autograd_not_implemented_fallback.cpp:63.)
|
| 497 |
+
[default5]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
|
| 498 |
+
[default6]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: c10d::allreduce_: an autograd kernel was not registered to the Autograd key(s) but we are trying to backprop through it. This may lead to silently incorrect behavior. This behavior is deprecated and will be removed in a future version of PyTorch. If your operator is differentiable, please ensure you have registered an autograd kernel to the correct Autograd key (e.g. DispatchKey::Autograd, DispatchKey::CompositeImplicitAutograd). If your operator is not differentiable, or to squash this warning and use the previous behavior, please register torch::CppFunction::makeFallthrough() to DispatchKey::Autograd. (Triggered internally at ../torch/csrc/autograd/autograd_not_implemented_fallback.cpp:63.)
|
| 499 |
+
[default6]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
|
| 500 |
+
[default7]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: c10d::allreduce_: an autograd kernel was not registered to the Autograd key(s) but we are trying to backprop through it. This may lead to silently incorrect behavior. This behavior is deprecated and will be removed in a future version of PyTorch. If your operator is differentiable, please ensure you have registered an autograd kernel to the correct Autograd key (e.g. DispatchKey::Autograd, DispatchKey::CompositeImplicitAutograd). If your operator is not differentiable, or to squash this warning and use the previous behavior, please register torch::CppFunction::makeFallthrough() to DispatchKey::Autograd. (Triggered internally at ../torch/csrc/autograd/autograd_not_implemented_fallback.cpp:63.)
|
| 501 |
+
[default7]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
|
| 502 |
+
E0702 14:16:15.183000 140497670219584 torch/distributed/elastic/multiprocessing/api.py:826] failed (exitcode: 1) local_rank: 0 (pid: 2695547) of binary: /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/python3.10
|
| 503 |
+
Traceback (most recent call last):
|
| 504 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/torchrun", line 8, in <module>
|
| 505 |
+
sys.exit(main())
|
| 506 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 347, in wrapper
|
| 507 |
+
return f(*args, **kwargs)
|
| 508 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 879, in main
|
| 509 |
+
run(args)
|
| 510 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 870, in run
|
| 511 |
+
elastic_launch(
|
| 512 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 132, in __call__
|
| 513 |
+
return launch_agent(self._config, self._entrypoint, list(args))
|
| 514 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 263, in launch_agent
|
| 515 |
+
raise ChildFailedError(
|
| 516 |
+
torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
|
| 517 |
+
============================================================
|
| 518 |
+
/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py FAILED
|
| 519 |
+
------------------------------------------------------------
|
| 520 |
+
Failures:
|
| 521 |
+
[1]:
|
| 522 |
+
time : 2024-07-02_14:16:13
|
| 523 |
+
host : ip-26-0-170-31.ec2.internal
|
| 524 |
+
rank : 1 (local_rank: 1)
|
| 525 |
+
exitcode : 1 (pid: 2695548)
|
| 526 |
+
error_file: <N/A>
|
| 527 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 528 |
+
[2]:
|
| 529 |
+
time : 2024-07-02_14:16:13
|
| 530 |
+
host : ip-26-0-170-31.ec2.internal
|
| 531 |
+
rank : 2 (local_rank: 2)
|
| 532 |
+
exitcode : 1 (pid: 2695549)
|
| 533 |
+
error_file: <N/A>
|
| 534 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 535 |
+
[3]:
|
| 536 |
+
time : 2024-07-02_14:16:13
|
| 537 |
+
host : ip-26-0-170-31.ec2.internal
|
| 538 |
+
rank : 3 (local_rank: 3)
|
| 539 |
+
exitcode : 1 (pid: 2695550)
|
| 540 |
+
error_file: <N/A>
|
| 541 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 542 |
+
------------------------------------------------------------
|
| 543 |
+
Root Cause (first observed failure):
|
| 544 |
+
[0]:
|
| 545 |
+
time : 2024-07-02_14:16:13
|
| 546 |
+
host : ip-26-0-170-31.ec2.internal
|
| 547 |
+
rank : 0 (local_rank: 0)
|
| 548 |
+
exitcode : 1 (pid: 2695547)
|
| 549 |
+
error_file: <N/A>
|
| 550 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 551 |
+
============================================================
|
| 552 |
+
srun: error: ip-26-0-170-31: task 0: Exited with exit code 1
|
| 553 |
+
[default0]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: c10d::allreduce_: an autograd kernel was not registered to the Autograd key(s) but we are trying to backprop through it. This may lead to silently incorrect behavior. This behavior is deprecated and will be removed in a future version of PyTorch. If your operator is differentiable, please ensure you have registered an autograd kernel to the correct Autograd key (e.g. DispatchKey::Autograd, DispatchKey::CompositeImplicitAutograd). If your operator is not differentiable, or to squash this warning and use the previous behavior, please register torch::CppFunction::makeFallthrough() to DispatchKey::Autograd. (Triggered internally at ../torch/csrc/autograd/autograd_not_implemented_fallback.cpp:63.)
|
| 554 |
+
[default0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
|
| 555 |
+
[default2]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: c10d::allreduce_: an autograd kernel was not registered to the Autograd key(s) but we are trying to backprop through it. This may lead to silently incorrect behavior. This behavior is deprecated and will be removed in a future version of PyTorch. If your operator is differentiable, please ensure you have registered an autograd kernel to the correct Autograd key (e.g. DispatchKey::Autograd, DispatchKey::CompositeImplicitAutograd). If your operator is not differentiable, or to squash this warning and use the previous behavior, please register torch::CppFunction::makeFallthrough() to DispatchKey::Autograd. (Triggered internally at ../torch/csrc/autograd/autograd_not_implemented_fallback.cpp:63.)
|
| 556 |
+
[default2]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
|
| 557 |
+
[default3]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: c10d::allreduce_: an autograd kernel was not registered to the Autograd key(s) but we are trying to backprop through it. This may lead to silently incorrect behavior. This behavior is deprecated and will be removed in a future version of PyTorch. If your operator is differentiable, please ensure you have registered an autograd kernel to the correct Autograd key (e.g. DispatchKey::Autograd, DispatchKey::CompositeImplicitAutograd). If your operator is not differentiable, or to squash this warning and use the previous behavior, please register torch::CppFunction::makeFallthrough() to DispatchKey::Autograd. (Triggered internally at ../torch/csrc/autograd/autograd_not_implemented_fallback.cpp:63.)
|
| 558 |
+
[default3]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
|
| 559 |
+
[default1]:/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/autograd/graph.py:744: UserWarning: c10d::allreduce_: an autograd kernel was not registered to the Autograd key(s) but we are trying to backprop through it. This may lead to silently incorrect behavior. This behavior is deprecated and will be removed in a future version of PyTorch. If your operator is differentiable, please ensure you have registered an autograd kernel to the correct Autograd key (e.g. DispatchKey::Autograd, DispatchKey::CompositeImplicitAutograd). If your operator is not differentiable, or to squash this warning and use the previous behavior, please register torch::CppFunction::makeFallthrough() to DispatchKey::Autograd. (Triggered internally at ../torch/csrc/autograd/autograd_not_implemented_fallback.cpp:63.)
|
| 560 |
+
[default1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
|
| 561 |
+
W0702 14:16:17.053000 140245491873536 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1252] The node 'ip-26-0-171-56.ec2.internal_2954948_0' has failed to send a keep-alive heartbeat to the rendezvous 'none' due to an error of type RendezvousConnectionError.
|
| 562 |
+
[default1]:[rank9]: Traceback (most recent call last):
|
| 563 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 564 |
+
[default2]:[rank10]: Traceback (most recent call last):
|
| 565 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 566 |
+
[default3]:[rank11]: Traceback (most recent call last):
|
| 567 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 568 |
+
[default3]:[rank11]: trainer.train(dataloader)
|
| 569 |
+
[default1]:[rank9]: trainer.train(dataloader)
|
| 570 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 571 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 572 |
+
[default2]:[rank10]: trainer.train(dataloader)
|
| 573 |
+
[default1]:[rank9]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 574 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 575 |
+
[default3]:[rank11]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 576 |
+
[default2]:[rank10]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 577 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 578 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 579 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 580 |
+
[default1]:[rank9]: outputs = self.pipeline_engine.train_batch_iter(
|
| 581 |
+
[default2]:[rank10]: outputs = self.pipeline_engine.train_batch_iter(
|
| 582 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 583 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 584 |
+
[default1]:[rank9]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 585 |
+
[default2]:[rank10]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 586 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 587 |
+
[default3]:[rank11]: outputs = self.pipeline_engine.train_batch_iter(
|
| 588 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 589 |
+
[default3]:[rank11]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 590 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 591 |
+
[default3]:[rank11]: output = model(**micro_batch)
|
| 592 |
+
[default2]:[rank10]: output = model(**micro_batch)
|
| 593 |
+
[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
|
| 594 |
+
[default2]:[rank10]: return self._call_impl(*args, **kwargs)
|
| 595 |
+
[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
|
| 596 |
+
[default2]:[rank10]: return forward_call(*args, **kwargs)
|
| 597 |
+
[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
|
| 598 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 599 |
+
[default2]:[rank10]: sharded_logits = self.model(
|
| 600 |
+
[default3]:[rank11]: return self._call_impl(*args, **kwargs)
|
| 601 |
+
[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
|
| 602 |
+
[default2]:[rank10]: return self._call_impl(*args, **kwargs)
|
| 603 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 604 |
+
[default1]:[rank9]: output = model(**micro_batch)
|
| 605 |
+
[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
|
| 606 |
+
[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
|
| 607 |
+
[default2]:[rank10]: return forward_call(*args, **kwargs)
|
| 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 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 610 |
+
[default1]:[rank9]: return self._call_impl(*args, **kwargs)
|
| 611 |
+
[default2]:[rank10]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 612 |
+
[default3]:[rank11]: return forward_call(*args, **kwargs)
|
| 613 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 614 |
+
[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
|
| 615 |
+
[default2]:[rank10]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 616 |
+
[default1]:[rank9]: return forward_call(*args, **kwargs)
|
| 617 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 618 |
+
[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
|
| 619 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 620 |
+
[default3]:[rank11]: sharded_logits = self.model(
|
| 621 |
+
[default2]:[rank10]: return self._call_impl(*args, **kwargs)
|
| 622 |
+
[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
|
| 623 |
+
[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
|
| 624 |
+
[default1]:[rank9]: sharded_logits = self.model(
|
| 625 |
+
[default2]:[rank10]: return forward_call(*args, **kwargs)
|
| 626 |
+
[default3]:[rank11]: return self._call_impl(*args, **kwargs)
|
| 627 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 126, in forward
|
| 628 |
+
[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
|
| 629 |
+
[default1]:[rank9]: return self._call_impl(*args, **kwargs)
|
| 630 |
+
[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
|
| 631 |
+
[default1]:[rank9]: return forward_call(*args, **kwargs)
|
| 632 |
+
[default2]:[rank10]: new_kwargs[name] = recv_from_pipeline_state_buffer(
|
| 633 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/functional.py", line 117, in recv_from_pipeline_state_buffer
|
| 634 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 635 |
+
[default2]:[rank10]: pipeline_state.run_communication()
|
| 636 |
+
[default1]:[rank9]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 637 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/state.py", line 160, in run_communication
|
| 638 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 639 |
+
[default2]:[rank10]: send_grad()
|
| 640 |
+
[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
|
| 641 |
+
[default1]:[rank9]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 642 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/state.py", line 41, in __call__
|
| 643 |
+
[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
|
| 644 |
+
[default3]:[rank11]: return forward_call(*args, **kwargs)
|
| 645 |
+
[default1]:[rank9]: return self._call_impl(*args, **kwargs)
|
| 646 |
+
[default2]:[rank10]: self.p2p.send_tensors([self.grad], to_rank=self.to_rank)
|
| 647 |
+
[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
|
| 648 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/p2p.py", line 348, in send_tensors
|
| 649 |
+
[default2]:[rank10]: futures = self.isend_tensors(tensors=tensors, to_rank=to_rank, tag=tag)
|
| 650 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 651 |
+
[default1]:[rank9]: return forward_call(*args, **kwargs)
|
| 652 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/p2p.py", line 295, in isend_tensors
|
| 653 |
+
[default3]:[rank11]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 654 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 126, in forward
|
| 655 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 656 |
+
[default1]:[rank9]: new_kwargs[name] = recv_from_pipeline_state_buffer(
|
| 657 |
+
[default3]:[rank11]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 658 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/functional.py", line 117, in recv_from_pipeline_state_buffer
|
| 659 |
+
[default2]:[rank10]: self._send_meta(tensor, to_rank=to_rank, tag=tag)
|
| 660 |
+
[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
|
| 661 |
+
[default1]:[rank9]: pipeline_state.run_communication()
|
| 662 |
+
[default3]:[rank11]: return self._call_impl(*args, **kwargs)
|
| 663 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/state.py", line 160, in run_communication
|
| 664 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/p2p.py", line 221, in _send_meta
|
| 665 |
+
[default1]:[rank9]: send_grad()
|
| 666 |
+
[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
|
| 667 |
+
[default2]:[rank10]: dist.send(
|
| 668 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/state.py", line 41, in __call__
|
| 669 |
+
[default1]:[rank9]: self.p2p.send_tensors([self.grad], to_rank=self.to_rank)
|
| 670 |
+
[default3]:[rank11]: return forward_call(*args, **kwargs)
|
| 671 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/c10d_logger.py", line 75, in wrapper
|
| 672 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/p2p.py", line 348, in send_tensors
|
| 673 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 126, in forward
|
| 674 |
+
[default1]:[rank9]: futures = self.isend_tensors(tensors=tensors, to_rank=to_rank, tag=tag)
|
| 675 |
+
[default3]:[rank11]: new_kwargs[name] = recv_from_pipeline_state_buffer(
|
| 676 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/p2p.py", line 295, in isend_tensors
|
| 677 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/functional.py", line 117, in recv_from_pipeline_state_buffer
|
| 678 |
+
[default1]:[rank9]: self._send_meta(tensor, to_rank=to_rank, tag=tag)
|
| 679 |
+
[default3]:[rank11]: pipeline_state.run_communication()
|
| 680 |
+
[default2]:[rank10]: return func(*args, **kwargs)
|
| 681 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/p2p.py", line 221, in _send_meta
|
| 682 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/state.py", line 160, in run_communication
|
| 683 |
+
[default3]:[rank11]: send_grad()
|
| 684 |
+
[default1]:[rank9]: dist.send(
|
| 685 |
+
[default2]:[rank10]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py", line 1886, in send
|
| 686 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/state.py", line 41, in __call__
|
| 687 |
+
[default2]:[rank10]: group.send([tensor], group_dst_rank, tag).wait()
|
| 688 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/c10d_logger.py", line 75, in wrapper
|
| 689 |
+
[default2]:[rank10]: torch.distributed.DistBackendError: NCCL communicator was aborted on rank 1.
|
| 690 |
+
[default1]:[rank9]: return func(*args, **kwargs)
|
| 691 |
+
[default3]:[rank11]: self.p2p.send_tensors([self.grad], to_rank=self.to_rank)
|
| 692 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/p2p.py", line 348, in send_tensors
|
| 693 |
+
[default3]:[rank11]: futures = self.isend_tensors(tensors=tensors, to_rank=to_rank, tag=tag)
|
| 694 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/p2p.py", line 295, in isend_tensors
|
| 695 |
+
[default1]:[rank9]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py", line 1886, in send
|
| 696 |
+
[default3]:[rank11]: self._send_meta(tensor, to_rank=to_rank, tag=tag)
|
| 697 |
+
[default1]:[rank9]: group.send([tensor], group_dst_rank, tag).wait()
|
| 698 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/p2p.py", line 221, in _send_meta
|
| 699 |
+
[default1]:[rank9]: torch.distributed.DistBackendError: NCCL communicator was aborted on rank 1.
|
| 700 |
+
[default3]:[rank11]: dist.send(
|
| 701 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/c10d_logger.py", line 75, in wrapper
|
| 702 |
+
[default3]:[rank11]: return func(*args, **kwargs)
|
| 703 |
+
[default3]:[rank11]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py", line 1886, in send
|
| 704 |
+
[default3]:[rank11]: group.send([tensor], group_dst_rank, tag).wait()
|
| 705 |
+
[default3]:[rank11]: torch.distributed.DistBackendError: NCCL communicator was aborted on rank 1.
|
| 706 |
+
[default0]:[rank8]: Traceback (most recent call last):
|
| 707 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 708 |
+
[default0]:[rank8]: trainer.train(dataloader)
|
| 709 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 710 |
+
[default0]:[rank8]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 711 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 712 |
+
[default0]:[rank8]: outputs = self.pipeline_engine.train_batch_iter(
|
| 713 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 714 |
+
[default0]:[rank8]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 715 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 716 |
+
[default0]:[rank8]: output = model(**micro_batch)
|
| 717 |
+
[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
|
| 718 |
+
[default0]:[rank8]: return self._call_impl(*args, **kwargs)
|
| 719 |
+
[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
|
| 720 |
+
[default0]:[rank8]: return forward_call(*args, **kwargs)
|
| 721 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 722 |
+
[default0]:[rank8]: sharded_logits = self.model(
|
| 723 |
+
[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
|
| 724 |
+
[default0]:[rank8]: return self._call_impl(*args, **kwargs)
|
| 725 |
+
[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
|
| 726 |
+
[default0]:[rank8]: return forward_call(*args, **kwargs)
|
| 727 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 728 |
+
[default0]:[rank8]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 729 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 730 |
+
[default0]:[rank8]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 731 |
+
[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
|
| 732 |
+
[default0]:[rank8]: return self._call_impl(*args, **kwargs)
|
| 733 |
+
[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
|
| 734 |
+
[default0]:[rank8]: return forward_call(*args, **kwargs)
|
| 735 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 126, in forward
|
| 736 |
+
[default0]:[rank8]: new_kwargs[name] = recv_from_pipeline_state_buffer(
|
| 737 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/functional.py", line 117, in recv_from_pipeline_state_buffer
|
| 738 |
+
[default0]:[rank8]: pipeline_state.run_communication()
|
| 739 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/state.py", line 160, in run_communication
|
| 740 |
+
[default0]:[rank8]: send_grad()
|
| 741 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/state.py", line 41, in __call__
|
| 742 |
+
[default0]:[rank8]: self.p2p.send_tensors([self.grad], to_rank=self.to_rank)
|
| 743 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/p2p.py", line 348, in send_tensors
|
| 744 |
+
[default0]:[rank8]: futures = self.isend_tensors(tensors=tensors, to_rank=to_rank, tag=tag)
|
| 745 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/p2p.py", line 295, in isend_tensors
|
| 746 |
+
[default0]:[rank8]: self._send_meta(tensor, to_rank=to_rank, tag=tag)
|
| 747 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/p2p.py", line 221, in _send_meta
|
| 748 |
+
[default0]:[rank8]: dist.send(
|
| 749 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/c10d_logger.py", line 75, in wrapper
|
| 750 |
+
[default0]:[rank8]: return func(*args, **kwargs)
|
| 751 |
+
[default0]:[rank8]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py", line 1886, in send
|
| 752 |
+
[default0]:[rank8]: group.send([tensor], group_dst_rank, tag).wait()
|
| 753 |
+
[default0]:[rank8]: torch.distributed.DistBackendError: NCCL communicator was aborted on rank 1.
|
| 754 |
+
W0702 14:16:18.050000 140251158693696 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2955028 closing signal SIGTERM
|
| 755 |
+
W0702 14:16:18.050000 140251158693696 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2955029 closing signal SIGTERM
|
| 756 |
+
W0702 14:16:18.053000 140251158693696 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2955030 closing signal SIGTERM
|
| 757 |
+
W0702 14:16:18.054000 140251158693696 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2955031 closing signal SIGTERM
|
| 758 |
+
W0702 14:16:18.054000 140251158693696 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2955032 closing signal SIGTERM
|
| 759 |
+
W0702 14:16:18.056000 140251158693696 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2955033 closing signal SIGTERM
|
| 760 |
+
W0702 14:16:18.064000 140251158693696 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2955034 closing signal SIGTERM
|
| 761 |
+
W0702 14:16:18.071000 140251158693696 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 2955035 closing signal SIGTERM
|
| 762 |
+
W0702 14:16:21.689000 140251158693696 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1203] The node 'ip-26-0-171-56.ec2.internal_2954948_0' has failed to shutdown the rendezvous 'none' due to an error of type RendezvousConnectionError.
|
| 763 |
+
W0702 14:16:21.700000 140251158693696 torch/distributed/elastic/rendezvous/dynamic_rendezvous.py:1203] The node 'ip-26-0-171-56.ec2.internal_2954948_0' has failed to shutdown the rendezvous 'none' due to an error of type RendezvousConnectionError.
|
| 764 |
+
Traceback (most recent call last):
|
| 765 |
+
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
|
| 766 |
+
return getattr(self._store, store_op)(*args, **kwargs)
|
| 767 |
+
torch.distributed.DistNetworkError: Broken pipe
|
| 768 |
+
|
| 769 |
+
The above exception was the direct cause of the following exception:
|
| 770 |
+
|
| 771 |
+
Traceback (most recent call last):
|
| 772 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/torchrun", line 8, in <module>
|
| 773 |
+
sys.exit(main())
|
| 774 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 347, in wrapper
|
| 775 |
+
return f(*args, **kwargs)
|
| 776 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 879, in main
|
| 777 |
+
run(args)
|
| 778 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 870, in run
|
| 779 |
+
elastic_launch(
|
| 780 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 132, in __call__
|
| 781 |
+
return launch_agent(self._config, self._entrypoint, list(args))
|
| 782 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 254, in launch_agent
|
| 783 |
+
result = agent.run()
|
| 784 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/metrics/api.py", line 123, in wrapper
|
| 785 |
+
result = f(*args, **kwargs)
|
| 786 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/agent/server/api.py", line 733, in run
|
| 787 |
+
result = self._invoke_run(role)
|
| 788 |
+
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
|
| 789 |
+
num_nodes_waiting = rdzv_handler.num_nodes_waiting()
|
| 790 |
+
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
|
| 791 |
+
self._state_holder.sync()
|
| 792 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/rendezvous/dynamic_rendezvous.py", line 419, in sync
|
| 793 |
+
get_response = self._backend.get_state()
|
| 794 |
+
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
|
| 795 |
+
base64_state: bytes = self._call_store("get", self._key)
|
| 796 |
+
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
|
| 797 |
+
raise RendezvousConnectionError(
|
| 798 |
+
torch.distributed.elastic.rendezvous.api.RendezvousConnectionError: The connection to the C10d store has failed. See inner exception for details.
|
| 799 |
+
srun: error: ip-26-0-171-56: task 1: Exited with exit code 1
|
| 800 |
+
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-1_tp-4_pp-4_mbz-32/status.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
oom
|