Upload llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16
Browse files
llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/bench.slurm
ADDED
|
@@ -0,0 +1,111 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/bin/bash
|
| 2 |
+
|
| 3 |
+
#SBATCH --job-name=bench_cluster
|
| 4 |
+
#SBATCH --time=02:00:00
|
| 5 |
+
#SBATCH --partition=hopper-prod
|
| 6 |
+
#SBATCH --nodes=1
|
| 7 |
+
#SBATCH --gres=gpu:8
|
| 8 |
+
#SBATCH --qos=normal
|
| 9 |
+
#SBATCH --ntasks-per-node=1
|
| 10 |
+
#SBATCH --cpus-per-task=96
|
| 11 |
+
#SBATCH --exclusive
|
| 12 |
+
#SBATCH --output=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/log.out
|
| 13 |
+
#SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/log.out
|
| 14 |
+
|
| 15 |
+
# Function to update status based on squeue output
|
| 16 |
+
update_status() {
|
| 17 |
+
job_id=$1
|
| 18 |
+
status_file=$2
|
| 19 |
+
# For unknown reasons, it doenst update status for pending. It only works for running
|
| 20 |
+
while true; do
|
| 21 |
+
job_status=$(squeue --job $job_id --noheader --format=%T)
|
| 22 |
+
echo "Job status: $job_status"
|
| 23 |
+
if [ -z "$job_status" ]; then
|
| 24 |
+
# Job has finished or is not found
|
| 25 |
+
break
|
| 26 |
+
elif [ "$job_status" = "RUNNING" ]; then
|
| 27 |
+
printf "running" > $status_file
|
| 28 |
+
break
|
| 29 |
+
fi
|
| 30 |
+
sleep 10
|
| 31 |
+
done
|
| 32 |
+
}
|
| 33 |
+
|
| 34 |
+
# Misc initializations.
|
| 35 |
+
echo "========================"
|
| 36 |
+
echo "START TIME: $(date)"
|
| 37 |
+
source /fsx/ferdinandmom/miniforge3/etc/profile.d/conda.sh
|
| 38 |
+
conda activate /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster
|
| 39 |
+
echo python3 version = $(python3 --version)
|
| 40 |
+
echo "========================"
|
| 41 |
+
|
| 42 |
+
# Slurm stuff
|
| 43 |
+
export HOSTNAMES=$(scontrol show hostnames "$SLURM_JOB_NODELIST")
|
| 44 |
+
export MASTER_ADDR=$(scontrol show hostnames "$SLURM_JOB_NODELIST" | head -n 1)
|
| 45 |
+
export MASTER_PORT=$((1024 + RANDOM % 64511))
|
| 46 |
+
|
| 47 |
+
export TMPDIR=/scratch
|
| 48 |
+
export HF_DATASETS_CACHE="/admin/home/ferdinand_mom/.cache"
|
| 49 |
+
export CUBLAS_WORKSPACE_CONFIG=":4096:8"
|
| 50 |
+
export CUDA_DEVICE_MAX_CONNECTIONS="1"
|
| 51 |
+
|
| 52 |
+
huggingface-cli login --token $HUGGINGFACE_TOKEN
|
| 53 |
+
|
| 54 |
+
|
| 55 |
+
NANOTRON_REPO="/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron"
|
| 56 |
+
CMD="$NANOTRON_REPO/run_train.py --config-file /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/config.yaml"
|
| 57 |
+
|
| 58 |
+
LAUNCHER="torchrun \
|
| 59 |
+
--nproc_per_node 8 \
|
| 60 |
+
--nnodes 1 \
|
| 61 |
+
--rdzv_endpoint ${MASTER_ADDR}:${MASTER_PORT} \
|
| 62 |
+
--rdzv_backend c10d \
|
| 63 |
+
--max_restarts 0 \
|
| 64 |
+
--tee 3 \
|
| 65 |
+
--node_rank ${SLURM_PROCID}"
|
| 66 |
+
|
| 67 |
+
# Checkout the bench_cluster branch
|
| 68 |
+
cd $NANOTRON_REPO
|
| 69 |
+
git checkout bench_cluster
|
| 70 |
+
cd ..
|
| 71 |
+
# Get the current job ID
|
| 72 |
+
job_id=${SLURM_JOB_ID}
|
| 73 |
+
|
| 74 |
+
# Update status to "pending" or "running" in the background
|
| 75 |
+
update_status $job_id /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/status.txt &
|
| 76 |
+
|
| 77 |
+
# Run the main command
|
| 78 |
+
srun -u $LAUNCHER $CMD
|
| 79 |
+
exit_status=$?
|
| 80 |
+
|
| 81 |
+
# Update status based on the exit status of `srun`
|
| 82 |
+
if [ $exit_status -eq 0 ]; then
|
| 83 |
+
printf "completed" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/status.txt
|
| 84 |
+
else
|
| 85 |
+
if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/log.out; then
|
| 86 |
+
printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/status.txt
|
| 87 |
+
elif grep -q " CUDA error: an illegal memory access" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/log.out; then
|
| 88 |
+
printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/status.txt
|
| 89 |
+
elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/log.out; then
|
| 90 |
+
printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/status.txt
|
| 91 |
+
else
|
| 92 |
+
printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/status.txt
|
| 93 |
+
fi
|
| 94 |
+
fi
|
| 95 |
+
|
| 96 |
+
# Run the report script if the job completed successfully
|
| 97 |
+
if [ $exit_status -eq 0 ]; then
|
| 98 |
+
python /fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py report --inp_dir /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16 --is_logs
|
| 99 |
+
python /fsx/ferdinandmom/ferdinand-hf/bench_cluster/main.py report --inp_dir /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16 --is_profiler
|
| 100 |
+
fi
|
| 101 |
+
|
| 102 |
+
|
| 103 |
+
# Push to hub the folder using huggingface_cli
|
| 104 |
+
huggingface-cli upload nanotron/bench_cluster /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16 llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16 --commit-message "Upload llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16"
|
| 105 |
+
|
| 106 |
+
# Verify the upload
|
| 107 |
+
if [ $? -eq 0 ]; then
|
| 108 |
+
echo "Uploading to Huggingface Hub successful"
|
| 109 |
+
else
|
| 110 |
+
echo "Failed to upload to Huggingface Hub"
|
| 111 |
+
fi
|
llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/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: 8
|
| 49 |
+
expert_parallel_size: 1
|
| 50 |
+
pp: 1
|
| 51 |
+
pp_engine: 1f1b
|
| 52 |
+
tp: 1
|
| 53 |
+
tp_linear_async_communication: false
|
| 54 |
+
tp_mode: REDUCE_SCATTER
|
| 55 |
+
profiler:
|
| 56 |
+
profiler_export_path: /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16
|
| 57 |
+
tokenizer:
|
| 58 |
+
tokenizer_max_length: null
|
| 59 |
+
tokenizer_name_or_path: openai-community/gpt2
|
| 60 |
+
tokenizer_revision: null
|
| 61 |
+
data_stages:
|
| 62 |
+
- name: Training Stage
|
| 63 |
+
start_training_step: 1
|
| 64 |
+
data:
|
| 65 |
+
dataset:
|
| 66 |
+
dataset_overwrite_cache: false
|
| 67 |
+
dataset_processing_num_proc_per_process: 64
|
| 68 |
+
hf_dataset_config_name: null
|
| 69 |
+
hf_dataset_or_datasets: roneneldan/TinyStories
|
| 70 |
+
hf_dataset_splits: train
|
| 71 |
+
text_column_name: text
|
| 72 |
+
num_loading_workers: 0
|
| 73 |
+
seed: 42
|
| 74 |
+
lighteval: null
|
| 75 |
+
tokens:
|
| 76 |
+
train_steps: 20
|
| 77 |
+
val_check_interval: -1
|
| 78 |
+
batch_accumulation_per_replica: 8
|
| 79 |
+
limit_test_batches: 0
|
| 80 |
+
limit_val_batches: 0
|
| 81 |
+
micro_batch_size: 16
|
| 82 |
+
sequence_length: 4096
|
| 83 |
+
logging:
|
| 84 |
+
iteration_step_info_interval: 1
|
| 85 |
+
log_level: info
|
| 86 |
+
log_level_replica: info
|
| 87 |
+
checkpoints:
|
| 88 |
+
checkpoint_interval: 100000
|
| 89 |
+
checkpoints_path: /dev/null
|
| 90 |
+
resume_checkpoint_path: null
|
llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/log.out
ADDED
|
@@ -0,0 +1,663 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
========================
|
| 2 |
+
START TIME: Wed Jul 3 22:49:00 UTC 2024
|
| 3 |
+
python3 version = Python 3.10.14
|
| 4 |
+
========================
|
| 5 |
+
The token has not been saved to the git credentials helper. Pass `add_to_git_credential=True` in this function directly or `--add-to-git-credential` if using via `huggingface-cli` if you want to set the git credential as well.
|
| 6 |
+
Token is valid (permission: write).
|
| 7 |
+
Your token has been saved to /admin/home/ferdinand_mom/.cache/huggingface/token
|
| 8 |
+
Login successful
|
| 9 |
+
Already on 'bench_cluster'
|
| 10 |
+
M examples/config_tiny_llama.py
|
| 11 |
+
M examples/config_tiny_llama.yaml
|
| 12 |
+
M examples/train_tiny_llama.sh
|
| 13 |
+
M src/nanotron/models/llama.py
|
| 14 |
+
M src/nanotron/trainer.py
|
| 15 |
+
Your branch is up to date with 'origin/bench_cluster'.
|
| 16 |
+
Job status: RUNNING
|
| 17 |
+
W0703 22:49:02.624000 139633926510400 torch/distributed/run.py:757]
|
| 18 |
+
W0703 22:49:02.624000 139633926510400 torch/distributed/run.py:757] *****************************************
|
| 19 |
+
W0703 22:49:02.624000 139633926510400 torch/distributed/run.py:757] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
|
| 20 |
+
W0703 22:49:02.624000 139633926510400 torch/distributed/run.py:757] *****************************************
|
| 21 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: Config:
|
| 22 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: Config(general=GeneralArgs(project='bench_cluster',
|
| 23 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: run='%date_%jobid',
|
| 24 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: seed=42,
|
| 25 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: step=None,
|
| 26 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: consumed_train_samples=None,
|
| 27 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: benchmark_csv_path=None,
|
| 28 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: ignore_sanity_checks=True),
|
| 29 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: parallelism=ParallelismArgs(dp=8,
|
| 30 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: pp=1,
|
| 31 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: tp=1,
|
| 32 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7fc9030b48b0>,
|
| 33 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
|
| 34 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: tp_linear_async_communication=False,
|
| 35 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: expert_parallel_size=1),
|
| 36 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
|
| 37 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: eos_token_id=2,
|
| 38 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: hidden_act='silu',
|
| 39 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: hidden_size=2048,
|
| 40 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: initializer_range=0.02,
|
| 41 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: intermediate_size=4096,
|
| 42 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: is_llama_config=True,
|
| 43 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: max_position_embeddings=4096,
|
| 44 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: num_attention_heads=32,
|
| 45 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: num_hidden_layers=24,
|
| 46 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: num_key_value_heads=32,
|
| 47 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: pad_token_id=None,
|
| 48 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: pretraining_tp=1,
|
| 49 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: rms_norm_eps=1e-05,
|
| 50 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: rope_scaling=None,
|
| 51 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: rope_theta=10000.0,
|
| 52 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: tie_word_embeddings=True,
|
| 53 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: use_cache=True,
|
| 54 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: vocab_size=50257),
|
| 55 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: init_method=RandomInit(std=0.025),
|
| 56 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: dtype=torch.bfloat16,
|
| 57 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: make_vocab_size_divisible_by=1,
|
| 58 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: ddp_bucket_cap_mb=25),
|
| 59 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
|
| 60 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: tokenizer_revision=None,
|
| 61 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: tokenizer_max_length=None),
|
| 62 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
|
| 63 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: checkpoint_interval=100000,
|
| 64 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: save_initial_state=False,
|
| 65 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: resume_checkpoint_path=None,
|
| 66 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: checkpoints_path_is_shared_file_system=False),
|
| 67 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: logging=LoggingArgs(log_level='info',
|
| 68 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: log_level_replica='info',
|
| 69 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: iteration_step_info_interval=1),
|
| 70 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: tokens=TokensArgs(sequence_length=4096,
|
| 71 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: train_steps=20,
|
| 72 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: micro_batch_size=16,
|
| 73 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: batch_accumulation_per_replica=8,
|
| 74 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: val_check_interval=-1,
|
| 75 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: limit_val_batches=0,
|
| 76 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: limit_test_batches=0),
|
| 77 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
|
| 78 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: adam_beta1=0.9,
|
| 79 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: adam_beta2=0.95,
|
| 80 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: torch_adam_is_fused=True,
|
| 81 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: name='adamW'),
|
| 82 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: zero_stage=1,
|
| 83 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: weight_decay=0.01,
|
| 84 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: clip_grad=1.0,
|
| 85 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: accumulate_grad_in_fp32=True,
|
| 86 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
|
| 87 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: lr_warmup_steps=1,
|
| 88 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: lr_warmup_style='linear',
|
| 89 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: lr_decay_style='linear',
|
| 90 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: lr_decay_steps=19,
|
| 91 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: lr_decay_starting_step=None,
|
| 92 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: min_decay_lr=1e-05)),
|
| 93 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: data_stages=[DatasetStageArgs(name='Training Stage',
|
| 94 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: start_training_step=1,
|
| 95 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
|
| 96 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: hf_dataset_splits='train',
|
| 97 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: hf_dataset_config_name=None,
|
| 98 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: dataset_processing_num_proc_per_process=64,
|
| 99 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: dataset_overwrite_cache=False,
|
| 100 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: text_column_name='text'),
|
| 101 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: seed=42,
|
| 102 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: num_loading_workers=0))],
|
| 103 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16')),
|
| 104 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: lighteval=None)
|
| 105 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: Model Config:
|
| 106 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: LlamaConfig(bos_token_id=1,
|
| 107 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: eos_token_id=2,
|
| 108 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: hidden_act='silu',
|
| 109 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: hidden_size=2048,
|
| 110 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: initializer_range=0.02,
|
| 111 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: intermediate_size=4096,
|
| 112 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: is_llama_config=True,
|
| 113 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: max_position_embeddings=4096,
|
| 114 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: num_attention_heads=32,
|
| 115 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: num_hidden_layers=24,
|
| 116 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: num_key_value_heads=32,
|
| 117 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: pad_token_id=None,
|
| 118 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: pretraining_tp=1,
|
| 119 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: rms_norm_eps=1e-05,
|
| 120 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: rope_scaling=None,
|
| 121 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: rope_theta=10000.0,
|
| 122 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: tie_word_embeddings=True,
|
| 123 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: use_cache=True,
|
| 124 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: vocab_size=50257)
|
| 125 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: Building model..
|
| 126 |
+
[default0]:07/03/2024 22:49:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: Setting PP block ranks...
|
| 127 |
+
[default1]:07/03/2024 22:49:27 [INFO|DP=1|PP=0|TP=0|ip-26-0-161-178]: No checkpoint path provided.
|
| 128 |
+
[default0]:07/03/2024 22:49:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: Total number of parameters: 1.11G (2116.51MiB)
|
| 129 |
+
[default0]:07/03/2024 22:49:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: Local number of parameters: 1.11G (2116.51MiB)
|
| 130 |
+
[default0]:07/03/2024 22:49:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: [After model building] Memory usage: 2140.53MiB. Peak allocated: 2338.88MiB Peak reserved: 2392.00MiB
|
| 131 |
+
[default0]:07/03/2024 22:49:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: No checkpoint path provided.
|
| 132 |
+
[default0]:07/03/2024 22:49:27 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: Parametrizing model parameters using StandardParametrizator
|
| 133 |
+
[default5]:07/03/2024 22:49:27 [INFO|DP=5|PP=0|TP=0|ip-26-0-161-178]: No checkpoint path provided.
|
| 134 |
+
[default6]:07/03/2024 22:49:27 [INFO|DP=6|PP=0|TP=0|ip-26-0-161-178]: No checkpoint path provided.
|
| 135 |
+
[default4]:07/03/2024 22:49:27 [INFO|DP=4|PP=0|TP=0|ip-26-0-161-178]: No checkpoint path provided.
|
| 136 |
+
[default2]:07/03/2024 22:49:27 [INFO|DP=2|PP=0|TP=0|ip-26-0-161-178]: No checkpoint path provided.
|
| 137 |
+
[default3]:07/03/2024 22:49:27 [INFO|DP=3|PP=0|TP=0|ip-26-0-161-178]: No checkpoint path provided.
|
| 138 |
+
[default7]:07/03/2024 22:49:27 [INFO|DP=7|PP=0|TP=0|ip-26-0-161-178]: No checkpoint path provided.
|
| 139 |
+
[default0]:07/03/2024 22:49:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: [Optimizer Building] Using LearningRateForSP as learning rate
|
| 140 |
+
[default0]:07/03/2024 22:49:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: [ZeRO sharding] Size of optimizer params per rank:
|
| 141 |
+
[default0]:07/03/2024 22:49:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: [ZeRO sharding] DP Rank 0 has 139M out of 1.11G (12.50%) params' optimizer states
|
| 142 |
+
[default0]:07/03/2024 22:49:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: [ZeRO sharding] DP Rank 1 has 139M out of 1.11G (12.50%) params' optimizer states
|
| 143 |
+
[default0]:07/03/2024 22:49:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: [ZeRO sharding] DP Rank 2 has 139M out of 1.11G (12.50%) params' optimizer states
|
| 144 |
+
[default0]:07/03/2024 22:49:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: [ZeRO sharding] DP Rank 3 has 139M out of 1.11G (12.50%) params' optimizer states
|
| 145 |
+
[default0]:07/03/2024 22:49:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: [ZeRO sharding] DP Rank 4 has 139M out of 1.11G (12.50%) params' optimizer states
|
| 146 |
+
[default0]:07/03/2024 22:49:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: [ZeRO sharding] DP Rank 5 has 139M out of 1.11G (12.50%) params' optimizer states
|
| 147 |
+
[default0]:07/03/2024 22:49:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: [ZeRO sharding] DP Rank 6 has 139M out of 1.11G (12.50%) params' optimizer states
|
| 148 |
+
[default0]:07/03/2024 22:49:34 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: [ZeRO sharding] DP Rank 7 has 139M out of 1.11G (12.50%) params' optimizer states
|
| 149 |
+
[default0]:07/03/2024 22:49:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
|
| 150 |
+
[default0]:07/03/2024 22:49:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: Using `datasets` library
|
| 151 |
+
[default0]:07/03/2024 22:49:35 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
|
| 152 |
+
[default0]:07/03/2024 22:49:35 [WARNING|DP=0|PP=0|TP=0|ip-26-0-161-178]: Repo card metadata block was not found. Setting CardData to empty.
|
| 153 |
+
[default0]:Repo card metadata block was not found. Setting CardData to empty.
|
| 154 |
+
[default0]:07/03/2024 22:49:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: [Training Plan] There are 1 training stages
|
| 155 |
+
[default0]:07/03/2024 22:49:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: [Stage Training Stage] start from step 1
|
| 156 |
+
[default0]:07/03/2024 22:49:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]:
|
| 157 |
+
[default0]:07/03/2024 22:49:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: [Start training] datetime: 2024-07-03 22:49:36.207595 | mbs: 16 | grad_accum: 8 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
|
| 158 |
+
[default0]:07/03/2024 22:49:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
|
| 159 |
+
[default0]:07/03/2024 22:49:36 [INFO|DP=0|PP=0|TP=0|ip-26-0-161-178]: Memory usage: 6904.53MiB. Peak allocated 6904.53MiB. Peak reserved: 7156.00MiB
|
| 160 |
+
[default1]:07/03/2024 22:49:36 [WARNING|DP=1|PP=0|TP=0|ip-26-0-161-178]: Repo card metadata block was not found. Setting CardData to empty.
|
| 161 |
+
[default5]:07/03/2024 22:49:36 [WARNING|DP=5|PP=0|TP=0|ip-26-0-161-178]: Repo card metadata block was not found. Setting CardData to empty.
|
| 162 |
+
[default1]:Repo card metadata block was not found. Setting CardData to empty.
|
| 163 |
+
[default4]:07/03/2024 22:49:36 [WARNING|DP=4|PP=0|TP=0|ip-26-0-161-178]: Repo card metadata block was not found. Setting CardData to empty.
|
| 164 |
+
[default6]:Repo card metadata block was not found. Setting CardData to empty.
|
| 165 |
+
[default7]:Repo card metadata block was not found. Setting CardData to empty.
|
| 166 |
+
[default6]:07/03/2024 22:49:36 [WARNING|DP=6|PP=0|TP=0|ip-26-0-161-178]: Repo card metadata block was not found. Setting CardData to empty.
|
| 167 |
+
[default7]:07/03/2024 22:49:36 [WARNING|DP=7|PP=0|TP=0|ip-26-0-161-178]: Repo card metadata block was not found. Setting CardData to empty.
|
| 168 |
+
[default3]:Repo card metadata block was not found. Setting CardData to empty.
|
| 169 |
+
[default2]:07/03/2024 22:49:36 [WARNING|DP=2|PP=0|TP=0|ip-26-0-161-178]: Repo card metadata block was not found. Setting CardData to empty.
|
| 170 |
+
[default2]:Repo card metadata block was not found. Setting CardData to empty.
|
| 171 |
+
[default4]:Repo card metadata block was not found. Setting CardData to empty.
|
| 172 |
+
[default3]:07/03/2024 22:49:36 [WARNING|DP=3|PP=0|TP=0|ip-26-0-161-178]: Repo card metadata block was not found. Setting CardData to empty.
|
| 173 |
+
[default5]:Repo card metadata block was not found. Setting CardData to empty.
|
| 174 |
+
[default0]:[rank0]: Traceback (most recent call last):
|
| 175 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 176 |
+
[default0]:[rank0]: trainer.train(dataloader)
|
| 177 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 178 |
+
[default0]:[rank0]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 179 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 180 |
+
[default0]:[rank0]: outputs = self.pipeline_engine.train_batch_iter(
|
| 181 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 182 |
+
[default0]:[rank0]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 183 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 184 |
+
[default0]:[rank0]: output = model(**micro_batch)
|
| 185 |
+
[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
|
| 186 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 187 |
+
[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
|
| 188 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 189 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 190 |
+
[default0]:[rank0]: sharded_logits = self.model(
|
| 191 |
+
[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
|
| 192 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 193 |
+
[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
|
| 194 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 195 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 196 |
+
[default0]:[rank0]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 197 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 198 |
+
[default0]:[rank0]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 199 |
+
[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
|
| 200 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 201 |
+
[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
|
| 202 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 203 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 204 |
+
[default0]:[rank0]: output = self.pp_block(**new_kwargs)
|
| 205 |
+
[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
|
| 206 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 207 |
+
[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
|
| 208 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 209 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 210 |
+
[default0]:[rank0]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 211 |
+
[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
|
| 212 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 213 |
+
[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
|
| 214 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 215 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 216 |
+
[default0]:[rank0]: merged_states = self.gate_up_proj(hidden_states)
|
| 217 |
+
[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
|
| 218 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 219 |
+
[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
|
| 220 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 221 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 222 |
+
[default0]:[rank0]: return column_linear(
|
| 223 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 224 |
+
[default0]:[rank0]: return F.linear(input, weight, bias)
|
| 225 |
+
[default0]:[rank0]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 1024.00 MiB. GPU
|
| 226 |
+
[default1]:[rank1]: Traceback (most recent call last):
|
| 227 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 228 |
+
[default1]:[rank1]: trainer.train(dataloader)
|
| 229 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 230 |
+
[default1]:[rank1]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 231 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 232 |
+
[default1]:[rank1]: outputs = self.pipeline_engine.train_batch_iter(
|
| 233 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 234 |
+
[default1]:[rank1]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 235 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 236 |
+
[default1]:[rank1]: output = model(**micro_batch)
|
| 237 |
+
[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
|
| 238 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 239 |
+
[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
|
| 240 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 241 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 242 |
+
[default1]:[rank1]: sharded_logits = self.model(
|
| 243 |
+
[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
|
| 244 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 245 |
+
[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
|
| 246 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 247 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 248 |
+
[default1]:[rank1]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 249 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 250 |
+
[default1]:[rank1]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 251 |
+
[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
|
| 252 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 253 |
+
[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
|
| 254 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 255 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 256 |
+
[default1]:[rank1]: output = self.pp_block(**new_kwargs)
|
| 257 |
+
[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
|
| 258 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 259 |
+
[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
|
| 260 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 261 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 262 |
+
[default1]:[rank1]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 263 |
+
[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
|
| 264 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 265 |
+
[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
|
| 266 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 267 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 268 |
+
[default1]:[rank1]: merged_states = self.gate_up_proj(hidden_states)
|
| 269 |
+
[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
|
| 270 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 271 |
+
[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
|
| 272 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 273 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 274 |
+
[default1]:[rank1]: return column_linear(
|
| 275 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 276 |
+
[default1]:[rank1]: return F.linear(input, weight, bias)
|
| 277 |
+
[default1]:[rank1]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 1024.00 MiB. GPU has a total capacity of 79.33 GiB of which 289.94 MiB is free. Including non-PyTorch memory, this process has 79.04 GiB memory in use. Of the allocated memory 66.89 GiB is allocated by PyTorch, and 417.33 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)
|
| 278 |
+
[default7]:[rank7]: Traceback (most recent call last):
|
| 279 |
+
[default6]:[rank6]: Traceback (most recent call last):
|
| 280 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 281 |
+
[default3]:[rank3]: Traceback (most recent call last):
|
| 282 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 283 |
+
[default6]:[rank6]: trainer.train(dataloader)
|
| 284 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 285 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 286 |
+
[default7]:[rank7]: trainer.train(dataloader)
|
| 287 |
+
[default6]:[rank6]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 288 |
+
[default3]:[rank3]: trainer.train(dataloader)
|
| 289 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 290 |
+
[default7]:[rank7]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 291 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 292 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 293 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 294 |
+
[default6]:[rank6]: outputs = self.pipeline_engine.train_batch_iter(
|
| 295 |
+
[default3]:[rank3]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 296 |
+
[default7]:[rank7]: outputs = self.pipeline_engine.train_batch_iter(
|
| 297 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 298 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 299 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 300 |
+
[default7]:[rank7]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 301 |
+
[default3]:[rank3]: outputs = self.pipeline_engine.train_batch_iter(
|
| 302 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 303 |
+
[default6]:[rank6]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 304 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 305 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 306 |
+
[default3]:[rank3]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 307 |
+
[default6]:[rank6]: output = model(**micro_batch)
|
| 308 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 309 |
+
[default7]:[rank7]: output = model(**micro_batch)
|
| 310 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 311 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 312 |
+
[default3]:[rank3]: output = model(**micro_batch)
|
| 313 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 314 |
+
[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
|
| 315 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 316 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 317 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 318 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 319 |
+
[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
|
| 320 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 321 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 322 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 323 |
+
[default7]:[rank7]: sharded_logits = self.model(
|
| 324 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 325 |
+
[default6]:[rank6]: sharded_logits = self.model(
|
| 326 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 327 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 328 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 329 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 330 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 331 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 332 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 333 |
+
[default4]:[rank4]: Traceback (most recent call last):
|
| 334 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 335 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 336 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 337 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 338 |
+
[default3]:[rank3]: sharded_logits = self.model(
|
| 339 |
+
[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
|
| 340 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 341 |
+
[default4]:[rank4]: trainer.train(dataloader)
|
| 342 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 343 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 344 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 345 |
+
[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
|
| 346 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 347 |
+
[default6]:[rank6]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 348 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 349 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 350 |
+
[default2]:[rank2]: Traceback (most recent call last):
|
| 351 |
+
[default7]:[rank7]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 352 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 353 |
+
[default3]:[rank3]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 354 |
+
[default4]:[rank4]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 355 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 356 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 357 |
+
[default3]:[rank3]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 358 |
+
[default7]:[rank7]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 359 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 360 |
+
[default2]:[rank2]: trainer.train(dataloader)
|
| 361 |
+
[default4]:[rank4]: outputs = self.pipeline_engine.train_batch_iter(
|
| 362 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 363 |
+
[default6]:[rank6]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 364 |
+
[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
|
| 365 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 366 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 367 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 368 |
+
[default4]:[rank4]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 369 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 370 |
+
[default2]:[rank2]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 371 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 372 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 373 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 374 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 375 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 376 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 377 |
+
[default4]:[rank4]: output = model(**micro_batch)
|
| 378 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 379 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 380 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 381 |
+
[default2]:[rank2]: outputs = self.pipeline_engine.train_batch_iter(
|
| 382 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 383 |
+
[default3]:[rank3]: output = self.pp_block(**new_kwargs)
|
| 384 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 385 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 386 |
+
[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
|
| 387 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 388 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 389 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 390 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 391 |
+
[default2]:[rank2]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 392 |
+
[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
|
| 393 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 394 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 395 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 396 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 397 |
+
[default2]:[rank2]: output = model(**micro_batch)
|
| 398 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 399 |
+
[default3]:[rank3]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 400 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 401 |
+
[default7]:[rank7]: output = self.pp_block(**new_kwargs)
|
| 402 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 403 |
+
[default6]:[rank6]: output = self.pp_block(**new_kwargs)
|
| 404 |
+
[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
|
| 405 |
+
[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
|
| 406 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 407 |
+
[default4]:[rank4]: sharded_logits = self.model(
|
| 408 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 409 |
+
[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
|
| 410 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 411 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 412 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 413 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 414 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 415 |
+
[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
|
| 416 |
+
[default3]:[rank3]: merged_states = self.gate_up_proj(hidden_states)
|
| 417 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 418 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 419 |
+
[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
|
| 420 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 421 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 422 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 423 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 424 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 425 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 426 |
+
[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
|
| 427 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 428 |
+
[default6]:[rank6]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 429 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 430 |
+
[default4]:[rank4]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 431 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 432 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 433 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 434 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 435 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 436 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 437 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 438 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 439 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 440 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 441 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 442 |
+
[default4]:[rank4]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 443 |
+
[default3]:[rank3]: return column_linear(
|
| 444 |
+
[default2]:[rank2]: sharded_logits = self.model(
|
| 445 |
+
[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
|
| 446 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 447 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 448 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 449 |
+
[default7]:[rank7]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 450 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 451 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 452 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 453 |
+
[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
|
| 454 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 455 |
+
[default3]:[rank3]: return F.linear(input, weight, bias)
|
| 456 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 457 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 458 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 459 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 460 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 461 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 462 |
+
[default2]:[rank2]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 463 |
+
[default3]:[rank3]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 1024.00 MiB. GPU has a total capacity of 79.33 GiB of which 289.94 MiB is free. Including non-PyTorch memory, this process has 79.04 GiB memory in use. Of the allocated memory 66.89 GiB is allocated by PyTorch, and 417.33 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)
|
| 464 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 465 |
+
[default6]:[rank6]: merged_states = self.gate_up_proj(hidden_states)
|
| 466 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 467 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 468 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 469 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 470 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 471 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 472 |
+
[default7]:[rank7]: merged_states = self.gate_up_proj(hidden_states)
|
| 473 |
+
[default4]:[rank4]: output = self.pp_block(**new_kwargs)
|
| 474 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 475 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 476 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 477 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 478 |
+
[default2]:[rank2]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 479 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 480 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 481 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 482 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 483 |
+
[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
|
| 484 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 485 |
+
[default6]:[rank6]: return column_linear(
|
| 486 |
+
[default7]:[rank7]: return column_linear(
|
| 487 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 488 |
+
[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
|
| 489 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 490 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 491 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 492 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 493 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 494 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 495 |
+
[default2]:[rank2]: output = self.pp_block(**new_kwargs)
|
| 496 |
+
[default7]:[rank7]: return F.linear(input, weight, bias)
|
| 497 |
+
[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
|
| 498 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 499 |
+
[default6]:[rank6]: return F.linear(input, weight, bias)
|
| 500 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 501 |
+
[default7]:[rank7]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 1024.00 MiB. GPU has a total capacity of 79.33 GiB of which 1009.94 MiB is free. Including non-PyTorch memory, this process has 78.33 GiB memory in use. Of the allocated memory 66.89 GiB is allocated by PyTorch, and 417.33 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)
|
| 502 |
+
[default4]:[rank4]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 503 |
+
[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
|
| 504 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 505 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 506 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 507 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 508 |
+
[default6]:[rank6]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 1024.00 MiB. GPU has a total capacity of 79.33 GiB of which 289.94 MiB is free. Including non-PyTorch memory, this process has 79.04 GiB memory in use. Of the allocated memory 66.89 GiB is allocated by PyTorch, and 417.33 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)
|
| 509 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 510 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 511 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 512 |
+
[default4]:[rank4]: merged_states = self.gate_up_proj(hidden_states)
|
| 513 |
+
[default2]:[rank2]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 514 |
+
[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
|
| 515 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 516 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 517 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 518 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 519 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 520 |
+
[default4]:[rank4]: return column_linear(
|
| 521 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 522 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 523 |
+
[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
|
| 524 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 525 |
+
[default4]:[rank4]: return F.linear(input, weight, bias)
|
| 526 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 527 |
+
[default2]:[rank2]: merged_states = self.gate_up_proj(hidden_states)
|
| 528 |
+
[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
|
| 529 |
+
[default4]:[rank4]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 1024.00 MiB. GPU has a total capacity of 79.33 GiB of which 289.94 MiB is free. Including non-PyTorch memory, this process has 79.04 GiB memory in use. Of the allocated memory 66.89 GiB is allocated by PyTorch, and 417.33 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)
|
| 530 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 531 |
+
[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
|
| 532 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 533 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 534 |
+
[default2]:[rank2]: return column_linear(
|
| 535 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 536 |
+
[default2]:[rank2]: return F.linear(input, weight, bias)
|
| 537 |
+
[default2]:[rank2]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 1024.00 MiB. GPU has a total capacity of 79.33 GiB of which 289.94 MiB is free. Including non-PyTorch memory, this process has 79.04 GiB memory in use. Of the allocated memory 66.89 GiB is allocated by PyTorch, and 417.33 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)
|
| 538 |
+
[default5]:[rank5]: Traceback (most recent call last):
|
| 539 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 540 |
+
[default5]:[rank5]: trainer.train(dataloader)
|
| 541 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 542 |
+
[default5]:[rank5]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 543 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 544 |
+
[default5]:[rank5]: outputs = self.pipeline_engine.train_batch_iter(
|
| 545 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 546 |
+
[default5]:[rank5]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 547 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 548 |
+
[default5]:[rank5]: output = model(**micro_batch)
|
| 549 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 550 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 551 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 552 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 553 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 554 |
+
[default5]:[rank5]: sharded_logits = self.model(
|
| 555 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 556 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 557 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 558 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 559 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 560 |
+
[default5]:[rank5]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 561 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 562 |
+
[default5]:[rank5]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 563 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 564 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 565 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 566 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 567 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 568 |
+
[default5]:[rank5]: output = self.pp_block(**new_kwargs)
|
| 569 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 570 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 571 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 572 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 573 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 574 |
+
[default5]:[rank5]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 575 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 576 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 577 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 578 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 579 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 580 |
+
[default5]:[rank5]: merged_states = self.gate_up_proj(hidden_states)
|
| 581 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
|
| 582 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 583 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 584 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 585 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 586 |
+
[default5]:[rank5]: return column_linear(
|
| 587 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 588 |
+
[default5]:[rank5]: return F.linear(input, weight, bias)
|
| 589 |
+
[default5]:[rank5]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 1024.00 MiB. GPU has a total capacity of 79.33 GiB of which 289.94 MiB is free. Including non-PyTorch memory, this process has 79.04 GiB memory in use. Of the allocated memory 66.89 GiB is allocated by PyTorch, and 417.33 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)
|
| 590 |
+
W0703 22:49:42.779000 139633926510400 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 1032681 closing signal SIGTERM
|
| 591 |
+
E0703 22:49:42.991000 139633926510400 torch/distributed/elastic/multiprocessing/api.py:826] failed (exitcode: 1) local_rank: 0 (pid: 1032676) of binary: /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/python3.10
|
| 592 |
+
Traceback (most recent call last):
|
| 593 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/torchrun", line 8, in <module>
|
| 594 |
+
sys.exit(main())
|
| 595 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 347, in wrapper
|
| 596 |
+
return f(*args, **kwargs)
|
| 597 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 879, in main
|
| 598 |
+
run(args)
|
| 599 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 870, in run
|
| 600 |
+
elastic_launch(
|
| 601 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 132, in __call__
|
| 602 |
+
return launch_agent(self._config, self._entrypoint, list(args))
|
| 603 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 263, in launch_agent
|
| 604 |
+
raise ChildFailedError(
|
| 605 |
+
torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
|
| 606 |
+
============================================================
|
| 607 |
+
/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py FAILED
|
| 608 |
+
------------------------------------------------------------
|
| 609 |
+
Failures:
|
| 610 |
+
[1]:
|
| 611 |
+
time : 2024-07-03_22:49:42
|
| 612 |
+
host : ip-26-0-161-178.ec2.internal
|
| 613 |
+
rank : 1 (local_rank: 1)
|
| 614 |
+
exitcode : 1 (pid: 1032677)
|
| 615 |
+
error_file: <N/A>
|
| 616 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 617 |
+
[2]:
|
| 618 |
+
time : 2024-07-03_22:49:42
|
| 619 |
+
host : ip-26-0-161-178.ec2.internal
|
| 620 |
+
rank : 2 (local_rank: 2)
|
| 621 |
+
exitcode : 1 (pid: 1032678)
|
| 622 |
+
error_file: <N/A>
|
| 623 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 624 |
+
[3]:
|
| 625 |
+
time : 2024-07-03_22:49:42
|
| 626 |
+
host : ip-26-0-161-178.ec2.internal
|
| 627 |
+
rank : 3 (local_rank: 3)
|
| 628 |
+
exitcode : 1 (pid: 1032679)
|
| 629 |
+
error_file: <N/A>
|
| 630 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 631 |
+
[4]:
|
| 632 |
+
time : 2024-07-03_22:49:42
|
| 633 |
+
host : ip-26-0-161-178.ec2.internal
|
| 634 |
+
rank : 4 (local_rank: 4)
|
| 635 |
+
exitcode : 1 (pid: 1032680)
|
| 636 |
+
error_file: <N/A>
|
| 637 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 638 |
+
[5]:
|
| 639 |
+
time : 2024-07-03_22:49:42
|
| 640 |
+
host : ip-26-0-161-178.ec2.internal
|
| 641 |
+
rank : 6 (local_rank: 6)
|
| 642 |
+
exitcode : 1 (pid: 1032682)
|
| 643 |
+
error_file: <N/A>
|
| 644 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 645 |
+
[6]:
|
| 646 |
+
time : 2024-07-03_22:49:42
|
| 647 |
+
host : ip-26-0-161-178.ec2.internal
|
| 648 |
+
rank : 7 (local_rank: 7)
|
| 649 |
+
exitcode : 1 (pid: 1032683)
|
| 650 |
+
error_file: <N/A>
|
| 651 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 652 |
+
------------------------------------------------------------
|
| 653 |
+
Root Cause (first observed failure):
|
| 654 |
+
[0]:
|
| 655 |
+
time : 2024-07-03_22:49:42
|
| 656 |
+
host : ip-26-0-161-178.ec2.internal
|
| 657 |
+
rank : 0 (local_rank: 0)
|
| 658 |
+
exitcode : 1 (pid: 1032676)
|
| 659 |
+
error_file: <N/A>
|
| 660 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 661 |
+
============================================================
|
| 662 |
+
srun: error: ip-26-0-161-178: task 0: Exited with exit code 1
|
| 663 |
+
Consider using `hf_transfer` for faster uploads. This solution comes with some limitations. See https://huggingface.co/docs/huggingface_hub/hf_transfer for more details.
|
llama-1B/8_GPUS/dp-8_tp-1_pp-1_mbz-16/status.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
oom
|