Upload llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-512
Browse files
llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-512/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-1_tp-8_pp-1_mbz-512/log.out
|
| 13 |
+
#SBATCH --error=/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-512/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-1_tp-8_pp-1_mbz-512/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-1_tp-8_pp-1_mbz-512/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-1_tp-8_pp-1_mbz-512/status.txt
|
| 84 |
+
else
|
| 85 |
+
if grep -q "OutOfMemoryError" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-512/log.out; then
|
| 86 |
+
printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-512/status.txt
|
| 87 |
+
elif grep -q " CUDA error: an illegal memory access" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-512/log.out; then
|
| 88 |
+
printf "oom" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-512/status.txt
|
| 89 |
+
elif grep -q "Timeout at NCCL" /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-512/log.out; then
|
| 90 |
+
printf "timeout" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-512/status.txt
|
| 91 |
+
else
|
| 92 |
+
printf "fail" > /fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-512/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-1_tp-8_pp-1_mbz-512 --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-1_tp-8_pp-1_mbz-512 --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-1_tp-8_pp-1_mbz-512 llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-512 --commit-message "Upload llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-512"
|
| 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-1_tp-8_pp-1_mbz-512/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: 1
|
| 51 |
+
pp_engine: 1f1b
|
| 52 |
+
tp: 8
|
| 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-1_tp-8_pp-1_mbz-512
|
| 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: 2
|
| 79 |
+
limit_test_batches: 0
|
| 80 |
+
limit_val_batches: 0
|
| 81 |
+
micro_batch_size: 512
|
| 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-1_tp-8_pp-1_mbz-512/log.out
ADDED
|
@@ -0,0 +1,671 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
========================
|
| 2 |
+
START TIME: Thu Jul 4 00:05:41 UTC 2024
|
| 3 |
+
python3 version = Python 3.10.14
|
| 4 |
+
========================
|
| 5 |
+
The token has not been saved to the git credentials helper. Pass `add_to_git_credential=True` in this function directly or `--add-to-git-credential` if using via `huggingface-cli` if you want to set the git credential as well.
|
| 6 |
+
Token is valid (permission: write).
|
| 7 |
+
Your token has been saved to /admin/home/ferdinand_mom/.cache/huggingface/token
|
| 8 |
+
Login successful
|
| 9 |
+
fatal: Unable to create '/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/.git/index.lock': File exists.
|
| 10 |
+
|
| 11 |
+
Another git process seems to be running in this repository, e.g.
|
| 12 |
+
an editor opened by 'git commit'. Please make sure all processes
|
| 13 |
+
are terminated then try again. If it still fails, a git process
|
| 14 |
+
may have crashed in this repository earlier:
|
| 15 |
+
remove the file manually to continue.
|
| 16 |
+
Job status: RUNNING
|
| 17 |
+
W0704 00:05:44.480000 140468343650112 torch/distributed/run.py:757]
|
| 18 |
+
W0704 00:05:44.480000 140468343650112 torch/distributed/run.py:757] *****************************************
|
| 19 |
+
W0704 00:05:44.480000 140468343650112 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 |
+
W0704 00:05:44.480000 140468343650112 torch/distributed/run.py:757] *****************************************
|
| 21 |
+
[default0]:07/04/2024 00:06:03 [WARNING|DP=0|PP=0|TP=0|ip-26-0-164-187]: [Vocab Size Padding] Padded vocab (size: 50257) with 7 dummy tokens (new size: 50264)
|
| 22 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Config:
|
| 23 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Config(general=GeneralArgs(project='bench_cluster',
|
| 24 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: run='%date_%jobid',
|
| 25 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: seed=42,
|
| 26 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: step=None,
|
| 27 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: consumed_train_samples=None,
|
| 28 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: benchmark_csv_path=None,
|
| 29 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: ignore_sanity_checks=True),
|
| 30 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: parallelism=ParallelismArgs(dp=1,
|
| 31 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: pp=1,
|
| 32 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tp=8,
|
| 33 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: pp_engine=<nanotron.parallel.pipeline_parallel.engine.OneForwardOneBackwardPipelineEngine object at 0x7f672bd00880>,
|
| 34 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tp_mode=<TensorParallelLinearMode.REDUCE_SCATTER: 2>,
|
| 35 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tp_linear_async_communication=False,
|
| 36 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: expert_parallel_size=1),
|
| 37 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: model=ModelArgs(model_config=LlamaConfig(bos_token_id=1,
|
| 38 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: eos_token_id=2,
|
| 39 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: hidden_act='silu',
|
| 40 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: hidden_size=2048,
|
| 41 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: initializer_range=0.02,
|
| 42 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: intermediate_size=4096,
|
| 43 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: is_llama_config=True,
|
| 44 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: max_position_embeddings=4096,
|
| 45 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: num_attention_heads=32,
|
| 46 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: num_hidden_layers=24,
|
| 47 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: num_key_value_heads=32,
|
| 48 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: pad_token_id=None,
|
| 49 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: pretraining_tp=1,
|
| 50 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: rms_norm_eps=1e-05,
|
| 51 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: rope_scaling=None,
|
| 52 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: rope_theta=10000.0,
|
| 53 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tie_word_embeddings=True,
|
| 54 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: use_cache=True,
|
| 55 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: vocab_size=50264),
|
| 56 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: init_method=RandomInit(std=0.025),
|
| 57 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: dtype=torch.bfloat16,
|
| 58 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: make_vocab_size_divisible_by=1,
|
| 59 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: ddp_bucket_cap_mb=25),
|
| 60 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tokenizer=TokenizerArgs(tokenizer_name_or_path='openai-community/gpt2',
|
| 61 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tokenizer_revision=None,
|
| 62 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tokenizer_max_length=None),
|
| 63 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: checkpoints=CheckpointsArgs(checkpoints_path=Path('/dev/null'),
|
| 64 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: checkpoint_interval=100000,
|
| 65 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: save_initial_state=False,
|
| 66 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: resume_checkpoint_path=None,
|
| 67 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: checkpoints_path_is_shared_file_system=False),
|
| 68 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: logging=LoggingArgs(log_level='info',
|
| 69 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: log_level_replica='info',
|
| 70 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: iteration_step_info_interval=1),
|
| 71 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tokens=TokensArgs(sequence_length=4096,
|
| 72 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: train_steps=20,
|
| 73 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: micro_batch_size=512,
|
| 74 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: batch_accumulation_per_replica=2,
|
| 75 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: val_check_interval=-1,
|
| 76 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: limit_val_batches=0,
|
| 77 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: limit_test_batches=0),
|
| 78 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: optimizer=OptimizerArgs(optimizer_factory=AdamWOptimizerArgs(adam_eps=1e-08,
|
| 79 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: adam_beta1=0.9,
|
| 80 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: adam_beta2=0.95,
|
| 81 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: torch_adam_is_fused=True,
|
| 82 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: name='adamW'),
|
| 83 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: zero_stage=1,
|
| 84 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: weight_decay=0.01,
|
| 85 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: clip_grad=1.0,
|
| 86 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: accumulate_grad_in_fp32=True,
|
| 87 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: learning_rate_scheduler=LRSchedulerArgs(learning_rate=0.0001,
|
| 88 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: lr_warmup_steps=1,
|
| 89 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: lr_warmup_style='linear',
|
| 90 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: lr_decay_style='linear',
|
| 91 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: lr_decay_steps=19,
|
| 92 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: lr_decay_starting_step=None,
|
| 93 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: min_decay_lr=1e-05)),
|
| 94 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: data_stages=[DatasetStageArgs(name='Training Stage',
|
| 95 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: start_training_step=1,
|
| 96 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: data=DataArgs(dataset=PretrainDatasetsArgs(hf_dataset_or_datasets='roneneldan/TinyStories',
|
| 97 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: hf_dataset_splits='train',
|
| 98 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: hf_dataset_config_name=None,
|
| 99 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: dataset_processing_num_proc_per_process=64,
|
| 100 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: dataset_overwrite_cache=False,
|
| 101 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: text_column_name='text'),
|
| 102 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: seed=42,
|
| 103 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: num_loading_workers=0))],
|
| 104 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: profiler=ProfilerArgs(profiler_export_path=Path('/fsx/ferdinandmom/ferdinand-hf/bench_cluster/results/llama-1B/8_GPUS/dp-1_tp-8_pp-1_mbz-512')),
|
| 105 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: lighteval=None)
|
| 106 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Model Config:
|
| 107 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: LlamaConfig(bos_token_id=1,
|
| 108 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: eos_token_id=2,
|
| 109 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: hidden_act='silu',
|
| 110 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: hidden_size=2048,
|
| 111 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: initializer_range=0.02,
|
| 112 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: intermediate_size=4096,
|
| 113 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: is_llama_config=True,
|
| 114 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: max_position_embeddings=4096,
|
| 115 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: num_attention_heads=32,
|
| 116 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: num_hidden_layers=24,
|
| 117 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: num_key_value_heads=32,
|
| 118 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: pad_token_id=None,
|
| 119 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: pretraining_tp=1,
|
| 120 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: rms_norm_eps=1e-05,
|
| 121 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: rope_scaling=None,
|
| 122 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: rope_theta=10000.0,
|
| 123 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: tie_word_embeddings=True,
|
| 124 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: use_cache=True,
|
| 125 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: vocab_size=50264)
|
| 126 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Building model..
|
| 127 |
+
[default0]:07/04/2024 00:06:03 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Setting PP block ranks...
|
| 128 |
+
[default6]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=6|ip-26-0-164-187]: Local number of parameters: 139M (264.73MiB)
|
| 129 |
+
[default6]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=6|ip-26-0-164-187]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
|
| 130 |
+
[default6]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=6|ip-26-0-164-187]: No checkpoint path provided.
|
| 131 |
+
[default5]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=5|ip-26-0-164-187]: Local number of parameters: 139M (264.73MiB)
|
| 132 |
+
[default5]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=5|ip-26-0-164-187]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
|
| 133 |
+
[default5]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=5|ip-26-0-164-187]: No checkpoint path provided.
|
| 134 |
+
[default0]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Total number of parameters: 1.11G (2117.88MiB)
|
| 135 |
+
[default0]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Local number of parameters: 139M (264.73MiB)
|
| 136 |
+
[default0]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
|
| 137 |
+
[default0]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: No checkpoint path provided.
|
| 138 |
+
[default0]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Parametrizing model parameters using StandardParametrizator
|
| 139 |
+
[default0]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [Optimizer Building] Using LearningRateForSP as learning rate
|
| 140 |
+
[default0]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [ZeRO sharding] Size of optimizer params per rank:
|
| 141 |
+
[default0]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [ZeRO sharding] DP Rank 0 has 139M out of 139M (100.00%) params' optimizer states
|
| 142 |
+
[default1]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=1|ip-26-0-164-187]: Local number of parameters: 139M (264.73MiB)
|
| 143 |
+
[default1]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=1|ip-26-0-164-187]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
|
| 144 |
+
[default1]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=1|ip-26-0-164-187]: No checkpoint path provided.
|
| 145 |
+
[default7]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=7|ip-26-0-164-187]: Local number of parameters: 139M (264.73MiB)
|
| 146 |
+
[default7]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=7|ip-26-0-164-187]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
|
| 147 |
+
[default7]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=7|ip-26-0-164-187]: No checkpoint path provided.
|
| 148 |
+
[default3]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=3|ip-26-0-164-187]: Local number of parameters: 139M (264.73MiB)
|
| 149 |
+
[default3]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=3|ip-26-0-164-187]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
|
| 150 |
+
[default3]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=3|ip-26-0-164-187]: No checkpoint path provided.
|
| 151 |
+
[default2]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=2|ip-26-0-164-187]: Local number of parameters: 139M (264.73MiB)
|
| 152 |
+
[default2]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=2|ip-26-0-164-187]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
|
| 153 |
+
[default2]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=2|ip-26-0-164-187]: No checkpoint path provided.
|
| 154 |
+
[default4]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=4|ip-26-0-164-187]: Local number of parameters: 139M (264.73MiB)
|
| 155 |
+
[default4]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=4|ip-26-0-164-187]: [After model building] Memory usage: 290.76MiB. Peak allocated: 317.33MiB Peak reserved: 324.00MiB
|
| 156 |
+
[default4]:07/04/2024 00:06:18 [INFO|DP=0|PP=0|TP=4|ip-26-0-164-187]: No checkpoint path provided.
|
| 157 |
+
[default0]:07/04/2024 00:06:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [Training Plan] Stage Training Stage has 19 remaining training steps and has consumed 0 samples
|
| 158 |
+
[default0]:07/04/2024 00:06:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Using `datasets` library
|
| 159 |
+
[default0]:07/04/2024 00:06:20 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Loading tokenizer from openai-community/gpt2 and transformers/hf_hub versions ('4.41.2', '0.23.4')
|
| 160 |
+
[default0]:07/04/2024 00:06:20 [WARNING|DP=0|PP=0|TP=0|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 161 |
+
[default0]:Repo card metadata block was not found. Setting CardData to empty.
|
| 162 |
+
[default0]:07/04/2024 00:06:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [Training Plan] There are 1 training stages
|
| 163 |
+
[default0]:07/04/2024 00:06:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [Stage Training Stage] start from step 1
|
| 164 |
+
[default0]:07/04/2024 00:06:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]:
|
| 165 |
+
[default0]:07/04/2024 00:06:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: [Start training] datetime: 2024-07-04 00:06:22.734159 | mbs: 512 | grad_accum: 2 | global_batch_size: 1024 | sequence_length: 4096 | train_steps: 20 | start_iteration_step: 0 | consumed_train_samples: 0
|
| 166 |
+
[default0]:07/04/2024 00:06:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Resuming training from stage Training Stage, it has trained for 0 samples and has 19 remaining train steps
|
| 167 |
+
[default0]:07/04/2024 00:06:22 [INFO|DP=0|PP=0|TP=0|ip-26-0-164-187]: Memory usage: 1350.75MiB. Peak allocated 1350.76MiB. Peak reserved: 1384.00MiB
|
| 168 |
+
[default4]:Repo card metadata block was not found. Setting CardData to empty.
|
| 169 |
+
[default2]:07/04/2024 00:06:22 [WARNING|DP=0|PP=0|TP=2|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 170 |
+
[default4]:07/04/2024 00:06:22 [WARNING|DP=0|PP=0|TP=4|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 171 |
+
[default2]:Repo card metadata block was not found. Setting CardData to empty.
|
| 172 |
+
[default7]:07/04/2024 00:06:23 [WARNING|DP=0|PP=0|TP=7|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 173 |
+
[default7]:Repo card metadata block was not found. Setting CardData to empty.
|
| 174 |
+
[default6]:07/04/2024 00:06:23 [WARNING|DP=0|PP=0|TP=6|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 175 |
+
[default1]:07/04/2024 00:06:23 [WARNING|DP=0|PP=0|TP=1|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 176 |
+
[default1]:Repo card metadata block was not found. Setting CardData to empty.
|
| 177 |
+
[default6]:Repo card metadata block was not found. Setting CardData to empty.
|
| 178 |
+
[default5]:07/04/2024 00:06:23 [WARNING|DP=0|PP=0|TP=5|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 179 |
+
[default5]:Repo card metadata block was not found. Setting CardData to empty.
|
| 180 |
+
[default3]:Repo card metadata block was not found. Setting CardData to empty.
|
| 181 |
+
[default3]:07/04/2024 00:06:23 [WARNING|DP=0|PP=0|TP=3|ip-26-0-164-187]: Repo card metadata block was not found. Setting CardData to empty.
|
| 182 |
+
[default4]:[rank4]: Traceback (most recent call last):
|
| 183 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 184 |
+
[default4]:[rank4]: trainer.train(dataloader)
|
| 185 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 186 |
+
[default4]:[rank4]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 187 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 188 |
+
[default4]:[rank4]: outputs = self.pipeline_engine.train_batch_iter(
|
| 189 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 190 |
+
[default4]:[rank4]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 191 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 192 |
+
[default4]:[rank4]: output = model(**micro_batch)
|
| 193 |
+
[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
|
| 194 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 195 |
+
[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
|
| 196 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 197 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 198 |
+
[default4]:[rank4]: sharded_logits = self.model(
|
| 199 |
+
[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
|
| 200 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 201 |
+
[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
|
| 202 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 203 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 204 |
+
[default4]:[rank4]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 205 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 206 |
+
[default4]:[rank4]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 207 |
+
[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
|
| 208 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 209 |
+
[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
|
| 210 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 211 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 212 |
+
[default4]:[rank4]: output = self.pp_block(**new_kwargs)
|
| 213 |
+
[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
|
| 214 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 215 |
+
[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
|
| 216 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 217 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 218 |
+
[default4]:[rank4]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 219 |
+
[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
|
| 220 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 221 |
+
[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
|
| 222 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 223 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 224 |
+
[default4]:[rank4]: merged_states = self.gate_up_proj(hidden_states)
|
| 225 |
+
[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
|
| 226 |
+
[default4]:[rank4]: return self._call_impl(*args, **kwargs)
|
| 227 |
+
[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
|
| 228 |
+
[default4]:[rank4]: return forward_call(*args, **kwargs)
|
| 229 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 230 |
+
[default4]:[rank4]: return column_linear(
|
| 231 |
+
[default4]:[rank4]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 232 |
+
[default4]:[rank4]: return F.linear(input, weight, bias)
|
| 233 |
+
[default4]:[rank4]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 4.00 GiB. GPU has a total capacity of 79.33 GiB of which 3.18 GiB is free. Including non-PyTorch memory, this process has 76.13 GiB memory in use. Of the allocated memory 61.47 GiB is allocated by PyTorch, and 2.93 GiB 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)
|
| 234 |
+
[default1]:[rank1]: Traceback (most recent call last):
|
| 235 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 236 |
+
[default1]:[rank1]: trainer.train(dataloader)
|
| 237 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 238 |
+
[default1]:[rank1]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 239 |
+
[default3]:[rank3]: Traceback (most recent call last):
|
| 240 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 241 |
+
[default3]:[rank3]: trainer.train(dataloader)
|
| 242 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 243 |
+
[default1]:[rank1]: outputs = self.pipeline_engine.train_batch_iter(
|
| 244 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 245 |
+
[default3]:[rank3]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 246 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 247 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 248 |
+
[default1]:[rank1]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 249 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 250 |
+
[default1]:[rank1]: output = model(**micro_batch)
|
| 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/models/llama.py", line 891, in forward
|
| 256 |
+
[default1]:[rank1]: sharded_logits = self.model(
|
| 257 |
+
[default3]:[rank3]: outputs = self.pipeline_engine.train_batch_iter(
|
| 258 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 259 |
+
[default3]:[rank3]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 260 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 261 |
+
[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
|
| 262 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 263 |
+
[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
|
| 264 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 265 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 266 |
+
[default1]:[rank1]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 267 |
+
[default3]:[rank3]: output = model(**micro_batch)
|
| 268 |
+
[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
|
| 269 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 270 |
+
[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
|
| 271 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 272 |
+
[default1]:[rank1]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 273 |
+
[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
|
| 274 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 275 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 276 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 277 |
+
[default3]:[rank3]: sharded_logits = self.model(
|
| 278 |
+
[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
|
| 279 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 280 |
+
[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
|
| 281 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 282 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 283 |
+
[default3]:[rank3]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 284 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 285 |
+
[default3]:[rank3]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 286 |
+
[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
|
| 287 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 288 |
+
[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
|
| 289 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 290 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 291 |
+
[default3]:[rank3]: output = self.pp_block(**new_kwargs)
|
| 292 |
+
[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
|
| 293 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 294 |
+
[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
|
| 295 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 296 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 297 |
+
[default3]:[rank3]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 298 |
+
[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
|
| 299 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 300 |
+
[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
|
| 301 |
+
[default0]:[rank0]: Traceback (most recent call last):
|
| 302 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 303 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 304 |
+
[default3]:[rank3]: merged_states = self.gate_up_proj(hidden_states)
|
| 305 |
+
[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
|
| 306 |
+
[default7]:[rank7]: Traceback (most recent call last):
|
| 307 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 308 |
+
[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
|
| 309 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 310 |
+
[default3]:[rank3]: return self._call_impl(*args, **kwargs)
|
| 311 |
+
[default7]:[rank7]: trainer.train(dataloader)
|
| 312 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 313 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 314 |
+
[default7]:[rank7]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 315 |
+
[default1]:[rank1]: output = self.pp_block(**new_kwargs)
|
| 316 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 317 |
+
[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
|
| 318 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 319 |
+
[default0]:[rank0]: trainer.train(dataloader)
|
| 320 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 321 |
+
[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
|
| 322 |
+
[default7]:[rank7]: outputs = self.pipeline_engine.train_batch_iter(
|
| 323 |
+
[default0]:[rank0]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 324 |
+
[default3]:[rank3]: return forward_call(*args, **kwargs)
|
| 325 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 326 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 327 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 328 |
+
[default0]:[rank0]: outputs = self.pipeline_engine.train_batch_iter(
|
| 329 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 330 |
+
[default3]:[rank3]: return column_linear(
|
| 331 |
+
[default3]:[rank3]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 332 |
+
[default7]:[rank7]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 333 |
+
[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
|
| 334 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 335 |
+
[default0]:[rank0]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 336 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 337 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 338 |
+
[default0]:[rank0]: output = model(**micro_batch)
|
| 339 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 340 |
+
[default7]:[rank7]: output = model(**micro_batch)
|
| 341 |
+
[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
|
| 342 |
+
[default3]:[rank3]: return F.linear(input, weight, bias)
|
| 343 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 344 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 345 |
+
[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
|
| 346 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 347 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 348 |
+
[default1]:[rank1]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 349 |
+
[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
|
| 350 |
+
[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
|
| 351 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 352 |
+
[default7]:[rank7]: sharded_logits = self.model(
|
| 353 |
+
[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
|
| 354 |
+
[default3]:[rank3]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 4.00 GiB. GPU has a total capacity of 79.33 GiB of which 3.18 GiB is free. Including non-PyTorch memory, this process has 76.13 GiB memory in use. Of the allocated memory 61.47 GiB is allocated by PyTorch, and 2.93 GiB 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)
|
| 355 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 356 |
+
[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
|
| 357 |
+
[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
|
| 358 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 359 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 360 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 361 |
+
[default0]:[rank0]: sharded_logits = self.model(
|
| 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 171, in forward
|
| 364 |
+
[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
|
| 365 |
+
[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
|
| 366 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 367 |
+
[default1]:[rank1]: merged_states = self.gate_up_proj(hidden_states)
|
| 368 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 369 |
+
[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
|
| 370 |
+
[default1]:[rank1]: return self._call_impl(*args, **kwargs)
|
| 371 |
+
[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
|
| 372 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 373 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 374 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 375 |
+
[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
|
| 376 |
+
[default1]:[rank1]: return forward_call(*args, **kwargs)
|
| 377 |
+
[default0]:[rank0]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 378 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 379 |
+
[default7]:[rank7]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 380 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 381 |
+
[default0]:[rank0]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 382 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 383 |
+
[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
|
| 384 |
+
[default1]:[rank1]: return column_linear(
|
| 385 |
+
[default1]:[rank1]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 386 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 387 |
+
[default7]:[rank7]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 388 |
+
[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
|
| 389 |
+
[default1]:[rank1]: return F.linear(input, weight, bias)
|
| 390 |
+
[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
|
| 391 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 392 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 393 |
+
[default1]:[rank1]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 4.00 GiB. GPU has a total capacity of 79.33 GiB of which 3.18 GiB is free. Including non-PyTorch memory, this process has 76.13 GiB memory in use. Of the allocated memory 61.47 GiB is allocated by PyTorch, and 2.93 GiB 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)
|
| 394 |
+
[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
|
| 395 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 396 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 397 |
+
[default0]:[rank0]: output = self.pp_block(**new_kwargs)
|
| 398 |
+
[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
|
| 399 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 400 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 401 |
+
[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
|
| 402 |
+
[default7]:[rank7]: output = self.pp_block(**new_kwargs)
|
| 403 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 404 |
+
[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
|
| 405 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 406 |
+
[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
|
| 407 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 408 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 409 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 410 |
+
[default7]:[rank7]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 411 |
+
[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
|
| 412 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 413 |
+
[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
|
| 414 |
+
[default0]:[rank0]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 415 |
+
[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
|
| 416 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 417 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 418 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 419 |
+
[default7]:[rank7]: merged_states = self.gate_up_proj(hidden_states)
|
| 420 |
+
[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
|
| 421 |
+
[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
|
| 422 |
+
[default7]:[rank7]: return self._call_impl(*args, **kwargs)
|
| 423 |
+
[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
|
| 424 |
+
[default7]:[rank7]: return forward_call(*args, **kwargs)
|
| 425 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 426 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 427 |
+
[default7]:[rank7]: return column_linear(
|
| 428 |
+
[default7]:[rank7]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 429 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 430 |
+
[default0]:[rank0]: merged_states = self.gate_up_proj(hidden_states)
|
| 431 |
+
[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
|
| 432 |
+
[default0]:[rank0]: return self._call_impl(*args, **kwargs)
|
| 433 |
+
[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
|
| 434 |
+
[default7]:[rank7]: return F.linear(input, weight, bias)
|
| 435 |
+
[default7]:[rank7]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 4.00 GiB. GPU has a total capacity of 79.33 GiB of which 3.89 GiB is free. Including non-PyTorch memory, this process has 75.43 GiB memory in use. Of the allocated memory 61.47 GiB is allocated by PyTorch, and 2.93 GiB 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)
|
| 436 |
+
[default0]:[rank0]: return forward_call(*args, **kwargs)
|
| 437 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 438 |
+
[default0]:[rank0]: return column_linear(
|
| 439 |
+
[default0]:[rank0]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 440 |
+
[default0]:[rank0]: return F.linear(input, weight, bias)
|
| 441 |
+
[default0]:[rank0]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 4.00 GiB. GPU
|
| 442 |
+
[default6]:[rank6]: Traceback (most recent call last):
|
| 443 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 444 |
+
[default5]:[rank5]: Traceback (most recent call last):
|
| 445 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 446 |
+
[default6]:[rank6]: trainer.train(dataloader)
|
| 447 |
+
[default2]:[rank2]: Traceback (most recent call last):
|
| 448 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py", line 237, in <module>
|
| 449 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 450 |
+
[default5]:[rank5]: trainer.train(dataloader)
|
| 451 |
+
[default6]:[rank6]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 452 |
+
[default2]:[rank2]: trainer.train(dataloader)
|
| 453 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 454 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 455 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 429, in train
|
| 456 |
+
[default5]:[rank5]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 457 |
+
[default2]:[rank2]: outputs, loss_avg = self.training_step(dataloader=self.current_dataloader)
|
| 458 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 459 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/trainer.py", line 462, in training_step
|
| 460 |
+
[default2]:[rank2]: outputs = self.pipeline_engine.train_batch_iter(
|
| 461 |
+
[default6]:[rank6]: outputs = self.pipeline_engine.train_batch_iter(
|
| 462 |
+
[default5]:[rank5]: outputs = self.pipeline_engine.train_batch_iter(
|
| 463 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 464 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 465 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 278, in train_batch_iter
|
| 466 |
+
[default5]:[rank5]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 467 |
+
[default6]:[rank6]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 468 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 469 |
+
[default2]:[rank2]: output = self.forward(context=context, state=state, micro_batch=micro_batch, model=model)
|
| 470 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 471 |
+
[default5]:[rank5]: output = model(**micro_batch)
|
| 472 |
+
[default6]:[rank6]: output = model(**micro_batch)
|
| 473 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/engine.py", line 44, in forward
|
| 474 |
+
[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
|
| 475 |
+
[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
|
| 476 |
+
[default2]:[rank2]: output = model(**micro_batch)
|
| 477 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 478 |
+
[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
|
| 479 |
+
[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
|
| 480 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 481 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 482 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 483 |
+
[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
|
| 484 |
+
[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
|
| 485 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 486 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 487 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 488 |
+
[default6]:[rank6]: sharded_logits = self.model(
|
| 489 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 490 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 891, in forward
|
| 491 |
+
[default5]:[rank5]: sharded_logits = self.model(
|
| 492 |
+
[default2]:[rank2]: sharded_logits = self.model(
|
| 493 |
+
[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
|
| 494 |
+
[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
|
| 495 |
+
[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
|
| 496 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 497 |
+
[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
|
| 498 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 499 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 500 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 501 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
|
| 502 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 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 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 505 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 506 |
+
[default5]:[rank5]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 507 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 508 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 509 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 764, in forward
|
| 510 |
+
[default2]:[rank2]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 511 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 512 |
+
[default5]:[rank5]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 513 |
+
[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
|
| 514 |
+
[default6]:[rank6]: return self.forward_with_hidden_states(input_ids=input_ids, input_mask=input_mask)[0]
|
| 515 |
+
[default2]:[rank2]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 516 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 517 |
+
[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
|
| 518 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 780, in forward_with_hidden_states
|
| 519 |
+
[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
|
| 520 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 521 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 522 |
+
[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
|
| 523 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 524 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 525 |
+
[default5]:[rank5]: output = self.pp_block(**new_kwargs)
|
| 526 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 527 |
+
[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
|
| 528 |
+
[default6]:[rank6]: hidden_encoder_states = encoder_block(**hidden_encoder_states)
|
| 529 |
+
[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
|
| 530 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 531 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 532 |
+
[default2]:[rank2]: output = self.pp_block(**new_kwargs)
|
| 533 |
+
[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
|
| 534 |
+
[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
|
| 535 |
+
[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
|
| 536 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 537 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 538 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/pipeline_parallel/block.py", line 151, in forward
|
| 539 |
+
[default6]:[rank6]: output = self.pp_block(**new_kwargs)
|
| 540 |
+
[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
|
| 541 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 542 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 543 |
+
[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
|
| 544 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 545 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 546 |
+
[default2]:[rank2]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 547 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 548 |
+
[default5]:[rank5]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 549 |
+
[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
|
| 550 |
+
[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
|
| 551 |
+
[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
|
| 552 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 553 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 554 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 555 |
+
[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
|
| 556 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 637, in forward
|
| 557 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 558 |
+
[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
|
| 559 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 560 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 561 |
+
[default6]:[rank6]: hidden_states = self.mlp(hidden_states=hidden_states)["hidden_states"]
|
| 562 |
+
[default2]:[rank2]: merged_states = self.gate_up_proj(hidden_states)
|
| 563 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 564 |
+
[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
|
| 565 |
+
[default5]:[rank5]: merged_states = self.gate_up_proj(hidden_states)
|
| 566 |
+
[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
|
| 567 |
+
[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
|
| 568 |
+
[default2]:[rank2]: return self._call_impl(*args, **kwargs)
|
| 569 |
+
[default5]:[rank5]: return self._call_impl(*args, **kwargs)
|
| 570 |
+
[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
|
| 571 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 572 |
+
[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
|
| 573 |
+
[default5]:[rank5]: return forward_call(*args, **kwargs)
|
| 574 |
+
[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
|
| 575 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 576 |
+
[default5]:[rank5]: return column_linear(
|
| 577 |
+
[default2]:[rank2]: return forward_call(*args, **kwargs)
|
| 578 |
+
[default5]:[rank5]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 579 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 580 |
+
[default5]:[rank5]: return F.linear(input, weight, bias)
|
| 581 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 582 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/models/llama.py", line 171, in forward
|
| 583 |
+
[default5]:[rank5]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 4.00 GiB. GPU has a total capacity of 79.33 GiB of which 3.18 GiB is free. Including non-PyTorch memory, this process has 76.13 GiB memory in use. Of the allocated memory 61.47 GiB is allocated by PyTorch, and 2.93 GiB 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)
|
| 584 |
+
[default6]:[rank6]: merged_states = self.gate_up_proj(hidden_states)
|
| 585 |
+
[default2]:[rank2]: return column_linear(
|
| 586 |
+
[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
|
| 587 |
+
[default2]:[rank2]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 588 |
+
[default2]:[rank2]: return F.linear(input, weight, bias)
|
| 589 |
+
[default6]:[rank6]: return self._call_impl(*args, **kwargs)
|
| 590 |
+
[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
|
| 591 |
+
[default2]:[rank2]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 4.00 GiB. GPU has a total capacity of 79.33 GiB of which 3.18 GiB is free. Including non-PyTorch memory, this process has 76.13 GiB memory in use. Of the allocated memory 61.47 GiB is allocated by PyTorch, and 2.93 GiB 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)
|
| 592 |
+
[default6]:[rank6]: return forward_call(*args, **kwargs)
|
| 593 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/nn.py", line 87, in forward
|
| 594 |
+
[default6]:[rank6]: return column_linear(
|
| 595 |
+
[default6]:[rank6]: File "/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/src/nanotron/parallel/tensor_parallel/functional.py", line 359, in column_linear
|
| 596 |
+
[default6]:[rank6]: return F.linear(input, weight, bias)
|
| 597 |
+
[default6]:[rank6]: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 4.00 GiB. GPU has a total capacity of 79.33 GiB of which 3.18 GiB is free. Including non-PyTorch memory, this process has 76.13 GiB memory in use. Of the allocated memory 61.47 GiB is allocated by PyTorch, and 2.93 GiB 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)
|
| 598 |
+
W0704 00:06:34.768000 140468343650112 torch/distributed/elastic/multiprocessing/api.py:851] Sending process 38624 closing signal SIGTERM
|
| 599 |
+
E0704 00:06:35.390000 140468343650112 torch/distributed/elastic/multiprocessing/api.py:826] failed (exitcode: 1) local_rank: 0 (pid: 38618) of binary: /fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/python3.10
|
| 600 |
+
Traceback (most recent call last):
|
| 601 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/bin/torchrun", line 8, in <module>
|
| 602 |
+
sys.exit(main())
|
| 603 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 347, in wrapper
|
| 604 |
+
return f(*args, **kwargs)
|
| 605 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 879, in main
|
| 606 |
+
run(args)
|
| 607 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/run.py", line 870, in run
|
| 608 |
+
elastic_launch(
|
| 609 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 132, in __call__
|
| 610 |
+
return launch_agent(self._config, self._entrypoint, list(args))
|
| 611 |
+
File "/fsx/ferdinandmom/miniforge3/envs/env-bench-cluster/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 263, in launch_agent
|
| 612 |
+
raise ChildFailedError(
|
| 613 |
+
torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
|
| 614 |
+
============================================================
|
| 615 |
+
/fsx/ferdinandmom/ferdinand-hf/bench_cluster/nanotron/run_train.py FAILED
|
| 616 |
+
------------------------------------------------------------
|
| 617 |
+
Failures:
|
| 618 |
+
[1]:
|
| 619 |
+
time : 2024-07-04_00:06:34
|
| 620 |
+
host : ip-26-0-164-187.ec2.internal
|
| 621 |
+
rank : 1 (local_rank: 1)
|
| 622 |
+
exitcode : 1 (pid: 38619)
|
| 623 |
+
error_file: <N/A>
|
| 624 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 625 |
+
[2]:
|
| 626 |
+
time : 2024-07-04_00:06:34
|
| 627 |
+
host : ip-26-0-164-187.ec2.internal
|
| 628 |
+
rank : 2 (local_rank: 2)
|
| 629 |
+
exitcode : 1 (pid: 38620)
|
| 630 |
+
error_file: <N/A>
|
| 631 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 632 |
+
[3]:
|
| 633 |
+
time : 2024-07-04_00:06:34
|
| 634 |
+
host : ip-26-0-164-187.ec2.internal
|
| 635 |
+
rank : 3 (local_rank: 3)
|
| 636 |
+
exitcode : 1 (pid: 38621)
|
| 637 |
+
error_file: <N/A>
|
| 638 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 639 |
+
[4]:
|
| 640 |
+
time : 2024-07-04_00:06:34
|
| 641 |
+
host : ip-26-0-164-187.ec2.internal
|
| 642 |
+
rank : 4 (local_rank: 4)
|
| 643 |
+
exitcode : 1 (pid: 38622)
|
| 644 |
+
error_file: <N/A>
|
| 645 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 646 |
+
[5]:
|
| 647 |
+
time : 2024-07-04_00:06:34
|
| 648 |
+
host : ip-26-0-164-187.ec2.internal
|
| 649 |
+
rank : 5 (local_rank: 5)
|
| 650 |
+
exitcode : 1 (pid: 38623)
|
| 651 |
+
error_file: <N/A>
|
| 652 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 653 |
+
[6]:
|
| 654 |
+
time : 2024-07-04_00:06:34
|
| 655 |
+
host : ip-26-0-164-187.ec2.internal
|
| 656 |
+
rank : 7 (local_rank: 7)
|
| 657 |
+
exitcode : 1 (pid: 38625)
|
| 658 |
+
error_file: <N/A>
|
| 659 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 660 |
+
------------------------------------------------------------
|
| 661 |
+
Root Cause (first observed failure):
|
| 662 |
+
[0]:
|
| 663 |
+
time : 2024-07-04_00:06:34
|
| 664 |
+
host : ip-26-0-164-187.ec2.internal
|
| 665 |
+
rank : 0 (local_rank: 0)
|
| 666 |
+
exitcode : 1 (pid: 38618)
|
| 667 |
+
error_file: <N/A>
|
| 668 |
+
traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
|
| 669 |
+
============================================================
|
| 670 |
+
srun: error: ip-26-0-164-187: task 0: Exited with exit code 1
|
| 671 |
+
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-1_tp-8_pp-1_mbz-512/status.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
oom
|