ColossalAI/applications/ColossalChat/benchmarks/benchmark_orpo.sh

45 lines
1.4 KiB
Bash
Executable File

#!/bin/bash
set_n_least_used_CUDA_VISIBLE_DEVICES() {
local n=${1:-"9999"}
echo "GPU Memory Usage:"
local FIRST_N_GPU_IDS=$(nvidia-smi --query-gpu=memory.used --format=csv |
tail -n +2 |
nl -v 0 |
tee /dev/tty |
sort -g -k 2 |
awk '{print $1}' |
head -n $n)
export CUDA_VISIBLE_DEVICES=$(echo $FIRST_N_GPU_IDS | sed 's/ /,/g')
echo "Now CUDA_VISIBLE_DEVICES is set to:"
echo "CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES"
}
set_n_least_used_CUDA_VISIBLE_DEVICES 2
PROJECT_NAME="dpo"
PARENT_CONFIG_FILE="./benchmark_config" # Path to a folder to save training config logs
PRETRAINED_MODEL_PATH="" # huggingface or local model path
PRETRAINED_TOKENIZER_PATH="" # huggingface or local tokenizer path
TIMESTAMP=$(date +%Y-%m-%d-%H-%M-%S)
FULL_PROJECT_NAME="${PROJECT_NAME}-${TIMESTAMP}"
CONFIG_FILE="${PARENT_CONFIG_FILE}-${FULL_PROJECT_NAME}.json"
colossalai run --nproc_per_node 2 --master_port 31313 benchmark_orpo.py \
--pretrain $PRETRAINED_MODEL_PATH \
--tokenizer_dir $PRETRAINED_TOKENIZER_PATH \
--plugin "zero2" \
--config_file $CONFIG_FILE \
--max_epochs 1 \
--accumulation_steps 1 \
--batch_size 4 \
--lr 8e-6 \
--lam 0.5 \
--mixed_precision "bf16" \
--grad_clip 1.0 \
--max_length 2048 \
--weight_decay 0.01 \
--warmup_steps 60 \
--dataset_size 160 \
--grad_checkpoint \
--use_flash_attn