wehub-resource-sync
18 lines
464 B
Bash
18 lines
464 B
Bash
cp=$1
|
|
cn=$2
|
|
exp_name=$3
|
|
n_gpus=$4
|
|
split_id_offset=$5
|
|
|
|
for ((i=6;i<=$#;i++)); do
|
|
step=${!i}
|
|
echo $step
|
|
for j in {0..$(($n_gpus-1))}; do
|
|
TOKENIZERS_PARALLELISM=false CUDA_VISIBLE_DEVICES=$j python vllm_inference.py eval_sub_path=checkpoint-$step output_path_prefix=/mnt/fangkai_blob/reward_modeling/ fp16_bfloat16=False split_id=$(($j+$split_id_offset)) exp_name=$exp_name -cp $cp -cn $cn &
|
|
done
|
|
wait
|
|
echo "Step-$step Finished"
|
|
done
|
|
|
|
echo "Finished"
|