mirror of https://github.com/hpcaitech/ColossalAI
26 lines
718 B
Bash
Executable File
26 lines
718 B
Bash
Executable File
device_number=number of your devices
|
|
model_name="name of your model"
|
|
model_path="path to your model"
|
|
dataset="path to the question dataset"
|
|
answer_path="path to save the model answers"
|
|
|
|
torchrun --standalone --nproc_per_node=$device_number generate_answers.py \
|
|
--model 'llama' \
|
|
--strategy ddp \
|
|
--model_path $model_path \
|
|
--model_name $model_name \
|
|
--dataset $dataset \
|
|
--batch_size 8 \
|
|
--max_datasets_size 80 \
|
|
--answer_path $answer_path \
|
|
--max_length 512
|
|
|
|
python merge.py \
|
|
--model_name $model_name \
|
|
--shards $device_number \
|
|
--answer_path $answer_path \
|
|
|
|
for (( i=0; i<device_number; i++ )) do
|
|
rm -rf "${answer_path}/${model_name}_answers_rank${i}.json"
|
|
done
|