File size: 1,886 Bytes
d73500e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 | #!/usr/bin/bash
port="21804"
GPUs="0,1,2,3,4,5,6,7"
# Taking mistralai/Mistral-7B-v0.1 as an example.
model_names=("mistral") # The model to be compressed.
drop_modules=("mlp" "attn" "block") # The modules to be dropped.
drop_nums=("4" "8") # The number of dropped modules.
tasks=("boolq" "rte" "openbookqa" "piqa" "mmlu" "winogrande" "gsm8k" "hellaswag" "arc_challenge")
num_fewshots=("0" "0" "0" "0" "5" "5" "5" "10" "25")
for model_name in "${model_names[@]}"
do
# Download the model to a local directory.
git lfs install
git clone https://huggingface.co/mistralai/Mistral-7B-v0.1
mv Mistral-7B-v0.1 ./"$model_name"_model
for drop_module in "${drop_modules[@]}"
do
for drop_num in "${drop_nums[@]}"
do
cfg_path=./"$model_name"_drop"$drop_num"_"$drop_module"/config.json # PATH to the corresponding config.json file.
cp -f "$cfg_path" ./"$model_name"_model/config.json # Replace the original config.json file.
cp ./"$model_name"_drop"$drop_num"_"$drop_module"/*.py ./"$model_name"_model/ # Build the configuration and modeling files for remote code.
echo "Eval the config of:"
echo $cfg_path
num_tasks=${#tasks[@]}
for ((i=0; i<$num_tasks; i++)); do
CUDA_VISIBLE_DEVICES=$GPUs accelerate launch --main_process_port $port -m lm_eval \
--model hf \
--model_args pretrained=./${model_name}_model,trust_remote_code=True,dtype="bfloat16" \
--tasks ${tasks[$i]} \
--num_fewshot ${num_fewshots[$i]} \
--batch_size 1 \
--output_path ./${num_fewshots[$i]}shot_${tasks[$i]}_"$model_name"_drop"$drop_num"_"$drop_module".json >> output_"$model_name"_drop"$drop_num"_"$drop_module".out
done
done
done
done |