28 lines
675 B
Python
28 lines
675 B
Python
from datasets import load_dataset
|
|
from utils import process_dataset
|
|
|
|
|
|
|
|
def process_func(input_dict: dict):
|
|
conversations = input_dict["conversations"]
|
|
n = len(conversations) // 2
|
|
examples = []
|
|
|
|
for i in range(n):
|
|
user_msg = conversations[2*i]["value"]
|
|
assistant_msg = conversations[2*i+1]["value"]
|
|
examples.append({
|
|
"query": user_msg,
|
|
"response": assistant_msg
|
|
})
|
|
|
|
return examples
|
|
|
|
|
|
if __name__ == "__main__":
|
|
dataset = load_dataset("BelleGroup/train_3.5M_CN")
|
|
process_dataset(
|
|
dataset_dict=dataset,
|
|
output_subdir="belle-sft",
|
|
process_func=process_func,
|
|
) |