from datasets import load_dataset
ds = load_dataset("pxyyy/RLHFlow_mixture", split='train')
def filter_data(example):
messages = example['messages']
if len(messages) < 2:
return False
for i in range(len(messages)):
if messages[0]['role'] == 'system':
if i == 0:
turn = 'system'
elif i % 2 == 1:
turn = 'user'
else:
turn = 'assistant'
else:
if i % 2 == 0:… See the full description on the dataset page:
https://huggingface.co/datasets/pxyyy/RLHFlow_mixture_clean_empty_round.
