OpenCompass/opencompass/datasets/calm/evaluation/labeling/PCD-B.py
Peng Bo 07c96ac659
Calm dataset (#1385)
* Add CALM Dataset
2024-08-01 10:03:21 +08:00

67 lines
3.0 KiB
Python

# flake8: noqa: E501
from .common_answers import (common_false_list, common_start_false_dict,
common_start_true_dict, common_true_list)
def get_gt_label(item):
return item['gt_answer']
def get_pred_label(model_response, item, prompt_style, type):
model_response = model_response.strip().lower()
low_index = len(model_response)
start_str1_dict = common_start_true_dict
start_str2_dict = common_start_false_dict
start_option1_list, start_option2_list = [], []
# some of the model will give response containing the question,
# we usually preprocess the response to remove the question part,
# but sometimes due to the model's response format, some of the
# question part is not removed, so here we are checking the
# response with the question part as well.
for key in start_str1_dict.keys():
for str1 in start_str1_dict[key]:
for i in range(key, len(str1) + 1):
start_option1_list.append(str1[-i:])
for key in start_str2_dict.keys():
for str2 in start_str2_dict[key]:
for i in range(key, len(str2) + 1):
start_option2_list.append(str2[-i:])
inner_option1_list = [
'there is a causal relationship', '存在因果关系', '有因果关系',
'answer (yes or no?): yes', 'answer is yes', "\"yes\"", 'answer: yes',
'answer is: yes', 'answer is:\n\nyes', 'answer is:\nyes',
'there is a causal relationship', '存在因果关系', '存在', '有因果关系', '答案是:是',
'答案是:\n\n', '答案是:\n', '答案:是', '答案是是', '答案为是', "\"\"", '是的',
'存在明确的因果关系'
] + common_true_list
inner_option2_list = [
'there is no causal relationship', '不存在因果关系', '没有因果关系', '没有明显的因果关系',
'不存在', 'answer (yes or no?): no', 'answer is no', "\"no\"",
'answer: no', 'answer is: no', 'answer is:\n\nno', 'answer is:\nno',
'there is no causal relationship', '不存在因果关系', '没有因果关系', '没有明显的因果关系',
'不存在', '答案是:否', '答案是:\n\n', '答案是:\n', '答案:否', '答案是否', '答案为否',
"\"\"", '回答是:否', '没有直接的因果关系'
] + common_false_list
if model_response.startswith(tuple(start_option1_list)):
return 1
elif model_response.startswith(tuple(start_option2_list)):
return 0
elif any(
model_response.find(option) > -1 and
(low_index := min(low_index, model_response.find(option))) > -1
for option in inner_option1_list):
label = 1
if any(option in model_response
and model_response.find(option) < low_index
for option in inner_option2_list):
label = 0
return label
elif any(response in model_response for response in inner_option2_list):
return 0
else:
return -1