|
class MMPB(ImageBaseDataset): |
|
TYPE = 'MCQ' |
|
DATASET_URL = { |
|
'MMPB': '', |
|
} |
|
# DATASET_MD5 = { |
|
# 'MathVision': '93f6de14f7916e598aa1b7165589831e', |
|
# 'MathVision_MINI': '060fe4fa5d868987ce179307bd5f8a33' |
|
# } |
|
def __init__(self, dataset='MMBench', skip_noimg=True, **dataset_kwargs): |
|
self.dataset_name = dataset |
|
self.data_path = "/workspace/MMPB/dataset6.csv" |
|
# self.preference_path = "/workspace/MMPB/human/human_total_preference.json" |
|
self.dataset_path = "/workspace/MMPB" |
|
self.multi_turn_base_path = "/workspace/MMPB/multi_turn" |
|
self.eval_prompting_path = "/workspace/MMPB/eval_prompt/eval_prompt.csv" |
|
# with open(self.preference_path, "r", encoding="utf-8") as file: |
|
# self.total_preferences = json.load(file) |
|
|
|
self.data = load(self.data_path) |
|
# self.mt_data = load(self.multi_turn_path) |
|
self.model = dataset_kwargs.get("model", None)[0] |
Thank authors for sharing the dataset. I have observed that the dataset available on HuggingFace is in the ".parquet" format. Nevertheless, the method of reading the dataset in this repository does not align with this format. I would like to inquire whether there are any additional processing operations that are still necessary.
