this dataset is intended to use with openbmb/MiniCPM-V-2_6 model
import json
from itertools import product
from PIL import ImageDraw
import torch
from datasets import load_dataset
from PIL import Image
from transformers import AutoModel, AutoTokenizer
model = AutoModel.from_pretrained('openbmb/MiniCPM-V-2_6', trust_remote_code=True,
attn_implementation='sdpa', torch_dtype=torch.bfloat16,
low_cpu_mem_usage=True
) # sdpa or flash_attention_2, no eager
model =… See the full description on the dataset page:
https://huggingface.co/datasets/famousdetectiveadrianmonk/person-attributes-fewshot.