i want to fine tune the vision model.How can i prepare my owndataset

[primal sigil]

i want to fine tune the vision model.How can i prepare my owndataset without downloading from Huggin without downloading from Hugging face。I have already reshape the type like [
{ "role": "user",
"content": [{"type": "text", "text": Q}, {"type": "image", "image": image} ]
},
{ "role": "assistant",
"content": [{"type": "text", "text": A} ]
},
]but how can i save the image and use it

[ebon owl]

for our notebooks, we provide example datasets

follow the dataset's examples

just make sure the images are like the same size and less than 1000px