Finetune on multi-image VQA dataset

[rsr-droid]

Hi, is it possible to finetune this model on a multi-image VQA dataset, which has a format similar but not limited to the following where we can pass a list of images with a single prompt:
[
{
"system_prompt": "You are a helpful AI assistant",
"image": [
"/path/to/image_1,
"/path/to/image_2"
],
"conversations": [
{
"from": "human",
"value": "Provide a description of the images in attached."
},
{
"from": "gpt",
"value": "The first image shows a Labradoodle dog running energetically through a grassy field while holding a stick in its mouth. The dog has curly, light-colored fur and floppy ears flying in the air as it sprints with a joyful expression.

The second image features a close-up of an orange-colored cat with its mouth wide open in a large yawn, showing its sharp teeth and tongue. The background is black, making the cat's fur and facial features stand out vividly."

  }
]

},