Skip to main content

Required format for Dataset

  • image : Points to the location of the image (URL) that the model will access during training.
  • conversations: Contains the interaction where the “human” requests a caption and the “gpt” provides the caption or description.
The <image> tag in the conversation will be replaced by the actual image during training, allowing the model to generate the caption based on the image.

Sample Dataset containing a single reference datapoint