multi turn issue

[codeMonkey-shin]

The model seems to need to transmit as many images as there are questions.

Otherwise, you just keep repeating your previous answer.

I would like to have a conversation without images, but upload images only when necessary. Is this impossible?

[gxlover0625]

Might the code below will give you some ideas about how to construct conversations about pure text, single image, multiple images

def parse_input(self, query = None, imgs = None):
        """
        Parses the input query and images into the expected format.

        Parameters
        ----------
        query : str, optional
            The query from the user.
        imgs : list or str, optional
            A list of images or a single image path.

        Returns
        -------
        list of dict
            A list containing the messages formatted for the model.
        """
        if imgs is None:
            messages = [
                {"role": "user", "content": query},
            ]
            return messages
        
        if isinstance(imgs, str):
            imgs = [imgs]
        
        content = []
        for img in imgs:
            content.append({
                "type": "image",
                "image": img
            })
        content.append({
            "type": "text",
            "text": query
        })

        messages = [
            {"role": "user", "content": content},
        ]
        return messages