Questions about Gemini Finetuning Dataset

“Currently, I have fine-tuned using text_input and output. According to the AI, fine-tuning can also be done using a dataset with {"content":...}. However, I’m unsure if this is accurate information or just a hallucination from the AI. In the official documentation, I’ve only found references to text_input, output, or role types.”

Hi @itjimkr,

If you want to provide content in your fine-tuning dataset, you can format the text_input as shown in the example below:

{
        'text_input': ' content:    ***               input: *** ' ,
        'output': ' *** '
}