some question about dataset format
I have seen internvl2's fine-tuned data template, but I would like to ask whether there will be performance degradation if I do not follow this template for fine-tuning.
Specifically, our data set is shown below
"conversations": [
{
"from": "human",
"Value" : "What is the functionality of the element at: [0.6216, 0.9507, 0.7432, 1.0000]?"
},
{
"from": "gpt",
"value": "Click to navigate to the Account tab."
}
].
The template used by internvl2 is framed in the form of , and the coordinates are relative coordinates multiplied by 1000, so if I do not follow this form, will it cause performance degradation? And how to change to the correct fine-tuning template? Click to navigate to the Account tab Is the verb included with it?
In addition, I see that the data requiring a graph needs to be added \n before the first sentence, is it OK if I put it in another position?
Hello, thank you for your attention. My suggestion is that if your dataset is small (e.g., fewer than 10,000 samples), it is recommended to use the same format as InternVL2. If your dataset is larger, using a custom format should not be a problem.