Open 2000ZRL opened 3 months ago
You can do so by adding your meta file in the instruction_data file. The meta file is a json and follows the following schema:
[
{
"video": "string",
"QA": {
"i": "string",
"q": "string",
"a": "string"
}
},
{
"image": "string",
"QA": {
"i": "string",
"q": "string",
"a": "string"
}
},
...
]
Thanks for your excellent work. I am curious if there are any instructions for fine-tuning video-llava on my own dataset?