xtuner-train_internvideo2_5

How to finetuning InternVideo2.5?

Note: We only support the training with video data.

Install

cd xtuner-train_internvideo2_5
pip install -e .

Prepare your data

1. Prepare your data annotations like this, if you need to use data packing, all data item in annotation file shuld have duration.
2. List your training data in data/diy_ft_data.json.

Start to training

If you need to use data packing to speed up:

bash ft_internvideo_2_5_datapacking.sh

otherwise:

bash ft_internvideo_2_5.sh

Evaluation

Copy the python file in https://huggingface.co/OpenGVLab/InternVideo2_5_Chat_8B and use lmms_eval to evaluate.