Only train my dataset with huggingface checkpoint repo LISA-13B-llama2-v1

[Wbzb]

Great work! I have a question as following: As I have low GPUs, how can I train my dataset with your huggingface checkpoint repo LISA-13B-llama2-v1, instead of training from LLaVA version with all the datasets (include ade20k, refCOCO and so on)

[Amark-cheey]

Do you have some ideas？