how can I continue fine-tuning the model after SFT on my own data

[xsx1001]

Thanks for this awesome repo, and I am wandering how to continue fine-tune the model after SFT on my only data. Like the fintune_task_lora.sh in the official llava?

[xsx1001]

It seems that the finetuning config can only load the weights from pretrain. But I want load the fine-tuned model for further downstream fine-tuning

[xsx1001]

will you release the fine-tuned weights (before converted to hf format)?