[Feature] Is it possible to continue training from the last finetuned checkpoint?

[chunping-xt]

Thanks for this great project with very high sound quality, VQGAN is excellent. When I finetuned from 'fish-speech-1.2-demo' and had my finetuned checkpoint. After merge_lora from the last finetune checkpoint, I used this pretrained to continue finetuning on an extended dataset and the results were very bad. So every time I finetune I always have to finetune with fish-speech-1.2-demo to get good results. Is there any solution to save the effort of the last finetuned?