modelscope / 3D-Speaker

A Repository for Single- and Multi-modal Speaker Verification, Speaker Recognition and Speaker Diarization

fine-tune

JINzezhong7 opened this issue 5 months ago · comments

dsadsadsa commented 5 months ago

您好，我再训练完dino之后想用label数据fine-tune，如何加载之前的模型.pth

Chen Yafeng commented 5 months ago

可以参考https://github.com/alibaba-damo-academy/3D-Speaker/blob/main/speakerlab/bin/train_rdino.py 第115行restart_from_checkpoint函数。

dsadsadsa commented 5 months ago

我可以参考extract_rdino.py 中的加载模型的代码吗，因为fine-tune需要加classifier，不需要dino head

Chen Yafeng commented 5 months ago

都是可以的，哪个更适合你就选择哪个，have a try~