Regarding the training weights file issue

[ldy674]

When training Stage-1 with my own dataset, there are two pytorch_model.bin files, one is 3.5GB and the other is 1.8GB. However, when I use these files for Stage-2 training, it shows missing keys. What could be the reason for this?

[ldy674]

I found that the keys in my trained model are missing the `adapter_modules` and `controlnet` parts compared to the official ones. How can I solve this?

[Jeoyal]

Hi @ldy674 , thank you for your interest in our work. adapter_modules are loaded from ckpts trained from stage1 and controlnet is loaded from Unet.

[ldy674]

Hi @ldy674 , thank you for your interest in our work. adapter_modules are loaded from ckpts trained from stage1 and controlnet is loaded from Unet.

Thank you very much for your reply. After I finished training stage 1, the above five files appeared. During stage 2, do the --pretrained_ip_adapter_path and pretrained_style_encoder_path correspond to my pytorch_model.bin and pytorch_model_1.bin, respectively?

[ldy674]

Hi @ldy674 , thank you for your interest in our work. adapter_modules are loaded from ckpts trained from stage1 and controlnet is loaded from Unet.

Thank you very much for your reply. After I finished training stage 1, the above five files appeared. During stage 2, do the --pretrained_ip_adapter_path and pretrained_style_encoder_path correspond to my pytorch_model.bin and pytorch_model_1.bin, respectively?

Thank you very much, I tried it just now and it worked.