Currently, except the proj layers, other layers in model.diffusionmodel are not of type nn.linear. As a result, the lora layers cannot generate from those layers, which lead to mismatches between the names of the model weights and bias and those in the checkpoint files. I guess there are some updates in your model, can you upload the new model file(s)? Thank you!
Currently, except the proj layers, other layers in model.diffusionmodel are not of type nn.linear. As a result, the lora layers cannot generate from those layers, which lead to mismatches between the names of the model weights and bias and those in the checkpoint files. I guess there are some updates in your model, can you upload the new model file(s)? Thank you!