popoala

> It seems like this is the only mismatch between meta data and actual file path. I am investigating why this happened and will update the dataset once I conclude....

> > any solution? > > Solved: Add --precision full --no-half to the command line How to "Add --precision full --no-half" with function like: pipe = StableDiffusionPipeline.from_pretrained(model_id, scheduler=scheduler, torch_dtype=torch.float16, revision="fp16",...

> train.py中：改成 training_args.ddp_find_unused_parameters = True 就可以了 After setting training_args.ddp_find_unused_parameters = True, running into this error: _RuntimeError: Expected to mark a variable ready only once_ This happened both in single GPU...