Is it possible to train Dreambooth on an already finetuned model?
Meaning, instead of selecting v1-5 or v2, how could I choose checkpoints of an already trained model?
Thanks a lot for your work.
Cheers
Hi, yes this is possible but you would need to customize the code a little bit on your duplicated Space. A hacky way would be just replacing the repo_id
here from multimodalart/sd-fine-tunable
to the one you would like to Dreambooth on here: https://huggingface.co/spaces/multimodalart/dreambooth-training/blob/main/app.py#L34
ah awesome, great idea, thanks!
Hi, yes this is possible but you would need to customize the code a little bit on your duplicated Space. A hacky way would be just replacing the
repo_id
here frommultimodalart/sd-fine-tunable
to the one you would like to Dreambooth on here: https://huggingface.co/spaces/multimodalart/dreambooth-training/blob/main/app.py#L34
Ok Multimodelart, I got it. But why are .safetensors and .ckpt files excluded in app.py#34 ? In other words, what format does a space need to be in order to be eligible as a trained model? Can I include a direct link to a ckpt model in app.py#34 or does it have to be to a space which includes the ckpt file?
Does this also work with private models? Somehow I have not been able to get this to work. I always get only error messages. I am not a Python pro (yet).
Hi, yes this is possible but you would need to customize the code a little bit on your duplicated Space. A hacky way would be just replacing the
repo_id
here frommultimodalart/sd-fine-tunable
to the one you would like to Dreambooth on here: https://huggingface.co/spaces/multimodalart/dreambooth-training/blob/main/app.py#L34Ok Multimodelart, I got it. But why are .safetensors and .ckpt files excluded in app.py#34 ? In other words, what format does a space need to be in order to be eligible as a trained model? Can I include a direct link to a ckpt model in app.py#34 or does it have to be to a space which includes the ckpt file?
bump