About diffusers weight files
I use diffusers for inference and finetune. Could you upload diffusers weight files?
You can use this for example to convert them https://huggingface.co/spaces/diffusers/sd-to-diffusers It will be faster probably than waiting
I am just too lazy to do converting :) But it’s too long to wait..So I have done converting today and found out the fact that the text encoder is not contained in safetensor file but according release note that the text encoder is finetuned rather than frozen.
text encoder is not contained in safetensor file but according release note that the text encoder is finetuned rather than frozen
Wait what? That does not sounds right, it is missing one layer which is fine but it is there according to model toolkit
It could be a bug of converting script( https://github.com/huggingface/diffusers/blob/main/scripts/convert_original_stable_diffusion_to_diffusers.py ) which hugginface <'u'> provides. The script downloads OpenCLIP weight instead of using self-contained weight from original safetensors file.
Might be bug yeah, since the last layer is missing due to CLIP skip it may think it's corrupted or you may need config YAML for it or something