#SDXL Dreambooth - can’t load model
47 messages · Page 1 of 1 (latest)
Noo it’s not, I’m not sure why it is split in the logs
I checked through the web terminal and the file is there
Oh
~6 GB
It’s the second one
It seems like it can find the checkpoint but not loading it for some reason
Ooh
It says there's an error in the model loading
State dict, I'm not sure what does it means either
I'm not sure if DB endpoint supports SDXL as it's diffrent arch than SD 1.5 and 2.1
oh dreambooth i thought db=database mb lol
It's serverless channel so I though question is about serverless
yeahh
though it looks like our friend runs on pod
It is definitely serverless yeah
Dreambooth should work with SDXL though
And this is using sd-scripts which Kohya uses under the hood
Ohh
Soo.. How did you download it
Try searching that problem your experiencing on Google too
Maybe some part of the logs match with someone's problem
I have this line in my Dockerfile so two models get baked into the image
RUN mkdir -p /model_cache && \
wget "https://civitai.com/api/download/models/292213?type=Model&format=SafeTensor&size=full&fp=fp16&token=${CIVITAI_TOKEN}" -O /model_cache/hyperRealism_30.safetensors && \
wget "https://civitai.com/api/download/models/344487?type=Model&format=SafeTensor&size=pruned&fp=fp16&token=${CIVITAI_TOKEN}" -O /model_cache/realvisxlV40_v40.safetensors
Oh should be good
For some reason the shape is different or something
size mismatch for down_blocks.1.attentions.0.proj_in.weight: copying a param with shape torch.Size([640, 640]) from checkpoint, the shape in current model is torch.Size([640, 640, 1, 1]).
mayby it's fp16
I think it is yes
Ah apparently, the script that I am running, is only for SD1.5
This is the source code of kohya
I have to run a different script for SDXL apparently. Sorry guys, missed that one
Haha alright, nice