[Bug]: SD 3.5 Medium fine-tune safetensors size too large

[ronlamb]

What happened?

When saving the safetensors file for a SD3.5 Medium finetune the size is coming as 15.39 GB instead of the basemodel size of 4.99 GB.

For the options I chose the sd 3 config, with the following changes. Set the model dropdown to Stable Diffusion 3.5 and Fine Tune.

Only options changes I made was:
Weight Data Type: bfloat16,
Train DataType: bfloat16
Optimizer: ADAMW_8BIT.
Attention: SDP
Model: stabilityai/stable-diffusion-3-medium-diffusers

What did you expect would happen?

The file size should be about the same as the base model of 4.99GB

Relevant log output

No response

Output of pip freeze

absl-py==2.1.0
accelerate==1.0.1
aiohappyeyeballs==2.4.4
aiohttp==3.11.8
aiosignal==1.3.1
annotated-types==0.7.0
antlr4-python3-runtime==4.9.3
async-timeout==5.0.1
attrs==24.2.0
bitsandbytes==0.44.1
certifi==2024.8.30
charset-normalizer==3.4.0
cloudpickle==3.1.0
colorama==0.4.6
coloredlogs==15.0.1
contourpy==1.3.1
customtkinter==5.2.2
cycler==0.12.1
dadaptation==3.2
darkdetect==0.8.0
-e git+https://github.com/huggingface/diffusers.git@e45c25d03aeb0a967d8aaa0f6a79f280f6838e1f#egg=diffusers
filelock==3.16.1
flatbuffers==24.3.25
fonttools==4.55.0
frozenlist==1.5.0
fsspec==2024.10.0
ftfy==6.3.1
grpcio==1.68.0
huggingface-hub==0.26.2
humanfriendly==10.0
idna==3.10
importlib_metadata==8.5.0
intel-openmp==2021.4.0
invisible-watermark==0.2.0
Jinja2==3.1.4
kiwisolver==1.4.7
lightning-utilities==0.11.9
lion-pytorch==0.2.2
Markdown==3.7
markdown-it-py==3.0.0
MarkupSafe==3.0.2
matplotlib==3.9.2
mdurl==0.1.2
-e git+https://github.com/Nerogar/mgds.git@0723e42ceff37f3527e7e7f7dfaf7ad8f7252719#egg=mgds
mkl==2021.4.0
mpmath==1.3.0
multidict==6.1.0
networkx==3.4.2
numpy==1.26.4
nvidia-ml-py==12.560.30
omegaconf==2.3.0
onnxruntime-gpu==1.19.2
open_clip_torch==2.28.0
opencv-python==4.10.0.84
packaging==24.2
pillow==11.0.0
platformdirs==4.3.6
pooch==1.8.2
prodigyopt==1.0
propcache==0.2.0
protobuf==5.29.0
psutil==6.1.0
pydantic==2.10.2
pydantic_core==2.27.1
Pygments==2.18.0
pynvml==11.5.0
pyparsing==3.2.0
pyreadline3==3.5.4
python-dateutil==2.9.0.post0
pytorch-lightning==2.4.0
pytorch_optimizer==3.1.2
PyWavelets==1.7.0
PyYAML==6.0.2
regex==2024.11.6
requests==2.32.3
rich==13.9.4
safetensors==0.4.5
scalene==1.5.45
schedulefree==1.3
sentencepiece==0.2.0
six==1.16.0
sympy==1.13.1
tbb==2021.13.1
tensorboard==2.18.0
tensorboard-data-server==0.7.2
timm==1.0.11
tokenizers==0.20.3
torch==2.5.1+cu124
torchmetrics==1.6.0
torchvision==0.20.1+cu124
tqdm==4.66.6
transformers==4.46.0
typing_extensions==4.12.2
urllib3==2.2.3
wcwidth==0.2.13
Werkzeug==3.1.3
xformers==0.0.28.post3
yarl==1.18.0
zipp==3.21.0

[clayne]

Can independently verify this. The output model is around 16GB regardless of output precision (FP8, NF4, BF16, etc).