Support symlink=False for from_single_file download.

[jlonge4]

Is your feature request related to a problem? Please describe.

When downloading from single file, take flux-dev-fp8 for example, the other pipeline components are loaded, but with a multi directory layout including symlinks (eg snapshots, blobs, etc).

This is a problem when trying to compress the downloaded components for deployment (think model.tar.gz for sagemaker).

Describe the solution you'd like.

It would be awesome to have a symlinks=False parameter like snapshot_download used to have prior to being deprecated.

Describe alternatives you've considered.

The alternative for me in this case was to download from single file locally, remove symlinks manually, and save the fp8 version of the flux dev pipeline to the hub.

Additional context.
Jlonge4/flux-dev-fp8

[DN6]

Hi @jlonge4 To avoid using symlinks it's better to download the checkpoint and configs beforehand with snapshot download.
https://huggingface.co/docs/diffusers/v0.32.2/en/using-diffusers/other-formats#local-files-without-symlink

You just have to set local_dir=<my local dir> in the snapshot download method if you're using huggingface_hub>=0.23.0

[jlonge4]

@DN6 it seems the easiest way to achieve this is just to do this. Would love to see a better way with snapshot download / hf hub but when creating the pipeline there is always something missing when using those methods.

import torch
from diffusers import FluxPipeline, FluxTransformer2DModel

# Load the FP8 transformer
transformer = FluxTransformer2DModel.from_single_file(
    "https://huggingface.co/Comfy-Org/flux1-dev/blob/main/flux1-dev-fp8.safetensors",
)

# Create the pipeline with the FP8 transformer
pipe = FluxPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-dev",
    transformer=transformer,
)

# Enable CPU offloading to reduce VRAM usage
pipe.enable_model_cpu_offload()

pipe.save_pretrained(save_directory='./model')

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[DN6]

@jlonge4 Okay to close this issue as resolved?

[jlonge4]

@DN6 sure