Spaces:

multimodalart

/

flux-lora-the-explorer

like 1.21k

Running on Zero

I saw your post about ZEROGPU_V2 somewhere, but could not find it again, so forgive me for posting it here.😓

I am using an older version of the HF library in some spaces, pulled in by a library dependency that is required for those spaces, and I'm reporting an error I got when using ZEROGPU_V2 in those spaces.
On the other hand, a space with the new HF library does not work without it.

If it doesn't work, I just need to false the environment variable or leave it blank, so I'm not in trouble, but it would be better if it could be repaired in the future.

Case 1

• ZEROGPU_V2=true => Either torch's CUDA-related error prevented it from starting, or when it did start, it pointed out a problem with CUDA's initialization position when executing inference and did not work.
• nothing => Operates properly.

torch
spaces
diffusers=0.30.0
compel=2.0.2
invisible-watermark=0.2.0
transformers=4.41.2
accelerate=0.31.0

Case 2

• ZEROGPU_V2=true => Operates properly.
• nothing => torch error and doesn't start.

torch
torchvision
huggingface_hub
accelerate
git+https://github.com/huggingface/diffusers
spaces
transformers
peft
sentencepiece
timm
einops
controlnet-aux
kornia
numpy