search

black-forest-labs / flux

Official inference repo for FLUX.1 models
Apache License 2.0
14.56k stars 1.05k forks source link

Loading components for the flux1-dev checkpoint Huggingface #153

Open VTaPo opened 2 weeks ago

VTaPo commented 2 weeks ago

I cannot loading components for the flux1-dev checkpoint.

Here is my code: pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16) pipe.enable_model_cpu_offload() pipe.to(torch.float16) prompt = "Create a commercial poster for restaurant" out = pipe( prompt=prompt, guidance_scale=3.5, height=768, width=768, num_inference_steps=1, ).images[0] out.save("image.png")

The process always stops when 20% loading. I have 40GB CPU RAM and 24GB VRAM on RTX3090.

everythoughthelps commented 2 weeks ago

pip install accelerate

VTaPo commented 2 weeks ago

@everythoughthelps Of course I did. However, every time I run the code as I provided, the loading components always stops at 20%. Is it possible that my configuration is not enough for using Flux1-dev from diffusers.

My requirements.txt file: torch torchvision

tokenizers transformers diffusers accelerate sentencepiece bitsandbytes

numpy