search

huggingface / diffusers

🤗 Diffusers: State-of-the-art diffusion models for image and audio generation in PyTorch and FLAX.
https://huggingface.co/docs/diffusers
Apache License 2.0
25.41k stars 5.26k forks source link

PyTorch 2.3.0 Incompatibility with Current Diffusers Library #7950

Open JethroChow opened 4 months ago

JethroChow commented 4 months ago

Describe the bug

When running the StableDiffusionXLPipeline with a specific model file (Clay_SDXL.safetensors), inference works correctly on CPU but results in a segmentation fault when run on GPU. Below are the specific steps and configurations that lead to this issue.

Reproduction

import torch from diffusers import StableDiffusionXLPipeline

pipeline = StableDiffusionXLPipeline.from_single_file('/data1/tides/sd/SDXL.safetensors', torch_dtype=torch.float16) pipeline.to('cuda:2')

result_image = pipeline(prompt='girl')

Logs

/home/tides/miniconda3/envs/diffusers/lib/python3.9/site-packages/huggingface_hub/file_download.py:1132: FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use `force_download=True`.
  warnings.warn(
Some weights of the model checkpoint were not used when initializing CLIPTextModel: 
 ['text_model.embeddings.position_ids']
Some weights of the model checkpoint were not used when initializing CLIPTextModelWithProjection: 
 ['text_model.embeddings.position_ids']

Segmentation fault (core dumped)

System Info

Python Version: 3.9.19 PyTorch Version: 2.3.0+cu118 CUDA Version: 11.8 Diffusers Version: 0.27.2 Operating System: Ubuntu

Who can help?

No response

JethroChow commented 4 months ago

The solution is to downgrade the torch version to v2.1.2.

tolgacangoz commented 4 months ago

In Google Colab's T4, one is able to use diffusers with PyTorch 2.3. Could you be sure that PyTorch 2.3 and its CUDA parts are installed properly in its environment? Also, installation in a fresh new environment can be tried. Lastly, could I ask your GPU's model name and its vRAM?

JethroChow commented 4 months ago

(base) tides@VM-112-2-ubuntu:~/StableDiffusionNotebooks$ nvidia-smi Fri May 17 13:54:15 2024
+-----------------------------------------------------------------------------+ | NVIDIA-SMI 525.105.17 Driver Version: 525.105.17 CUDA Version: 12.0 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |===============================+======================+======================| | 0 NVIDIA A10 On | 00000000:0B:01.0 Off | 0 | | 0% 30C P8 14W / 150W | 0MiB / 23028MiB | 0% Default | | | | N/A | +-------------------------------+----------------------+----------------------+ | 1 NVIDIA A10 On | 00000000:0B:02.0 Off | 0 | | 0% 31C P8 14W / 150W | 2MiB / 23028MiB | 0% Default | | | | N/A | +-------------------------------+----------------------+----------------------+ | 2 NVIDIA A10 On | 00000000:0B:03.0 Off | 0 | | 0% 30C P8 14W / 150W | 0MiB / 23028MiB | 0% Default | | | | N/A | +-------------------------------+----------------------+----------------------+ | 3 NVIDIA A10 On | 00000000:0B:04.0 Off | 0 | | 0% 29C P8 13W / 150W | 0MiB / 23028MiB | 0% Default | | | | N/A | +-------------------------------+----------------------+----------------------+

tolgacangoz commented 4 months ago

You also tried installing the torch in a fresh new environment, right?

JethroChow commented 4 months ago

yes

github-actions[bot] commented 2 weeks ago

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.