Error while running in Google Colab (torchaudio 2.5.1+cu121 requires torch==2.5.1)

[deicool]

Bug Report
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
gcsfs 2024.10.0 requires fsspec==2024.10.0, but you have fsspec 2024.9.0 which is incompatible.
torchaudio 2.5.1+cu121 requires torch==2.5.1, but you have torch 2.4.1 which is incompatible.

Problem
OmniGen Module on Colab does not work as expected

What is expected to happen?
OmniGen should convert text to image

What does actually happen?
OmniGen is unable to create a image from text, giving error

Information
Command or Code
!git clone https://github.com/staoxiao/OmniGen.git
%cd OmniGen
!pip install -e .
!pip install gradio spaces
!python app.py --share

Environment, Platform, Device
Google Colab

[able2608]

That might not be an issue if you just want to get it to run. Most of the time the scripts will still run fine even if dependency mismatch or conflict occurs (like what you see) happens. Just let it run as is and see if it launches. If it does launch, then you're good to go.

[deicool]

Hi able2608,

It does run, but when I try to create a image, the following error comes up:

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
gcsfs 2024.10.0 requires fsspec==2024.10.0, but you have fsspec 2024.9.0 which is incompatible.
torchaudio 2.5.1+cu121 requires torch==2.5.1, but you have torch 2.4.1 which is incompatible.

[able2608]

That does not look like an error message that will pop up when a python script, more like one when you are installing libraries through pip. Let me check out what is happening then.

[able2608]

Tested with colab's free tier and generating images works fine. I can confirm that the error you see does indeed come from pip and it does not interfere with the image generation process. You can safely ignore that and keep going on.

BTW, if you are using colab free tier for this, you'll definitely need to switch to this PR (#151) so that you don't get OOM on GPU, though the gradio demo you are using now still needs some patchwork to use prequantized weight provided by the PR's author to prevent OOM on system RAM during quantization (which will crash colab and return ^C without any error message). Further more, the generation speed will be very SLOW, approximately 30 seconds per iteration for 1024x1024 without any image condition and model offload. If you have other cloud computing provider (that provides something newer than colab's T4) or have a machine with at least 6GB of VRAM and 16GB of RAM, I would strongly recommend you use those instead. If you just want to try it out, the author's own huggingface demo can be a good starting point, though you'll quickly run out of ZeroGPU compute time and need to wait for a while or upgrade your membership to get more compute time. You will definitely not want to wait for half an hour just for a single image or edit.

[deicool]

Thank you so much "able2608".