CogVideoX Support

[kursatdinc]

Will you add cogvideox support?

[kinam0252]

Hello, thank you for your question. Yes, we’re planning to implement support for CogVideoX in the near future. Please stay tuned for updates!

[kinam0252]

Will you add cogvideox support?

Thanks for waiting! 🙌 We’ve added STG support to the CogVideoX pipeline and updated it at CogVideoXSTGPipeline. Feel free to check it out! 🚀

[kursatdinc]

When I try to get inference with A100 40GB, I get Cuda OOM. Did you face this problem ?

[kinam0252]

In the experiment, there was no OOM with 80GB, but if you get an OOM with 40GB, you can try some tips from the diffusers CogVideoX docs - Memory Optimization part.

You can try removing .to("cuda") in pipe = CogVideoXSTGPipeline.from_pretrained(...).to("cuda")

and using some of the following options.
pipe.enable_model_cpu_offload()
pipe.enable_sequential_cpu_offload()
pipe.vae.enable_tiling()
pipe.vae.enable_slicing()

Hope this works!