CUDA out of memory. :( no hugging face no colab

antgroup / echomimic

EchoMimic: Lifelike Audio-Driven Portrait Animations through Editable Landmark Conditioning

https://antgroup.github.io/ai/echomimic/

Apache License 2.0

3.17k stars 368 forks source link

CUDA out of memory. :( no hugging face no colab #141

Closed medalawi closed 3 months ago

medalawi commented 3 months ago

“First, thank you for sharing this wonderful repository. -How to avoid cuda out of memory many repos has an option to set batch size how to set it in echomimic? Could you please create a lite version for users with lower-end graphic cards? -Additionally, Hugging Face closed the session before EchoMimic completed due to GPU time limits. -I tried using Google Colab but encountered a size limit while downloading the pretrained model, as there wasn’t enough space.”

nitinmukesh commented 3 months ago

duration of audio?

medalawi commented 3 months ago

@nitinmukesh Cuda out of memory show in terminal before launch host run in browser I tried huggingface with 2sec got GPU abort time limit

nitinmukesh commented 2 months ago

I am not sure why CUDA OOM error.

What is you PC specification? GPU / VRAM / RAM.

I just now converted 56s audio on RTX 4060 8 GB VRAM, it took a lot of time but no OOM issues