search

X-LANCE / AniTalker

[ACM MM 2024] This is the official code for "AniTalker: Animate Vivid and Diverse Talking Faces through Identity-Decoupled Facial Motion Encoding"
https://x-lance.github.io/AniTalker/
Apache License 2.0
1.36k stars 125 forks source link

I got a error #32

Open wildboy2arthur opened 1 month ago

wildboy2arthur commented 1 month ago

Thanks for sharing, it's very interesting, I also want to make a .npy file. Follow your instructions to perform the installation step by step, making no mistakes until the last step. My error message is as follows:

0%| | 0/1 [00:06<?, ?it/s] Traceback (most recent call last): File "D:\AniTalker\talking_face_preprocessing_back\extract_audio_features.py", line 52, in main(args) File "D:\AniTalker\talking_face_preprocessing_back\extract_audio_features.py", line 34, in main outputs = model(input_values, output_hidden_states=True) File "C:\Users\Fadawan.conda\envs\tfpw\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl return forward_call(*input, kwargs) File "C:\Users\Fadawan.conda\envs\tfpw\lib\site-packages\transformers\models\hubert\modeling_hubert.py", line 1074, in forward encoder_outputs = self.encoder( File "C:\Users\Fadawan.conda\envs\tfpw\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl return forward_call(*input, *kwargs) File "C:\Users\Fadawan.conda\envs\tfpw\lib\site-packages\transformers\models\hubert\modeling_hubert.py", line 800, in forward layer_outputs = layer( File "C:\Users\Fadawan.conda\envs\tfpw\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl return forward_call(input, kwargs) File "C:\Users\Fadawan.conda\envs\tfpw\lib\site-packages\transformers\models\hubert\modeling_hubert.py", line 630, in forward hidden_states, attnweights, = self.attention( File "C:\Users\Fadawan.conda\envs\tfpw\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl return forward_call(*input, **kwargs) File "C:\Users\Fadawan.conda\envs\tfpw\lib\site-packages\transformers\models\hubert\modeling_hubert.py", line 504, in forward attn_weights = nn.functional.softmax(attn_weights, dim=-1) File "C:\Users\Fadawan.conda\envs\tfpw\lib\site-packages\torch\nn\functional.py", line 1818, in softmax ret = input.softmax(dim) RuntimeError: CUDA out of memory. Tried to allocate 8.12 GiB (GPU 0; 12.00 GiB total capacity; 9.68 GiB already allocated; 0 bytes free; 10.74 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF GPUz

nitinmukesh commented 1 month ago

Please wait for some time. There is a pull request once merged the manual step of creating audio feature will not be required

https://github.com/X-LANCE/AniTalker/pull/31

wildboy2arthur commented 1 month ago

Thank you! I am looking forward to.

liutaocode commented 1 month ago

Hello, it seems the GPU memory is full. May I ask what is the duration of your audio file? Additionally, to facilitate quick testing, you can first test with audio under 10 seconds or modify the Hubert audio extraction process to use CPU (slower but won't cause OOM). After extraction is complete, subsequent calls to Anitalker can normally use CUDA. (PS: The main GPU tested for this library is the 3090 with 24GB, which is larger than your 12GB, so some cases may not pass testing on your GPU.)

nitinmukesh commented 3 weeks ago

Manual process is removed and new WebUI Please follow https://github.com/X-LANCE/AniTalker/blob/main/md_docs/run_on_windows.md