Getting error while loading model
File "../python3.10/site-packages/transformers/modeling_utils.py", line 1824, in _check_and_enable_sdpa
raise ValueError(
ValueError: ConditionalChatTTS does not support an attention implementation through torch.nn.functional.scaled_dot_product_attention yet. Please request the support for this architecture: https://github.com/huggingface/transformers/issues/28005. If you believe this error is a bug, please open an issue in Transformers GitHub repository and load your model with the argument attn_implementation="eager" meanwhile. Example: model = AutoModel.from_pretrained("openai/whisper-tiny", attn_implementation="eager")
it solves if I switch to 'eager' but the usage note suggests to use either flash_attn2 or fsdp. Any suggestion?
check torch version
I have torch version 2.5.1. Is that an issue?
And one more question, If I am using it with 'eager' and it is running, what are the implications?
Pillow==10.1.0
torch==2.3.1
torchaudio==2.3.1
torchvision==0.18.1
transformers==4.44.2
librosa==0.9.0
soundfile==0.12.1
vector-quantize-pytorch==1.18.5
vocos==0.1.0
decord
moviepy
sorry about the delay, try pip install transformers==4.44.2 as stated in the readme, and I think it should be OK.
We use sdpa in the training stage, so sdpa inference will align with the training stage. Using eager may work, but it has a little precision gap between training and inference.