Jael Gu

> This item is too big, you can use command > > > git clone https://towhee.io/towhee/deepfake.git --depth 1 git clone --depth1 works. @reiase @Chiiizzzy But is this (--depth 1) enabled...

@angelapytao Can you check timm version on both machines? `pip show timm`

> > > diff timm, maybe the reason？ I think so. Can you try timm 0.5.4 on both machines?

To install pytorch with cuda, please refer to https://pytorch.org/get-started/locally/ for proper versions & installation commands. (I recommend to use `conda install` because it will install proper `cudatoolkit` for your environment.)

@jandy0414 Can you try to modify your path with options below: - option 1: r'D:\x\img.png' - option 2: str(os.path.join('D:/', 'x', 'img.png'))

@micrazy Towhee supports user model as well. You have two options below: 1. wrap the model inference with a function to work with other towhee operators. Here are some example...

@micrazy `tensor_normalize` does normalize the input. You can check docs here: https://towhee.readthedocs.io/en/latest/builtin%20operators/tensor%20like.html . In your case, I think you need to squeeze dim 0 of the vec before normalization. i.e....

4. question_answering_system: https://katacoda.com/jael/scenarios/question_answering_system

Added Data2VecAudio via #238 . More contributions are welcome!

@junjiejiangjjj @wxywb Any idea?