After cloning the repository, run the following commands to create a virtual environment in your backend directory:
MacOS/Linux:
python3 -m venv .venv
Windows:
python -m venv .venv
Change into the virtual environment directory:
cd .venv
Then, activate the virtual environment:
MacOS/Linux:
source venv/bin/activate
Windows:
.\Scripts\Activate.ps1
Install the required packages provided in the requirements.txt file:
MacOS/Linux:
pip install -r ../requirements.txt
Windows:
pip install -r ..\requirements.txt
The download of PyTorch depends on your system. If you have a CUDA-enabled GPU with Windows, you can install the GPU version of PyTorch. First download CUDA (Version 11) from here and then run the following command:
Windows with CUDA-GPU:
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
Windows without CUDA-GPU:
pip install torch torchvision torchaudio
If you use a Mac, you can´t use CUDA, so just run the following command:
Mac:
pip install torch torchvision torchaudio
To be able to access models on HuggingFace you need to have an access token. You can get one by signing up on HuggingFace. Navigate to your profile, then to 'Access Tokens' and create a new token with READ as role.
Then, create a file called .env in the root directory of the project and add the following line to it:
HF_TOKEN="YOUR_TOKEN"
NOTE: Files starting with a dot are hidden by default. To see them in your file explorer, you need to enable the option to show hidden files.
On HuggingFace, choose the model you want to use and click on the button next to the model name to copy the correct name.
Insert it into the model and tokenizer variables. Play around with the ìnput_text variable to get different results.
Then, run run.py.
If you use the model for the first time, it will be downloaded and saved in the following directories:
MacOS:
user/.cache/huggingface/hub
Windows:
C:\Users\user\.cache\huggingface\hub
Also, if you run out of storage, you can delete the models in this directory.