Download AI Models

AI Orchestrators on the Livepeer AI network can choose from various AI models to advertise on the network. A comprehensive list of currently supported models for each Livepeer AI pipeline is available on the Pipelines page. Before specifying which models to advertise on the Livepeer AI network, you must first download the models to your machine. The following steps will guide you through downloading the recommended models for Livepeer AI.

1. Install HuggingFace CLI

   Use the following command to install the HuggingFace CLI:

   pip install huggingface_hub
   

2. Generate Hugging Face Token

   Follow the instructions on the Hugging Face website to generate an access token with read permissions. Then, use the Hugging Face CLI to install the token:

   hf auth login
   

   You will be prompted to paste the access token you created on the Hugging Face website.

   If you have trouble locating hf in your path, it may be necessary to add /home/$USER/.local/bin/ to your local path.
3. Accept SVD1.1 Terms

   The currently recommended models include one gated model. Therefore, you must accept the terms of the SVD1.1 model on its model page before downloading.

4. Download AI Models

   Use the following command to download the recommended models for Livepeer AI:

   cd ~/.lpData
   curl -s https://raw.githubusercontent.com/livepeer/ai-worker/main/runner/dl_checkpoints.sh | bash -s -- --beta
   

   This command downloads the recommended models for Livepeer AI and stores them in your machine’s ~/.lpData/models directory. To obtain a complete set of models, omit the --beta flag. This will require additional disk space.