A small app meant to test the capabilities explored in the Socratic Models paper. The idea is to use a committee of large language models (LLM), video-language models (VLM), e.g. CLIP, audio-language models (ALM) like wav2clip, and speech to text models to supervise each other. This demo explores the interaction between LLMs and VLMs using YouTube videos through a Streamlit app.
To run the app, a valid HuggingFace API key is needed. Create a key at https://huggingface.co/settings/tokens and save to a file src/hf_api.key.
Navigate to the src dir and run streamlit run app.py.