Absolutely, LiveKit does need Torch to work properly. Torch is included as a dependency in the project files and plays a key role in handling tensor operations, managing models, and running tests. If you have any questions about it, feel free to ask! Sources: convert_hf_to_gguf.py | tests/rl/test_actor.py

tldr; you can safely ignore this warning.

None of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.

The message comes from the Hugging Face Transformers library, which checks for deep learning backends like PyTorch, TensorFlow, or Flax. If none are installed, it issues a warning. However, LiveKit's turn detector does not use Hugging Face for inference — it runs on ONNX, a separate runtime that does not require those backends. Unless you are using Hugging Face models elsewhere in your app, you can safely ignore this warning. You can import PyTorch to get rid of the warning but it is not necessary.