randombk/chatterbox-vllm

AI Summary: Replace the current Hugging Face-based model downloading and loading mechanism with a more robust and production-ready solution for downloading, storing, and loading the model weights in the Chatterbox TTS vLLM port. This involves implementing a more traditional model management system, potentially including features like caching and version control.

AI Summary: The task requires implementing learned speech positional embeddings in a vLLM port of the Chatterbox TTS model. This involves finding a method to obtain the position offset from the start of the first decoded token within the vLLM framework, addressing current instabilities at the end of generated audio snippets.