mu/LLM.md

mu

Full-length song generation (up to 4:45) with latent diffusion on a Diffusion Transformer backbone, with optional lyrics-conditioned alignment and an audio codec for reconstruction. Part of the Zen LM ecosystem (Hanzo AI).

Layout: model/ (DiT + codec), train/, infer/, g2p/ (grapheme-to-phoneme for lyrics), config/, dataset/.

• Setup: pip install -r requirements.txt

Full docs: README.md