A Music-Gen model that uses llama3 architecture (without grouped query attention and KV cache).

WORK IN PROGRESS

Some good samples during tests:

audio_70423_5bb3a71c1b86e5f150a2.wav

audio_112036_5bd73d048dc6e8110489.wav

audio_76825_43d63656857c1a47d726.wav

audio_124840_75e8f0fd40a25a3a3a52(1).wav

audio(2)(1).wav

audio(1)(1).wav

audio(3).wav

audio_102433_0a51665641927bc7eb77.wav

GitHub: https://github.com/WaveGenAI/phonira