Limitations

• Non-commercial only. The weights are CC BY-NC 4.0, from the original YuE2-3B and YuE2-Vae by m-a-p.
• Everything in the original model card's limitations applies too.
• A song is generated in a single context window. Longer songs, which the original splits into chunks, aren't supported yet.
• For the same seed, the 4-bit model doesn't reproduce the bf16 song note for note: its songs drift into a different, equally valid performance.
• Only the decoder half of the VAE is included, so it can't encode reference audio.