Wipe the misfiled Vevo v1 layout and upload the inference-only files from RMSnow/Vevo2: AR (posttrained), FM (fm_emilia101k_singnet7k_repa), Vocos vocoder, and content-style + prosody tokenizers. README rewritten to match. Training artifacts and the _text/pretrained variants are intentionally dropped.
Mirror contentstyle_modeling/Vq32ToVq8192/model.safetensors from amphion/Vevo
