README: version note — weights updated to stage-5 multi-task distill (2026-05-30)

fix config _name_or_path -> JuzeZhang/ViBES-Audio (drop internal step24000 path)

update ViBES-Audio to stage5_mt_model/final (latest GLM-4-Voice 0.5B distill, step24000)

card: drop the 9B/0.5B variant table (lives in the GitHub repo)

ViBES-Audio: distilled 0.5B GLM-4-Voice (speech backbone for ViBES)

initial commit