daje/Qwen3-TTS-12Hz-0.6B-Korean
Partial fine-tune vs Qwen/Qwen3-TTS-12Hz-0.6B-Base. Partial change: 49% of the weights changed (attention, embeddings, lm_head, mlp, norms), in 28 of 28 layers.
Tensors changed: lm_head 12/15, Embeddings 11/17, Norms 2/134, MLP 76/103, Attention 109/132, other 0/1