erkamk/qwen3-0.6b-tr
Full fine-tune vs Qwen/Qwen3-TTS-12Hz-0.6B-Base. Fine-tune: 99% of the weights changed, across 28 of 28 layers.
Tensors changed: other 1/77, lm_head 15/15, Embeddings 17/17, Norms 134/134, MLP 103/103, Attention 132/132
Full fine-tune vs Qwen/Qwen3-TTS-12Hz-0.6B-Base. Fine-tune: 99% of the weights changed, across 28 of 28 layers.
Tensors changed: other 1/77, lm_head 15/15, Embeddings 17/17, Norms 134/134, MLP 103/103, Attention 132/132