17slever17/Qwen3-ForcedAligner-Ultra
Partial fine-tune vs Qwen/Qwen3-ForcedAligner-0.6B. Partial change: 39% of the weights changed (attention, lm_head, mlp, other), in 28 of 28 layers.
Tensors changed: other 2/11, MLP 40/180, Norms 0/211, Attention 144/304, lm_head 1/1, Embeddings 0/1