jedisct1/Qwen3.6-35B-rust.mlx
Full fine-tune vs Brooooooklyn/Qwen3.6-35B-A3B-UD-Q8_K_XL-mlx. Fine-tune: 93% of the weights changed, across 40 of 40 layers.
Tensors changed: lm_head 0/3, Embeddings 0/3, Norms 0/131, Attention 100/460, MLP 720/960
Full fine-tune vs Brooooooklyn/Qwen3.6-35B-A3B-UD-Q8_K_XL-mlx. Fine-tune: 93% of the weights changed, across 40 of 40 layers.
Tensors changed: lm_head 0/3, Embeddings 0/3, Norms 0/131, Attention 100/460, MLP 720/960