nicolasramos/Qwen3.8-9B-Distill-MLX-bf16-mtp
Partial fine-tune (fewer layers) vs empero-ai/Qwen3.8-9B-Distill. nicolasramos/Qwen3.8-9B-Distill-MLX-bf16-mtp matches part of empero-ai/Qwen3.8-9B-Distill (402 of its 775 tensors; the rest of empero-ai/Qwen3.8-9B-Distill has no counterpart). On the shared part: Partial change: 2% of the weights changed (mlp, norms), in 32 of 32 layers.
Tensors changed: Embeddings 0/1, Norms 81/105, Attention 0/200, MLP 3/96