SiddhJagani/Qwen3.8-4B-mlx-fp16
Partial fine-tune (fewer layers) vs empero-ai/Qwen3.8-4B-Distill. SiddhJagani/Qwen3.8-4B-mlx-fp16 matches part of empero-ai/Qwen3.8-4B-Distill (402 of its 738 tensors; the rest of empero-ai/Qwen3.8-4B-Distill has no counterpart). On the shared part: Partial change: 0% of the weights changed (norms), in 32 of 32 layers.
Tensors changed: Embeddings 0/1, Norms 81/105, Attention 0/200, MLP 0/96