SiddhJagani/Qwen3.8-2B-mlx-fp16
Partial fine-tune (fewer layers) vs empero-ai/Qwen3.8-2B-Distill. SiddhJagani/Qwen3.8-2B-mlx-fp16 matches part of empero-ai/Qwen3.8-2B-Distill (302 of its 632 tensors; the rest of empero-ai/Qwen3.8-2B-Distill has no counterpart). On the shared part: Partial change: 0% of the weights changed (norms), in 24 of 24 layers.
Tensors changed: Embeddings 0/1, Norms 61/79, Attention 0/150, MLP 0/72